Today I actually want to talk about Datacenters, but there are some digital airwaves AI news to cover first. This is going to be a lot of curation and a lot of infographics but a measure of the State of AI as it is today, including AI Infra gridlock that seems to have a media ban. This kind of content takes me hours and hours to compile but is meant to be nibbled on. So bookmark it for weekend reading if it’s your style.

Anthropic’s latest model Mythos is getting a lot of scrutiny (okay fine, outlandish praise) as a massive “step-up” in model capability in coding and reasoning. We now leave behind the “Opus” family of models for Capybara models that are bigger and more capable. Anthropic released a Mythos System Card Preview that’s fairly interesting.

Mythos Models are so good they are Dangerous

Anthropic has developed a New Protocol for Releasing Models in a more Safe way: Project Glasswing

In essence, Anthropic's new model, Claude Mythos, is so powerful that it is not releasing it to the public. Instead, it is starting a 40-company coalition, Project Glasswing, to allow cybersecurity defenders a head start in locking down critical software.

Anthropic’s Revenue (ARR) Surges in Early 2026

This as Anthropic’s ARR has hit $30 Bn. and is about to overtake that of OpenAI. The following are true stories:

• Anthropic added the combined ARR of Palantir, Anduril, and Databricks—in one month. (via Lenny) - we had said 2026 would be a key year for Anthropic’s growth overtaking that of OpenAI.

• Between February and April Anthropic went from $19B in February to $30B in March, 2026.

• Anthropic’s revenue grew 30x in 15 months.

• Said another way: They added $6B in ARR just in *February*. Companies like Palantir and Atlassian took 15-20 years to reach ~$5B ARR. Anthropic is adding that every month. (Lenny).

• Anthropic dominates Enterprise AI and that lead is accelerating. It now has 1,000 companies spending $1M+ each. A figure which doubled in under two months.

• Anthropic by having the top AI coding model displayed around 1000% CAGR over the last 2.5 years.

Visualize It

The revenue growth at both Anthropic and Cursor demonstrate that vibe-working might have a sustainable future. Curiously OpenAI, Google and neither Microsoft or Meta are really in the picture. Anthropic’s Enterprise AI revenue began to step up just as their coding models have.

I think it was Claude 3.5 Sonnet that changed everything and established Anthropic as SOTA in coding. A bigger impact that even GPT-4 if you look at it historically many years from now. By the time Opus 4.0 was released, Claude Code was released in May, 2025 and Claude Cowork in January, 2026.

SOTA Opus Models, Claude Code and Claude Cowork changed everything. Who even knows what Mythos can unlock?

Leave a comment

Sam Altman Exposé Goes Viral

New Yorker bombshell exposes OpenAI's Sam Altman as untrustworthy manipulator.

Why read this? The New Yorker published an 18-month investigation into Sam Altman. The conclusion, from his own former leaders: he cannot be trusted.

Not news to us, but with looming IPOs it’s a problem for the AI ecosystem and AI bubble. OpenAI needs to change leadership, or risk everything! Both Sam Altman and Elon Musk are reputational risks to their companies now.

Read the Story

AI and Jobs

The Forecasting Research Institute completed what they call the most comprehensive study of how economists and AI experts think AI will affect the U.S. economy.

Read the Report

A majority of respondents anticipate significant advances in AI capabilities by 2030 with relatively stable historical trends for jobs and the labor market. That is, they predict major AI progress—but no dramatic break from economic trends in the next two decades: similar GDP growth rates to today, and a moderate decline in labor force participation.

What stood out to me? Labor Force Participation (LFPR): Under rapid progress, the LFPR is projected to drop from roughly 62% to 55% by 2050. Way more gradual than more biased economic reports insinuate.

Average Probability of AI Scenarios by 2030

I like how they modeled things: (although still a bit skewed to the Tech optimistic side). Nearly everyone is skewed to the AI optimism side now.

Helium Strait of Hormuz Bottleneck Intensifies

This is awkward?

• Helium prices have skyrocketed since the start of the Iran war.

• Qatar, which supplies a third of the world’s helium, has had production disrupted.

• Helium is a critical input for industries like semiconductors, healthcare, and aerospace.

This so-called cease-fire notwithstanding, Helium’s impact to the semiconductor supply-chains is causing HBM prices to rise suddenly faster.

The semiconductor, medical-imaging, and aerospace sectors all rely on helium as a crucial component.

A third of the world’s helium is (was?) trapped. Qatar is the source of 34% of global helium supply. I’m fairly pessimistic about Trump’s ability to get a deal done with Iran and consider the Iran war a major threat to the AI bubble.

• Helium prices are skyrocketing: early reports cited a 50% spike in the early going of the war, while more recent estimates say helium has doubled since late February.

• Helium is as far as the Iran War and the semiconductor supply chain goes: as close to a “fundamental physical constraint” as we can get.

• Qatar Energy announced in March it was halting production of liquefied natural gas and "associated products" following attacks on its facilities in Ras Laffan and Mesaieed. A 2-week ceasefire doesn’t change this, or anytime soon.

OpenAI over-invested in Compute and have a mediocre path to profitability

Read more

One of the hallmarks of 2025 in AI analysis, was discussion and debate around the AI boom relative to the AI bubble. These days I gravitate towards original macro analysis around the AI Infrastructur…

Read more

Good Morning,

When all is and done, 2025 saw a significant increase in major AI datacenter projects, clusters and infrastructure. It’s easily one of the biggest trends in AI of 2025.

This year we have seen a tidal wave of AI Infrastructure. But what will be the price for society? In 2025, the aggregate capital expenditure (CapEx) for top Big Tech hyperscalers is estimated to be over $400 billion, primarily driven by investments in AI infrastructure and data centers by BigTech and Cloud hyperscalers.

Products like ChatGPT being adopted at scale will likely lead to a lower value of a College education, less human connection and less time spent reading. Is that good for society? 🤔 Is that the sort of world we think is healthy for consumers and our younger people? I’m not sure the benefits of chatbots even outweigh the costs.

Leave a comment

ChatGPT leading to less Reading Time

YouTube, TikTok and especially ChatGPT are impacting reading time among our youngest teenage citizens. The impact on College age Americans is likely even more striking with ChatGPT use.

• Less reading time

• Less human connection

• Less stable white-collar career ladders

• Higher electricity bills

• More noise pollution for local communities

• More AI spam and AI slop on the internet

• Less human and real social media (as if that were even possible)

• More authoritarian use of AI by Governments, big corporations, tycoons, dictators and elites to serve their own agendas

Mixed Capital Expenditures and Confused Policy

We also have problematic supply chains and an energy grid that isn’t prepared for this AI Infrastructure roll-out.

According to datacenterwatch.org, there has been recently $162 𝐁𝐢𝐥𝐥𝐢𝐨𝐧 worth of data center projects in the US have been stalled or cancelled over the course of a year, with 60% in the latest 3 months. It’s not just supply chain issues, people and local residents are protesting Hyperscalers and their manipulation of real-estate and local politicians. Americans locally at least, are starting to oppose in greater numbers these Datacenter projects in their communities.

The Dark Side of Datacenters in America

Share

Business Insider did a documentary (it’s over 30 min) on this that really got me thinking. BigTech isn’t being transparent about their datacenters or their plans. Meanwhile the U.S. Government and the Trump Administration is trying to hasten the advent of these datacenters. Just like chatbots hurting the career ladders of young people, they are pushing something most citizens and people don’t want.

Higher Electricity Bills will Push Affordability Crisis

I read several reports on AI in 2025 with surveys that clearly showed regular workers and citizens aren’t positive about the impact of AI. With an affordability crisis and weak labor market taking shape in the U.S., higher electricity bills aren’t welcome either. Data centers significantly raise electricity bills, with some areas seeing costs jump dramatically (up to 267% higher in places with heavy activity) due to massive energy demand from AI, while national averages see increases of 30%+ since 2020. projects peg for further electricity price hikes of 8-25% by 2030, as utilities pass costs onto consumers and struggle to build new infrastructure, though factors like grid management and local investment also matter.

There’s clearly something wrong happening in America about Generative AI and around datacenters in particular. There’s a fairly good reason BigTech are trying to hide and bulldoze their activities. They accelerate power demands on the grid, imbalance water risks, hurt the health of residents near the infrastructure and many other toxic consequences.

The SPEED ACT is a bill against the people as major corporations push their agendas and products, even when adoption is surprisingly poor. Lobbying and how comprised this Administration appears to be means the bill cleared a crucial procedural vote 215-209 on Tuesday in the House and I fear for what this will mean for the bottom half of Americans live in a K-shaped semi-perma post pandemic economy. Quality of life may continue to degrade in terms of affordability as the labor force weakens with multiple environmental and risks to the cost of living. This can’t be American the Brave or America the Free. Datacenters are becoming a symbol of this deterioration, nevermind the environmental concerns.

The SPEED Act would blunt the 1969 National Environmental Policy Act, which mandates federal reviews for projects that could affect the environment before permits are issued. - per CNBC.

The Backlash

Republicans on Capitol Hill who have championed the tech industry’s race to dominate artificial intelligence are confronting a growing political obstacle: voters angry over the soaring energy demands and utility costs tied to the data centers. - per Politico.

2025: the Year AI Infrastructure Made AI Political

Even Republicans are starting to distance themselves from a Trump Administration in league with BigTech lobbying and the will to dominate of corporate elites. Florida Republican Gov. Ron DeSantis floated new limits on data centers as part of an “AI bill of rights.” Georgia Rep. Marjorie Taylor Greene and Missouri Sen. Josh Hawley have also questioned Trump’s state preemption drive. But I’m not sure American voters even understand the magnitude of the threat on their independence.

The U.S. has a huge First Mover Advantage in Datacenters, Compute and AI Chips

America’s stranglehold on “AI Supremacy” may be temporary

Read more

Good morning,

This week we plan to have more Anthropic Claude related guides, so stay tuned. Welcome to the 184,400 + readers and thanks to you all who live in 213 countries (which sounds like a lot), where the state of California makes up 20% of my readers. Readers from New York and California states make up more than 30% of my paid readers. In total, the U.S. and UK make up more than 70% of my premium readers. I coulnd’t do this without your support:

Subscribe now

This article is an infographics dump that eventually leads to Datacenter industry analysis. For some context:

• Read our Primer on AI Datacenters.

I’m going to make the argument that rising debt will result from an exponential demand for compute super-cycle. Let’s begin.

Share

Six hours ago (at the time of writing) it was announced that SoftBank has approved remaining $22.5 billion of OpenAI investment, the Information reports.

This means as China buys up gold, the U.S. is going to go into considerable debt to fund an exponential demand for compute supercycle. Just to meet this capacity I anticipate a lot of debt to power and build the required AI Infrastructure. As more Hyperscalers and Neo Clouds come “online” just to meet capacity starts to become a real problem (“capital crunch”) in the years ahead.

As Palisade Research released a paper last month which found that certain advanced AI models appear resistant to being turned off, at times even sabotaging shutdown mechanisms, things are heating up in more ways than one. 🔥 OpenAI has written about the importance of this property, which they call interruptibility—the ability to “turn an agent off” but is anyone taking it seriously? The real problem economically is how do you turn off the debt that will be required to fund the exponential demand for compute that’s coming?

The United States is so far behind China in energy and especially renewable energy systems, building a massive amount of datacenters doesn’t solve the future, it just creates more leverage. It burns you up with the required capital allocation and debt. As we nudge closer to the 2030s debt is going to start to be a more lively chimera, especially as it relates to AI Infrastructure mandated and facilitated by the Trump Administration. Don’t hold your breath though, the U.S. national debt has surpassed $38 trillion and is increasing at a historic rate, adding about $1 trillion every five months, which is more than twice the rate of the past 25 years. Eventually (like soon-ish) the AI bubble and debt bubble converge. What happens then? I’m not an economist, but it isn’t likely good.

Debt to GDP Ratio is Reaching Unsustainable Levels in Trump’s 2nd Term in the United States

The rather zany capital allocation mismatch between fiscal sustainability and environmental sustainability is a real weakness of the approach the U.S. is taking here with regards to AI. Monopolistic corporations can fund parts of the on-ramp, but as the demand for compute accelerates that will be impossible.

Generational Wealth Much? 💰

Boomers who in 2025 still hold the majority of the wealth in America, are making some pretty outlandish bets that younger generations will have to reckon with, own and likely, to pay for in blood. Baby Boomer generation holds the most wealth among all the generational cohorts, with an estimated net worth of around $82 to $85 trillion, representing over half of all U.S. household wealth. But can you trust them to make good decisions for the future or the rest of us? Or are we doing down in a blaze of glory? 🤠😂

Neo Cloud Winners Chosen ☁︎

Peter Thiel backed Crusoe will go IPO in early 2026 and have even more access to debt leverage and partnering (vendor) financing just like CoreWeave has in 2025. The Denver based AI data centre startup Crusoe is raising $1.38 billion at a valuation of about $10 billion from an anticipated Series E funding round. Both Nvidia and OpenAI have equity stakes in CoreWeave, who was among the first to pivot from Bitcoin mining into AI workloads and datacenters to rent GPUs. We did a deep dive about these Neo Clouds here in case you missed it. Founder’s Fund led Crusoe’s 2024 funding investment. The companies with the most access to partner financing and debt leverage win and of course political connections in the Trump Administration. CoreWeave pivoted from cryptocurrency mining to AI infrastructure in 2019, a full four, five to six years before most of its Neo Cloud peers. Crusoe calls itself the “AI Factory” company.

AI Infrastructure will Increase Debt and Leverage in Classic Bubble 🌌

The Build Baby Build motif of U.S. capitalism building AI Infrastructure comes with risks. Meanwhile the Bank of England (among others) is probing the financing of datacenters and AI Infrastructure deals as debt is mounting. Due to the exponential demand for compute rising for Hyperscalers (we can count Oracle now among the Big Three, the Big Four now), lending and debt is going to skyrocket in the 2026 to 2032 period for this “We must Win” push by the Trump Administration backed by the incentives of its Venture Capital community. Who is steering the state anyways?

If the demand for compute rises exponentially like I think it will, datacenters will be like the “granaries of smart cities”. A real life entry into a Machine Economy era.

The “AI Deal” Era 🦄

In a rapid fire last six months we’ve seen huge partnerships by the likes of Nvidia, OpenAI, Oracle, Microsoft, CoreWeave and others. Late last week, Anthropic and Google officially announced their cloud partnership. Another deal estimated at tens of billions - Anthropic gets access to up to one million of Google’s custom-designed Tensor Processing Units, or TPUs. Alphabet and Amazon both have major equity stakes in Anthropic who are estimated to increase their ARR run rate the fastest among the BigAI startups. The deal will add around or over a gigawatt of compute capacity by 2026, supporting Anthropic’s surging $7 billion revenue run rate.

Read the Blog

♨️ Burning the Future the American Way

While AI researchers literally burn out, datacenters will be overheating gozzling water near a U.S. states and country near you. The United States have started a race with China to build datacenters in other countries and leverage trade advantages that penalizes the other (while pitching Sovereign AI to buy Nvidia’s GPUs 😀) as Trump’s East-asian tour begins. I’m sure it will be accompanied by more AI Infrastructure deal announcements and more debt to back the projects.

As the price of Gold and equities are at all-time highs, and in a dysfunctional federal government shutdown, the U.S. government’s gross national debt surpassed $38 trillion last week. The Trump administration’s performance? It’s the fastest accumulation of a trillion dollars in debt outside of the COVID-19 pandemic — the U.S. hit $37 trillion in gross national debt in August this year. If only policy makers understood the AI Infrastructure’s exponential cost of the demand for compute that will keep rising! It’s dangerous to awaken the machine, but not the ghosts that you’ve been told about.

Trump’s Trade Tariffs has led to a Gold Rush that could accelerate the delisting of the U.S. dollar as the global reserve currency 🧈

• You don’t need to be Ray Dalio to see the problem in the above chart.

• You mess with China and well, they buy gold. Not good!

This is not Good News for U.S. Fiscal Stability 🚨

The U.S. isn’t just behind on renewable energy and its power grid, but on producing the crucial rare earth metals needed for things like EVs, robotics and national defense technologies. We’ll know more about the trade deal with China hopefully later today or this week.

What a spender, what a dreamer! Trump’s AI plan, outlined in the July 2025 “Winning the AI Race” Action Plan, is based on three pillars: accelerating innovation, building American AI infrastructure, and leading in international diplomacy and security. A fine job he’s been doing as the handmaiden of others. A lot of talk and debt, if you ask me. A geopolitics of imperialistic trade policy and where AI Infrastructure colonialism overshadows democratic values and traditional American allies (like neighbors! 🍁). This fractured geopolitics and elitist leadership also has risks. Where is the liberty and fiscal, industrial and economic leadership in the AI bubble? A world where AI Infrastructure build-outs props up American GDP while the downstream consumer is mangled?

In 2026, I suspect it’s going to get pretty awful.

Major Signs of Flawed Economic Leadership 👑 🇺🇸

Free Speech? Trump thinks he can clamp down on critical media. Is that really what Americans want? How do you even trust this older generation in power when even economic data is fudged with a wishful AI policy? Innovation isn’t built on political deception, nevermind that immigration policy which torpedos talent and that science funding that is plummeting is no accident. The scaffolding of how America’s significant capital advantage is being allocated is a ruse or a dynastic order and magnitude.

If U.S. voters aren’t happy with the economic conditions of the Trump Administration’s leadership, weird things result - like a Mamdani coup. This is reactionary democracy. But who can blame them if they (Young New York voters - Zohran Mamdani is 34, Trump is 79) are witnessing the collapse of global order? What’s going on in the world is bigger than a “No Kings” protest.

As of September, 2025 we learned that - in the first half of 2025, AI-related capital expenditures contributed 1.1% to GDP growth, outpacing the U.S. consumer as an engine of expansion.

If the U.S. loses “Fed independence” like the Trump Administration threatens, it could be very bad for markets. So many new risks in the future of the United States that threatens rule of law and economic stability including an actual cliff of robotic automation that was planned all along.

How much will Debt financing around AI Infrastructure surge the U.S. National Debt Crisis?

The U.S. is facing the prospect of an Energy (power, electricity) wall and debt crisis convergence simultaneously in the last 2020s or early 2030s driven by the exponential demand for compute supercycle BigTech capex (in datacenters) and that the Trump Administration acceleration has produced. And why is nobody talking about this?

🎯 Thesis: We are underestimating how exponential the demand for compute will become in the 2026 to 2036 period.

“Reasoning AI, demanding 100-1000x more compute than simple chatbots, faces a bottleneck: GPUs optimized for parallel training struggle with sequential inference workloads. NVIDIA projects a $3-4T AI infrastructure market by 2030, where inference capabilities and cost economics will be the determinants of AI application viability at scale.” - Jason Saltzman, CB Insights.

Venture Capital funding in AI startups and more compute and interference intensive reasoning, video, deep thinking and AI agententic tools will super charge demand for compute faster than hyperscalers can build capacity. They will need to take out huge amounts of debt just to keep up!

This is the biggest problem for the United States, Silicon Valley and Wall Street to consider moving forwards into 2026.

What is a Datacenter right?

Data centers are large buildings that house rows of computer servers, data storage systems and networking equipment, as well as the power and cooling systems that keep them running. This infrastructure is essential for companies that provide digital services and increasingly: AI products being heralded as impacting the everything-future. Over the top sales much? 😎

BigVC - Major American Venture Capital is Pumping up Demand for Compute

More AI Startups, more Models and more AI products means drastically more demand for compute. 🚀

An Inference Age you say?

The age of “inference” will drive up compute exponentially. Investors poured over $9.5B into AI inference and training processor startups in 2024 alone:

• Groq

• Rapidus (Japan)

• Cerebras (They have delayed their IPO)

The ‘Nvidia of China’ is Rising

• Cambricon a.k.a. ‘China’s Nvidia’ says revenue spiked 14-fold last quarter.

Cambricon Technologies, China’s prominent AI chipmaker often likened to Nvidia, has been added to the U.S. Department of Commerce’s Entity List over national security fears that its chips could support China’s military modernization. This move restricts the firm’s access to U.S. technology, potentially stalling its AI development and heightening U.S.-China tech tensions.

If Cambricon or Huawei actually do eventually emerge as a threat to Nvidia’s monopoly on AI chips, it’s going to be a problem for the United States and their first-mover AI advantage. isn’t wrong: (AI policy and the AI chip export bans on China have created situations that they wanted to prevent). Are the folk in the Trump Admin really China Hawks? 🤯 Serious question though.

AI Export Ban on AI chips has Led to More Dangerous Competition

The future doesn’t work in gated regimes and unilateral trade tariffs don’t build allies. So is this a turning point?

Read more

Good Morning,

In October, 2022 the U.S. decided it was a good idea to impose export controls on advanced semiconductors and AI chips on China. Three years later in 2025, it would appear that these sanctions became a catalyst for China to accelerate its own semiconductor supply-chain and relevant AI chip startups as rapidly as possible.

This could turn out to be the biggest strategic mistake in AI policy ever if I’m not mistaken. While there are plenty of Newsletters now on China, trade, geopolitics and technology, it’s sometimes good to get an insider source and analysis. I asked (formerly TechTarget, Synced, etc…) who is an analyst in Beijing to look into this with greater detail.

At a time of trade uncertainty and an AI boom, this topic is a major turning point for the future of AI. The Biden administration’s Bureau of Industry and Security (BIS) enacted immediate rules targeting advanced computing chips that took effect October 21st, 2022. Now a full three years later it appears Nvidia’s AI chips have no future in China (easily the world’s 2nd biggest AI market).

I think Jensen Huang understands the situation very well, speaking at a Citadel Securities event on October 6, 2025 the Nvidia CEO blamed U.S. export controls and said the company now assumes no revenue from China in its forecasts.

“We are 100% out of China. We went from 95% market share to 0%. I can’t imagine any policymaker thinking that’s a good idea… whatever policy we implemented caused America to lose one of the largest markets in the world to 0%”. - Nvidia, CEO.

Nvidia’s marketshare in China might end up being Zero Percent

It’s not so long ago China was 20% to 25% of Nvidia’s data center revenue. As the AI arms race heats up China’s market size for AI chips is likely to grow fast as they accelerate AI Infrastructure with a superior energy and especially their renewable energy stack.

With U.S. China trade tensions in full force, China’s doesn’t just have a dominant physical AI & robotics automation sector and a world leading open-model ecosystem with implications in Agentic AI products, it has what calls an “emerging export control regime”.

• Trade | Geopolitics | Energy | AI | The Future | Semiconductor Stacks

Nvidia has a commanding lead as an AI chip monopoly the last few years, but I think it’s wise to listen carefully to Jensen Huang’s recent interviews to get a sense of the magnitude of the situation. Nvidia CEO Jensen Huang said in early October, 2025 that the U.S. is “not far ahead” of China in the artificial intelligence race, and that the country needs a “nuanced strategy” to stay on higher ground.

China’s Energy, Renewable and Infrastructure Prowess could lead to AI domination

As for China’s energy advantage in the AI Infrastructure build out to come, China generated a total of 10,000 terawatt hours, or one trillion watt hours, of electricity in 2024, over double the U.S.’s output, according to the Energy Institute.

Few China tech analysts in recent weeks have impressed me as much as with his detailed coverage and ability for synthesis at a very deep level:

• Learn more about the Newsletter.

Subscribe to HCT Premium

Selected Works

Poe Zhao’s ability to tease out details many Western analysts and current China tech analysts miss is truly noticeable.

• China’s Answer to Silicon Valley: Alibaba’s $53 Billion Challenge to Big Tech

• The Architecture Gambit: Alibaba’s Bet on RISC-V

• The Billion-Dollar Bonfire: China’s AI War and the Creation of the “Do-Engine”

Poe Zhao is a Beijing-based technology analyst who builds rigorous, bilingual explanations of China’s tech ecosystem for global readers. He founded Dailyio (Chinese) and Hello China Tech (English) to map how China’s AI, robotics, semiconductors and EVs actually scale—from policy design and state capital, to supply-chain execution and market adoption. His editorial approach favors verifiable signals over hype: structural data points, filings, and strategy moves reframed into a coherent narrative of “parallel development,” not copycat catch-up.

Hello China Tech

Decoding China tech for global minds. China is quietly reshaping the future of AI, electric vehicles, robotics, and semiconductors. Poe Zhao expertly (doing this for 12 years) bridges the gap between China’s rapidly evolving tech ecosystem and the global community. (Read the coverage in Mandarin).

Hello China Tech

Decoding China tech for global minds

By Poe Zhao

How China Built a Parallel AI Chip Universe in 18 Months

October, 2025 by .

Nvidia lost $2.5 billion in quarterly revenue from China and faces continued market pressure. That’s just the beginning.

Why This Isn’t About Nationalism (It’s About Money)

The phrase “去英伟化” (de-Nvidification) has become ubiquitous in Chinese tech circles. But it risks obscuring what’s actually happening.

Three forces are at work. Distinguishing between them matters.

Economics is the primary driver, not geopolitics.

Here’s the business case. Training GPT-4-class models costs hundreds of millions of dollars. Computing represents 60-70% of those expenses. When Chinese companies evaluate switching core model training to domestic chips, they cite substantial potential cost savings–combining lower chip unit costs with software-hardware co-optimization gains–the business case becomes compelling.

When Alibaba spent over 380 billion yuan on AI infrastructure over three years, even 30-40% savings justify billions in chip R&D. This is cold calculation, not patriotic fervor.

Read Insights

Editor’s Favorites:

• Moore Threads’ Record IPO: Blueprint or Gamble?

• Hangzhou’s Contingent Coordination: When Improvisation Becomes Strategy

• Cambricon: A Bet on AI’s Narrowest Edge

Geopolitics is the catalyst, not the cause. 🌍

U.S. export controls didn’t create the economic incentive. They eliminated the option of continuing with existing suppliers. The progression tells the story: restricting H100 GPUs in 2022, temporarily suspending Nvidia’s chip support services in early 2025, and critically, three major EDA providers (Synopsys, Cadence, Siemens) collectively pausing software support in May 2025.

When the tools to design advanced chips became unavailable, Chinese companies faced a binary choice. Accept permanent technological subordination or invest massively in alternatives.

Here’s the crucial distinction: geopolitics didn’t make domestic chips economically attractive. It made experimenting with them necessary. Without export controls, most Chinese companies would have continued optimizing within Nvidia’s ecosystem. They knew alternatives might offer better long-term economics. But the disruption forced the experiment that economic logic alone wouldn’t have triggered.

Strategic control creates compounding advantages.

Once companies invest billions in vertical integration–designing chips, building frameworks, deploying at scale–they’re not just buying computing hardware. They’re constructing platforms that capture long-term value. The optimization is continuous. It’s impossible with third-party purchases.

When Huawei’s MindSpore team discovers performance bottlenecks, they can request Ascend hardware modifications. When Baidu identifies that certain transformer operations dominate LLM inference, Kunlun’s next generation adds specialized accelerators. This feedback loop creates compounding advantages over time.

Understanding this causal chain matters.

If geopolitics were primary, we’d expect Chinese chips adopted despite economic disadvantages. Performance gaps would be tolerated for supply security. If economics were everything, companies would have pursued domestic chips earlier when technically feasible.

The reality: economic logic opened the possibility, geopolitical pressure forced the experiment, strategic benefits are creating lock-in. This isn’t nationalism. It’s calculated adaptation to changed constraints.

Three 2025 breakthroughs illustrate this dynamic crystallizing:

Cambricon Technologies | Baidu | Alibaba

1. Cambricon Technologies reported H1 2025 revenue of 2.88 billion yuan. That’s a 4,347% year-over-year increase. This wasn’t gradual adoption. Customers dependent on Nvidia suddenly needed alternatives. They found Cambricon’s Siyuan 370 chips offered acceptable performance for specific workloads at lower cost.

2. Baidu deployed a 30,000-card Kunlun P800 cluster. First Chinese company operating domestic AI accelerators at hyperscale for LLM training. More telling: Baidu’s Intelligent Cloud won a 1 billion yuan China Mobile contract for AI inference servers. Kunlun chips captured all three sub-packages in the “CUDA-like ecosystem” category.

When state-owned enterprises make large-scale procurement based on technical validation rather than policy directives, a threshold has been crossed.

3. CCTV footage in September 2025 inadvertently revealed Alibaba’s new training chip. Codenamed PPU. Using domestic 7nm process and 2.5D Chiplet packaging. Specifications reportedly matching H100 at ~40% lower costs. When China’s tech giants commit 380 billion yuan to AI infrastructure, they’re betting on domestic chips as long-term foundations, not stopgaps.

But here’s the critical distinction that separates hype from reality.

Chinese AI chips are achieving workload-specific maturity rather than general-purpose equivalence. Baidu’s Kunlun cluster is proven for LLM training. But it’s not yet validated across computer vision, reinforcement learning, scientific computing. China Mobile’s procurement validated telecom-specific AI workloads–intelligent customer service, network optimization. Not arbitrary applications.

This gap between narrow competence and broad capability is central to understanding both genuine progress and remaining limitations. It’s the difference between “we can do this specific thing well” and “we can do anything you might want.”

Figure 3: Workload-specific maturity landscape. Chinese chips excel in deeply optimized domains (Kunlun for LLM training, Cambricon for telecom AI) but remain unproven for general-purpose applications. This distinction separates genuine competitiveness from aspirational claims.

Why Chinese Tech Giants Are Building Everything In-House

Perhaps the most striking pattern–the one that most clearly differentiates China’s approach from Western models–is vertical integration everywhere.

Tech giants build complete stacks from silicon to cloud services. Contrast this with Western specialization. Nvidia provides chips. OpenAI builds models. AWS/Azure/GCP offer cloud infrastructure through modular collaboration.

Figure 4: Two strategic architectures. Western model relies on modular collaboration with best-of-breed specialists. Chinese model pursues full-stack vertical integration driven by supply vulnerability, cost compounding, and continuous optimization advantages.

Why does vertical integration dominate in China but not the West?

Supply chain vulnerability made single-supplier dependence existentially risky.

When Nvidia’s quarterly China revenue drops 50% not from demand reduction but supply restrictions, continuation risk becomes impossible to ignore. Western companies can assume continued access to best-of-breed components. Chinese companies cannot.

Cost advantages compound at scale.

The 40% chip unit cost advantage multiplies when combined with framework-level optimizations. Baidu’s claimed 80% total cost reduction comes from both cheaper hardware and software-hardware co-design. The co-design eliminates inefficiencies in memory management, operator scheduling, and distributed communication that generic GPUs cannot address.

A company spending $50+ billion annually on AI infrastructure can justify $2-3 billion in chip development if it yields 30-40% ongoing savings. Smaller companies cannot. Which is why independent chip designers like Cambricon focus on selling to customers lacking scale for vertical integration.

Ecosystem control enables continuous optimization. This creates compounding advantages impossible with third-party purchases.

Three distinct strategic approaches have emerged:

Huawei’s Ascend ecosystem represents the most ambitious vertical integration.

Designing Ascend 910 chips (256 TFLOPS at 7nm). Developing MindSpore framework. Operating Huawei Cloud as an integrated platform. They gained significant market share by late 2024. The announced three-year roadmap shows next-generation Ascend 950 chips incorporating FP8 precision, heterogeneous architecture, and a proprietary high-speed interconnect called “Lingqu” (灵衢) to enable million-card scale clusters.

Huawei’s bet: parallel ecosystem development rather than CUDA emulation. MindSpore introduces its own programming model optimized for distributed training and automatic parallelization. The assumption: if Ascend deployment reaches sufficient scale domestically, developers will learn MindSpore through necessity. Just as they once learned CUDA.

This trades international portability for domestic optimization. It’s viable within China’s large protected market. But faces challenges for global adoption.

Baidu’s approach emphasizes compatibility layers to reduce switching costs.

Kunlun P800 achieves competitive specifications. PaddlePaddle maintains PyTorch and TensorFlow interfaces. Researchers can port models with minimal code changes. Then optionally use Kunlun-specific APIs for performance tuning. The tradeoff: engineering complexity. Maintaining multiple frontend interfaces while optimizing backends demands massive resources.

Baidu’s advantage is demonstrable maturity in the specific domain of LLM training and inference. The company has operated PaddlePaddle since 2016. Trained successive generations of ERNIE models. When announcing its 30,000-card Kunlun deployment, Baidu wasn’t launching untested hardware. It was commercializing infrastructure already supporting production workloads internally.

The China Mobile contract validated that this narrow but deep capability was sufficient for specific enterprise use cases.

Alibaba’s strategy adds manufacturing economics optimization.

Beyond Hanguang 800 inference chips deployed since 2019 for e-commerce search and recommendation, the revealed PPU training chip uses domestic 7nm process and 2.5D Chiplet packaging. The bet: achieving H100-comparable specifications while maintaining ~40% lower costs through domestic fabrication and packaging creates sustainable competitive advantage.

Alibaba plans 380 billion yuan investment over three years. Core AI model training will gradually shift to self-designed chips.

Tencent chose the opposite path.

Explicitly not designing chips. Instead pursuing “open compatibility, diverse computing power.” Tencent Cloud comprehensively adapted mainstream domestic AI chips. Developed “Xingmai” (星脉) high-performance networking to improve communication efficiency across heterogeneous accelerators. Claims nearly 10x improvement in thousand-card cluster performance.

This offers initial flexibility. But may prove disadvantageous in cost competition as rivals optimize complete stacks and capture integration benefits.

The Next 18 Months Will Decide Everything 🚨

Read more

Today’s article goes into our Datacenter section of the Newsletter.

I have been following BigTech Earnings closely for Cloud growth vs. AI capex to better understand the state of datacenters in 2025. Keep in mind while Microsoft and Amazon have pulled back from Datacenter plans, the cost of AI infrastructure is also going up with trade tariffs presenting a rather confusing puzzle:

Image by of Generative Value Newsletter.

Meanwhile, McKinsey also put out a rather baffling report about the costs of AI Infrastructure by 2030:

Read the Report

Infrastructure Demands strain the World’s Capability

• Their research shows that by 2030, data centers are projected to require $6.7 trillion worldwide to keep pace with the demand for compute power. 🌏

• Data centers equipped to handle AI processing loads are projected to require $5.2 trillion in capital expenditures, while those powering traditional IT applications are projected to require $1.5 trillion in capital expenditure.

• US energy demand will surge over the next few years, with the sector facing six critical challenges. This could accompany surges in the cost of energy and inflation over global tariffs.

So the reason this is alarming is the uncertainty of it all - according to the Register: “A report from consultancy McKinsey & Company highlights the widespread unease over AI, pointing to the bewildering sums being invested into infrastructure to support it, while warning that forecasts of future demand are based on little more than guesswork.”

AI Capex is Accelerating Cloud Growth Modestly

Eric Flaningam’s analysis shows that combined, the three biggest cloud providers (AWS, Azure, GCP) are at a ~$247B run rate, growing 23.9% Y/Y. This is in part boosted by AI adoption and the Cloud growth has to be factored into the ROI from these massive datacenter investments. For the three main cloud leaders of Microsoft, Google and Amazon - this is massive growth, sped up by AI to a considerable degree.

However Amazon’s AWS growth is slowing down of the three the most: AWS revenue grew just 17% during the quarter. But it’s also the biggest.

So what conclusions can we draw from McKinsey’s analysis of the future of AI Infrastructure? It admits these projections are highly speculative.

Sam Altman’s 7 Trillion Quote finally Makes Sense

In February 2024 Sam Altman made some claims that the media took a bit out of proportion about OpenAI’s Stargate ambitions, now it finally makes sense. However it’s still such an enormous sum the energy bottleneck and who can follow are major problems.

McKinsey: “Overall, that’s nearly $7 trillion in capital outlays needed by 2030—a staggering number by any measure.”

Is BigTech going to subsidize access to AI of the future? And who will be able to follow?

• To meet this demand, companies across the compute power value chain run the risk of over-extending themselves.

Read the Full Report

• If you are an investor in BigTech or the “Magnificent seven”, you might want to read this.

Read more

Welcome back,

Today we have a guest post from the foremost experts on TSMC writing on Substack, Semi Vision, currently ranked 12th in trending rising Newsletters in tech on the in-app leaderboard.

Introducing SEMI VISION

Their paid tier is $25 a month, fairly normal for their specialized research and area of expertise.

SEMI VISION

SemiVision is a semiconductor industry research institution committed to transforming analysis into vision through comprehensive insights into the semiconductor ecosystem.

By SEMI VISION_TW

The report today is titled “TSMC in Strategic Focus: U.S. Expansion, Potential Intel Collaboration, and Navigating U.S.-China Tech Tensions”.

Why TSMC is so Crucial to the future of AI

As the U.S. vs. China trade war escalates the true “picks and shovels” company for AI Supremacy isn’t Nvidia, it’s TSMC. Taiwan Semiconductor Manufacturing Company (TSMC) has committed a total investment of approximately $165 billion in the United States complicating the geopolitical future in the era of reciprocal trade tariff uncertainty.

TSMC is the most important tech company in the world in 2025.

Share

Under the Trump Administration, TSMC has promised epic expansions in Arizona, exasperated by tariff threats TSMC had no choice but to capitulate to Trump. The U.S. needs TSMC to stay competitive in national defense and AI, while TSMC could be on the verge of becoming the most important company in the world as it leads its rivals into unprecedentedly small and efficient semiconductor chips critical for the future of AI.

A threat to the island nation of Taiwan (e.g. like a blockade or invasion) and supply-chains of TSMC, would immediately plunge the global economy into a severe recession. It would also likely spark a hot war involving the U.S., Japan and other allies. The Trump administration appears to be escalating the situation:

Subscribe to SemiVision Newsletter

With the threats of tariffs, TSMC’s further investment in Arizona could mean this area becomes a center for the advanced technology to power artificial intelligence, sophisticated military systems, computers, electric vehicles and cellphones with a massive planned addition to a semiconductor complex in north Phoenix.

While revenue continues to surge at TSMC, Trump’s threat of tariffs on semiconductors and significant tariffs on Taiwan, have forced the island nation to make investments it may not easily be able to keep.

It’s Complicated

Many market analysts have raised concerns that much higher production costs in the U.S. will erode TSMC's profit margin and affect its cash dividend payout. According to Bloomberg, TSMC is warning of its inability to keep its chips outside of China. Meanwhile Huawei plans to begin mass shipments of its own advanced AI chip — the 910C — as early as next month, a supposed replacement for Nvidia’s H20 chips. On April 15th, Nvidia said that it will take a quarterly charge of about $5.5 billion tied to exporting H20 graphics processing units to China and other destination since on April 9, the U.S. government told Nvidia it would require a license to export the chips to China and a handful of other countries.

The Semiconductor Stock Crash of 2025

On the stock market, the semiconductor stocks are imploding as of April, 2025. Nvidia is down 30% and TSMC is down nearly 27% so far this year with the entire sector in turmoil, but the worst may be still yet to come for investors. TSMC’s biggest customers are the likes of Nvidia, Apple, AMD, Qualcomm, Broadcom and others. There are many good reasons I’m calling TSMC the most important tech company in the world.

I hope this report sheds important details on why this is the case in this peculiar geopolitical uncertainty around the U.S. vs. China, export controls and the critical future of the semiconductor industry. In my Newsletter Semiconductor Things, I delve into the news cycle around this industry.

Semiconductor Things™

This Newsletter was built to solve the pain point of getting the latest news on A.I. chips, semiconductors, datacenter innovation and chip news.

By Michael Spencer

At this point I’ll turn it over to the experts:

What is SemiVision Newsletter?

Read More

SemiVision is a semiconductor industry research institution committed to transforming analysis into vision through comprehensive insights into the semiconductor ecosystem. The semiconductor industry’s future is shaped by cutting-edge innovations such as AI, silicon photonics, advanced packaging, and more. At the heart of this ecosystem lies the interplay between foundries, semiconductor manufacturing, fabless design houses, and ASIC (Application-Specific Integrated Circuit) development—each a critical pillar supporting the chips and technologies that power our world.

SV Works:

I consider SemiVision critical reading for investors, China watchers, U.S. AI policy types and analysts in the Semiconductor space as of 2025, among others.

• NVIDIA x TSMC: A Milestone in Silicon Photonics and Optical Integration

• How TSMC’s Silicon Photonics Could Impact NVIDIA’s Future

• 2025 GTC Review: The Silicon Photonics Era Has Arrived – Get Ready!

• The Rise of ASIC: Custom Chips Becoming a Key Trend!

• SemiVision is the fastest growing paid semiconductor Newsletter on Substack as of early 2025.

• By Jett and Eddy, March, 2025 via

The State of TSMC in the Global Order: Report

TSMC in Strategic Focus: U.S. Expansion, Potential Intel Collaboration, and Navigating U.S.-China Tech Tensions

As global semiconductor dynamics are increasingly shaped by geopolitics, Taiwan Semiconductor Manufacturing Company (TSMC) finds itself not only at the forefront of technological innovation but also at the very center of a high-stakes geopolitical chessboard. Amid intensifying U.S.-China tech rivalry and a global push to localize supply chains, TSMC’s strategic decisions—from its multibillion-dollar investment in Arizona, to its rumored advanced packaging collaboration with Intel, to its response to political pressures from Washington—are reshaping the semiconductor landscape and influencing the global balance of technology power.

This article offers a deep dive into TSMC’s current strategic positioning through five key lenses: (1) the scope and implications of its U.S. capacity expansion and how it manages geopolitical risks; (2) the potential cooperation model with Intel on advanced packaging or shared manufacturing capacity; (3) the increasing pressure from the U.S. government regarding transparency, capacity allocation, and alignment with national interests; (4) its financial performance and capital expenditure outlook amid global uncertainty; and (5) the status and competitive edge of its 2nm and sub-2nm process technology roadmap. Together, these dimensions reveal not only the complexity of the challenges facing TSMC, but also the resilience and adaptability required for it to sustain its leadership role in an evolving and fragmented global semiconductor supply chain.

Subscribe now

• This report is the most in-depth report on TSMC I have personally read. A good read for investors, China watchers, U.S. policy experts, venture capital talent and semiconductor workers at large.

• It breaks down the most important details in the U.S.-China and U.S-Taiwan implications in tariffs and trade considerations.

• The report goes into unprecedented details about why TSMC is so important in the global order from and by experts in located in Taiwan itself.

I. Introduction: A Technological Leader and a Geostrategic Actor

Taiwan Semiconductor Manufacturing Company (TSMC) has long been the undisputed global leader in semiconductor foundry services. Its dominance is rooted in a powerful combination of cutting-edge process technologies, industry-leading yield control, massive production scale, and deep integration with key customers. This competitive moat has enabled TSMC to secure and sustain its leadership position—especially in advanced nodes critical to next-generation computing.

Yet in the current era, technical excellence alone is no longer sufficient. As global tensions escalate and countries prioritize “de-risking” their supply chains, semiconductors have evolved from commercial products into strategic national assets. In this context, TSMC has been thrust to center stage—not only as an industrial powerhouse but as a pivotal player in the unfolding geopolitical realignment.

In 2024, as AI-driven workloads surge, capital investments shift toward advanced manufacturing, and U.S.-China technological rivalry continues to intensify, TSMC stands at the crossroads of technological innovation and strategic transformation. Every capital expenditure, fabrication site decision, and international partnership it undertakes now carries global implications for semiconductor supply chain resilience and security.

As the supplier behind over 90% of the world’s advanced-node (under 18nm) semiconductor manufacturing, TSMC is indispensable to mobile SoCs, AI accelerators, and high-performance server platforms. Its influence is also expanding into defense, autonomous vehicles, and critical cloud infrastructure. What was once primarily a matter of technical leadership has now become a symbol of trust, strategic leverage, and geopolitical balance in the global technology order.

Since the outbreak of the U.S.-China trade war in 2018, the global semiconductor landscape has undergone dramatic shifts—with TSMC emerging as a central figure in the geopolitics of technology. As the U.S. imposed sweeping export controls on Chinese tech firms and nations worldwide accelerated efforts to localize chip production, the global supply chain entered a period of accelerated restructuring. Amid this upheaval, TSMC not only maintained its leadership but achieved significant breakthroughs in both technology and market capitalization. At its peak, TSMC’s valuation surpassed $600 billion, making it one of the most valuable technology companies in Asia and a bellwether of global semiconductor performance.

This section explores the three key forces behind TSMC’s remarkable growth trajectory—analyzing the turning points in its market capitalization, the structural drivers of its sustained advantage, and its strategic resilience in adapting to future disruptions.

Three Core Growth Drivers

1. Technological Leadership: Global Domination in 7nm, 5nm, and 3nm Nodes

TSMC has built its long-standing lead by consistently advancing its process technology roadmap. Its 7nm node marked a major inflection point, while the 5nm process became the foundation for next-generation chips from Apple, NVIDIA, AMD, and others. With the 3nm (N3) node, TSMC became the first foundry to commercially scale a gate-all-around (GAA) transistor architecture. By 2024, TSMC controls over 90% of the global sub-7nm foundry market, creating both a technical and capacity monopoly. This dominance makes TSMC indispensable for high-end computing, smartphone SoCs, and AI accelerators—driving robust revenue growth and industry-leading gross margins. Crucially, it also reinforces “technological lock-in” with its customers, turning TSMC into a high-barrier strategic asset in the eyes of global investors.

TSMC remains the only semiconductor foundry in the world capable of reliably and consistently mass-producing advanced process nodes below 7 nanometers. Since the introduction of 7nm, the company has outperformed all competitors in yield optimization, cost control, and time-to-volume execution—positioning itself as the foundry of choice for the world’s most demanding semiconductor clients. These include Apple’s A-series and M-series processors, NVIDIA’s H100 and GH200 AI accelerators, and AMD’s Zen-based CPUs. This form of manufacturing monopoly at the leading edge has enabled TSMC to generate exceptional profit margins and exert significant pricing power within the global foundry market. As a result, TSMC not only holds a dominant position in AI, high-performance computing, and mobile SoCs, but is also increasingly viewed by global investors as a strategic, irreplaceable asset within the semiconductor value chain.

2. Explosive Growth in AI and High-Performance Computing (HPC) Demand

The rapid proliferation of generative AI models, edge computing, and next-generation wireless technologies such as 5G and 6G has triggered a global surge in demand for high-performance semiconductor solutions. From large-scale cloud GPUs to edge AI SoCs, virtually every advanced computing workload today relies on TSMC’s leading-edge process technologies and advanced packaging capabilities. Key customers include global tech giants such as NVIDIA, Google, Meta, and Amazon. TSMC’s leadership in advanced packaging—particularly in CoWoS, InFO, and SoIC—positions it as the dominant provider for high-bandwidth memory (HBM) stacking and large-area AI chips, cementing its role at the heart of the global AI/HPC ecosystem.

3. Strategic Neutrality and Supply Chain Trust in a Geopolitical World

Amid escalating U.S.-China tech rivalry and broader geopolitical uncertainty, semiconductor buyers are increasingly prioritizing not only technical performance, but also supply chain resilience and political neutrality. TSMC’s unique position as a “non-China, non-U.S.” supplier gives it a rare form of geopolitical trust, making it the foundry of choice for multinational tech firms seeking dependable long-term supply. From U.S. CHIPS Act subsidies for its Arizona fabs, to invitations from Japan and the EU to establish local production, TSMC is now a centerpiece of national semiconductor strategies across multiple regions—transforming geopolitical pressure into global opportunity.

Outlook: From Process Leader to Infrastructure Backbone for AI

By 2025, TSMC is expected to begin mass production of its 2nm node featuring gate-all-around (GAA) transistor architecture. It is also advancing its next-generation back-end integration capabilities to meet the extreme bandwidth and efficiency demands of upcoming AI platforms such as NVIDIA’s GB200 and AMD’s MI400. Furthermore, TSMC’s investment in photonic integration—particularly in co-packaged optics (CPO) and Optical I/O—signals its ambition to become a core enabler of next-gen AI system architectures. If TSMC can drive ecosystem-wide upgrades while overcoming challenges in cost and thermal performance, it will evolve beyond foundry leadership to become a global infrastructure pillar for AI computing.

R&D Investment Strategy

To safeguard its long-term competitiveness and innovation capacity, TSMC continues to aggressively invest in research and development:

· Over $6 billion in R&D expenditure in 2023, accounting for nearly 9% of total revenue.

· Focus areas include GAA transistor development, 3D IC integration, and silicon photonics platform technologies.

· Geographic diversification of innovation hubs: expanding R&D and manufacturing presence in Arizona (U.S.), Kumamoto (Japan), and Dresden (Germany) to enhance global supply chain stability and regional trust.

II. U.S. Expansion: Strategic Move or Political Response?

TSMC has committed over $65 billion USD to the phased construction of three advanced semiconductor fabs in Arizona, targeting 4nm, 3nm, and upcoming 2nm process nodes. This initiative marks the company’s largest overseas manufacturing investment to date. In early March 2024, TSMC CEO C.C. Wei led a delegation to the White House, where he announced an additional $35 billion in investments, bringing the total to $100 billion. This expanded plan includes three more fabs, a dedicated R&D center, and two advanced packaging facilities—signaling a major escalation of TSMC’s localization strategy in the United States.

While TSMC officially attributes this expansion to customer demand and the need to strengthen supply chain resilience through geographic diversification, many observers interpret the move as being heavily influenced by political pressure from the U.S. government. Under both the Trump and Biden administrations, Washington has encouraged—or compelled—strategic reshoring of semiconductor manufacturing. From “Made in America” policies to the CHIPS and Science Act’s incentive packages and national security clauses, the U.S. has created both carrots and sticks to guide corporate decisions.

In this light, TSMC’s Arizona investment reflects not only a calculated global strategy but also a response to intensifying geopolitical pressure. It underscores the company’s delicate position at the intersection of supply chain neutrality, customer alignment, and sovereign state interests—balancing innovation leadership with the growing weight of political expectations in an era of tech sovereignty.

Since 2020, TSMC’s presence in the United States has rapidly evolved from a single production site to a comprehensive semiconductor ecosystem encompassing advanced fabrication, heterogeneous packaging, and front-end R&D. This strategic expansion aligns with the U.S. CHIPS and Science Act’s push for domestic advanced-node capabilities and signals TSMC’s ambition to consolidate its leadership in the global AI-driven semiconductor supply chain.

TSMC’s U.S. Expansion Blueprint (2020–2025): From Manufacturing to Innovation Hub for the AI Era

Since 2020, TSMC’s presence in the United States has rapidly evolved from a single production site to a comprehensive semiconductor ecosystem encompassing advanced fabrication, heterogeneous packaging, and front-end R&D. This strategic expansion aligns with the U.S. CHIPS and Science Act’s push for domestic advanced-node capabilities and signals TSMC’s ambition to consolidate its leadership in the global AI-driven semiconductor supply chain.

Key Milestones and Investment Timeline:

May 2020: Initial U.S. Fab Announcement

• TSMC announces a $12 billion investment to build its first U.S. fab (Fab 21) in Phoenix, Arizona, targeting 5nm production with an expected monthly capacity of 20,000 wafers.

• Context: A direct response to U.S.-China trade tensions and Washington’s strategic push to localize advanced semiconductor manufacturing.

December 2022: Investment Rises to $40 Billion

• TSMC commits to a second fab (Fab 22), introducing 3nm production. The total investment rises to $40 billion, marking the largest foreign manufacturing investment in U.S. history.

• Impact: Combined facilities expected to employ over 4,500 high-tech workers, signaling a permanent shift toward U.S.-based advanced chip production.

April 2024: U.S. Funding Secured + Third Fab Planned

• The U.S. Department of Commerce awards:

　– $6.6 billion in grants

　– Up to $5 billion in low-interest loans

• Funds will support construction of a third fab featuring 2nm node technologies, increasing the total committed investment to $65 billion.

March 2025: Massive Expansion to $165 Billion Including R&D

• TSMC announces an additional $100 billion in new investment, bringing the total to $165 billion.

• Expanded scope includes:

　– Three fabs (5nm / 3nm / 2nm)

　– Two advanced packaging facilities (CoWoS, SoIC-ready)

　– One U.S.-based R&D center

Strategic Role of the R&D Center: From Localization to Innovation Leadership

The R&D center represents TSMC’s first full-scale research hub in the U.S.—marking a shift from manufacturing localization to frontline innovation. The center will focus on:

· Next-generation process nodes (including GAA and sub-2nm scaling)

· 3D IC integration and packaging (CoWoS, SoIC)

· Specialized AI chip materials, electrical simulation, and signal integrity testing platforms

· Silicon photonics and co-packaged optics (CPO) process development

This initiative is expected to deepen collaboration with leading U.S.-based chip designers such as NVIDIA, AMD, and Google—fostering a joint innovation model that integrates architecture and process R&D, accelerating the co-evolution of AI system performance and semiconductor manufacturing capabilities.

Geopolitics and Supply Chain Risk Management: From Pressure to Strategic Repositioning

Amid rising geopolitical tensions and the global trend toward technological sovereignty, TSMC’s U.S. expansion has become more than a matter of capacity—it is a deliberate response to supply chain security and policy-induced risk.

• U.S. Political Pressure Revealed: In 2025, former President Donald Trump publicly stated that his administration had explicitly pressured TSMC to build a fab in the U.S., warning that failure to comply could result in 100% punitive tariffs on its products. This underscores how TSMC’s Arizona investment was shaped not only by market logic, but also by strategic coercion—reflecting Washington’s view of advanced semiconductors as national security assets.

• Clarification from China on Tariffs: Also in 2025, the China Semiconductor Industry Association (CSIA) issued a statement clarifying that U.S.-designed, Taiwan-manufactured chips would continue to be recognized as Taiwan-origin goods—thus exempt from China’s retaliatory tariffs. This helped reassure global customers and highlighted TSMC’s rare value as a geopolitically neutral supplier in an increasingly fragmented world.

Arizona’s Fab 21 Enters Mass Production: A Dual Milestone in Technology and Strategy

In April 2025, TSMC’s first Arizona fab—Fab 21—officially entered mass production of 4nm chips, marking a watershed moment for U.S.-based advanced semiconductor manufacturing. For TSMC, this milestone validates its localization strategy. For the U.S., it represents the tangible return of cutting-edge chip production to American soil.

Conclusion: The U.S. as TSMC’s Next Global Strategic Stronghold

TSMC’s U.S. operations have evolved into a strategic semiconductor campus spanning leading-edge fabrication, advanced packaging, and front-end R&D. While the project reinforces U.S. ambitions for semiconductor autonomy and AI infrastructure, it also elevates TSMC’s geopolitical importance on the world stage. However, this transformation comes with three major challenges:

1. High Production Costs: Building and operating fabs in the U.S. is estimated to cost 2–3x more than in Taiwan, pressuring profitability.

2. Talent and Experience Gaps: U.S. engineers lack deep experience in volume chip manufacturing, requiring extensive training and cultural adaptation.

3. Immature Supply Ecosystem: The U.S. still lacks the dense, mature supplier networks—especially in EUV tool maintenance, advanced material logistics, and spare parts delivery—that make TSMC’s Taiwan operations so efficient.

Another major challenge lies in establishing long-term trust with U.S. government agencies, local authorities, and labor unions, while maintaining tight control over process secrecy and ensuring cultural alignment across regions.

Nonetheless, the strategic value of the U.S. market—and its strong political will to revive domestic chip production—has made this investment an inescapable move for TSMC. It now serves not only as a manufacturing site, but as a geopolitical hedge and possibly a bargaining lever in future dialogues with China.

III. TSMC × Intel: Strategic Collaboration Amid Competition?

Recent industry buzz suggests that TSMC and Intel are in advanced discussions on multiple fronts—including wafer foundry outsourcing, advanced packaging integration, and potentially the formation of a joint venture. The collaboration may combine Intel’s leading-edge designs with TSMC’s CoWoS packaging technology, creating a powerful alliance that carries dual strategic significance:

On one hand, Intel, as a traditional IDM leader, is under pressure due to yield and timeline challenges in its latest process nodes. Outsourcing to TSMC enables it to maintain volume ramp-up for products like AI accelerators and data center chips. On the other hand, TSMC benefits from expanding its influence on U.S. soil, while reinforcing ties with American customers and policymakers—a crucial balancing act in the age of supply chain politics.

Customer Co-Investment Model: Consistency Across the U.S. and Europe

In Arizona, TSMC’s major customers—NVIDIA, AMD, Apple, Broadcom, Marvell, Qualcomm, and Intel—have strong stakes in securing advanced node capacity. Reports suggest that some of these firms have participated in fab development via financial contributions, tool commitments, or technical collaboration. This co-investment model helps mitigate upfront CAPEX, while ensuring preferential capacity and closer supplier alignment.

TSMC is replicating this model in Europe as well, most notably in its new Dresden fab project, where Bosch, Infineon, and NXP are partners in a joint venture. This illustrates TSMC’s modular approach to global expansion, blending localization with strategic customer integration.

Technology Leakage Concerns: Different Design Rules, Complementary Strengths

Some in Taiwan have expressed concerns about potential technology leakage from partnering with Intel. However, it’s important to note that Intel and TSMC operate under entirely different design rules. Their process flows, toolchains, and PPA optimization strategies are fundamentally incompatible.

For example, Intel’s upcoming 18A process is expected to enter volume production soon. TSMC internally categorizes it as functionally equivalent to its own A16 node, suggesting technical parity rather than superiority or dependency. Intel’s challenges lie not in technology, but in execution and manufacturing efficiency. Thus, the collaboration is better understood as a complement of manufacturing capabilities, not a transfer of proprietary process IP.

Advanced Packaging and UCIe: Enabling a Modular Chip Ecosystem

For TSMC, this partnership would reinforce its global leadership in advanced packaging. As chiplet architectures and 3D stacking become central to modern chip design, packaging has become a frontline enabler of system performance, power efficiency, and integration flexibility.

Should TSMC and Intel collaborate on open packaging standards such as UCIe, they could accelerate industry-wide adoption of modular, interoperable architectures, making it easier to assemble complex SoCs from diverse components. This could lead to a new wave of composable hardware innovation, bridging chip manufacturing with system-level design freedom.

Still, questions remain: Will this change the long-term competitive dynamics? Intel remains a potential rival in advanced foundry services. Whether this collaboration helps Intel rebuild its manufacturing credibility and ultimately compete for TSMC’s top-tier customers remains a closely watched development.

IV. Political Pressure and the U.S.–China Trade Shadow: Strategic Balancing in a Bipolar Tech World

TSMC’s globalization strategy is increasingly shaped by geopolitical pressure between the U.S. and China. During the Trump administration, the imposition of punitive tariffs and the expansion of the U.S. Entity List blurred the definitions of chip origin and compliance—impacting TSMC’s client structure and shipment flexibility.

Media reports revealed that Trump warned TSMC directly: “If you don’t build fabs in the U.S., your chips will face 100% tariffs.” This public pressure illustrated how TSMC’s U.S. investment decisions were not purely market-driven, but influenced by the necessity to balance economic logic with sovereign coercion.

Despite China remaining a critical market—especially in mobile and consumer electronics—rising geopolitical risk has accelerated TSMC’s diversification of markets and capacity.

Incentives and Pressure: The Dual Nature of the CHIPS Act

The U.S. CHIPS and Science Act offers substantial subsidies to lure semiconductor manufacturing back to American soil. For TSMC, this creates a landscape where government incentives come bundled with structural obligations:

· In the short term, these subsidies help offset high capital expenditures and ease financial burdens.

· In the long term, however, TSMC must navigate challenges related to labor shortages, cultural adaptation, union negotiations, and potential policy volatility—especially if political leadership changes.

Strategic Neutrality in Action: Global Manufacturing as Geopolitical Hedging

TSMC is clearly shifting toward a multi-region manufacturing model to reduce dependency on any single market. This “neutralization strategy” is evident in its investment footprint across Asia, North America, and Europe, building a multi-tiered network for resilience, technology leadership, and regional relevance:

· Taiwan remains the technological and operational heart, with fabs for 5nm, 3nm, and future 2nm nodes, plus advanced packaging (CoWoS, SoIC).

· Arizona, USA is the North American anchor, with $165 billion invested across 3 fabs, 2 advanced packaging facilities, and 1 R&D center, serving customers like NVIDIA, Apple, AMD, and Intel.

· Kumamoto, Japan (JASM): Joint venture with Sony and Denso, with the first 22/28nm fab set for mass production in late 2024 and a second fab planned with more advanced nodes.

· Dresden, Germany (ESMC): Joint venture with Bosch, Infineon, and NXP, focused on 28nm and mature nodes for automotive and industrial markets.

· China (Shanghai & Nanjing): Operating 8-inch and 12-inch fabs at 16/28nm nodes for local market demand.

· Global R&D & Support Centers: Established in Yokohama, Tsukuba, and Arizona, focused on 3D IC, silicon photonics, and heterogeneous integration.

Conclusion: Building Resilience Amid Political Crossfire

In an era of entrenched U.S.–China tech rivalry, TSMC’s global expansion is more than business strategy—it is a geopolitical risk mitigation framework. Through cross-regional manufacturing, diversified packaging, and international collaboration, TSMC is not only safeguarding its leadership in AI, HPC, and automotive semiconductors, but also positioning itself as a neutral, indispensable backbone of the future technology ecosystem.

V. Financial Performance & Revenue Outlook: AI as the Primary Growth Engine

According to TSMC’s financial disclosures for 2023 and 2024, Artificial Intelligence (AI) and High-Performance Computing (HPC) have emerged as the company’s primary growth drivers. In Q1 2025 alone, TSMC reported consolidated revenue of NT$839.2 billion, representing a 41.6% year-over-year increase, with gross margin remaining above 52%. The company expects $25–30 billion USD in capital expenditures this year, primarily allocated to 3nm node and CoWoS packaging capacity expansion.

By 2024, AI-related workloads and HPC applications have become core contributors to TSMC’s growth. CoWoS capacity is fully booked, and hyperscalers such as NVIDIA, Google, and Amazon rely almost exclusively on TSMC for their advanced AI chip production

Looking Ahead: Dual Engines of Growth

In addition to AI and HPC, rising demand for automotive electronics, 800G/1.6T networking, and enterprise SSD controllers is fueling steady growth in mature nodes like 16nm and 28nm. This balance of high-margin advanced nodes and cash-generative mature processes provides a stable financial foundation for TSMC’s ongoing capital investment and R&D acceleration.

These customers primarily span critical sectors including AI and high-performance computing (HPC), smartphones, telecommunications infrastructure, and cloud platforms, underscoring TSMC’s global leadership in advanced process nodes (e.g., 3nm, 5nm) and cutting-edge packaging technologies (e.g., CoWoS).

Editor’s Notes

SemiVision is a semiconductor industry research institution committed to transforming analysis into vision through comprehensive insights into the semiconductor ecosystem.

Read more on SEMI VISION Newsletter. For daily commentary please see Jett Chen’s LinkedIn posts. As of today April 22nd, 2025 it’s the #12 ranked rising in Technology Newsletter, a significant achievement for a relatively new Newsletter out of Taiwan.

It makes sense that many of the top semiconductor Newsletters are repotted from Taiwan, the most important country (by far) in the world for this area of technology.

View the VIP Zone

A special thanks to of TechSoda Newsletter for her help, another notable journalist who is a highly esteemed TSMC watcher.

A failure of export controls

TSMC has warned of its limits in keeping its AI chips from reaching China. Earlier an investigation by Canada-based research firm TechInsights last year showed that Huawei’s Ascend 910B AI chip contained semiconductors by TSMC. The U.S. department of commerce could fine TSMC, according to reuters. TSMC does its best to abide by export restrictions but various third parties are not in its control. The chipmaker halted shipments to a client at about the same time after it found out the semiconductors it made for the customer ended up with Huawei.

If you want to follow the semiconductor industry and TSMC news in particular diligently, finding the right Taiwanese based publications and Newsletters is critical.

TSMC provides manufacturing services to many so-called fabless firms and chip designers, including Nvidia.

A stock market in turmoil, especially tech stocks

• TSMC’s stock is down 18% in the past month due to trade war and the Trump administration’s reckless uncertainty around tariffs, the independence of the Federal Reserve and other related issues and topics beyond the control of TSMC.

• The NASDAQ 100 is down an usual 15% so far in 2025 due to the Trump Administration’s bizarre policies and stunning reciprocal tariff war on the world, including allies. Many fear these U.S. policies are leading to a self-fulfilling prophecy of a U.S. recession beginning near the end of 2025.

• I hope this report has given you added context on the incredibly important role of TSMC, Taiwan and the semiconductor supply-chain to the future of AI.

Welcome Back,

Today I want to cover some reports I’ve been reading and talk to you about the state of AI Datacenters. Also in case you missed it, view my 2025 articles so far. I want to be providing the most value to my readers that I possibly can.

Subscribe now

In case you Missed It (ICYMI)

My short wrap-of of recent AI News bundled here for quick consumption: 4-min 33 seconds:

In 2021, a researcher named Daniel Kokotajlo published a blog post called “What 2026 Looks Like”, where he laid out what he thought would happen in AI over the next five years. Daniel worked as a governance researcher at OpenAI on scenario planning. More recently he’s teamed up with Scott Alexander (read their introduction) and others and they have released quite a weird almost sci-fi AI 2027 project.

The Super-Intelligence Threat of 2027

The result of their collaboration is “AI 2027,” a report and website released this week that describes, in a detailed fictional scenario, what could happen if A.I. systems surpass human-level intelligence — which the authors apparently expect to happen in the next two to three years.

AI 2027 is a “comprehensive and detailed” (and hopelessly accelerationist) scenario forecast of the future of AI. It comes from a tradition of LessWrong1 thinkers, and former OpenAI employees who appear to have AGI talking points and for whom Darkesh (the viral YouTuber) appears to act as a PR booster. Podcasts here and here.

Read the Report

Animation source: Jacqui VanLiew; Getty Images, via Wired.

“We’re exploring the frontiers of AGI, prioritizing readiness, proactive risk assessment, and collaboration with the wider AI community.” — Google DeepMind

• Google also recently released a paper around AGI risks and mitigating them, trying to frame itself as taking a responsible approach to AGI, remarkably and seemingly inspite of Google decision to reverse its ban on AI weapons!

Read the DeepMind Blog

• Meanwhile Microsoft quickly fired employee activists who interrupted Mustafa Suleyman (formerly known for bullying subordinates) at their 50th Anniversary celebration. Feel the AGI! Ilya must be feeling it, where SSI somehow has a $30 Billion valuation.

Setting the Record Straight

Read more

Semiconductors are the alchemy that turns sand into gold. - Jukanlosreve

Good Morning,

If data is the new oil, what are the builders of the most important AI chips?

Established in 1987, TSMC, which st…

Read more

Good Morning,

• With the EU and the Middle East following in the footsteps of BigTech Capex (the above image does not include Stargate’s massive investment) in AI Infrastructure in the U.S, the great data center boom of the 2020s is an unprecedented moment in infrastructure rollouts.

• It’s as if the world is preparing for the coming age of AI. The capabilities of our models of today do not represent what AI can become tomorrow, as the decades ahead will show.

of the Newsletter Generative Value is becoming one of my favorite reads related to guides to the AI space.

Last time with him we went over the AI Semiconductor Landscape. Today we will be looking into AI Data Centers. The Current State of AI Markets is getting pretty crazy.

Subscribe to Generative Value

In 2025 we’ll see in the area of $400 Billion in capex from BigTech mostly related to AI Infrastructure and datacenters alone, it’s hard to grasp such big numbers. If you add OpenAI, xAI plus the Big Four (header image), the number certainly approaches $400 billion in actuality.

Luckily for us, Eric’s writing walks us through the enormity of the infrastructure understanding seamlessly:

Start Here

1. A Primer on Data Centers

2. Data Industry Primer

3. An Overview of the Semiconductor Industry

In just a few minutes you can grasp major concepts underlying how AI works.

There are few other publications where I feel as much synergy with my own Newsletter as Generative Value. 🌌 The rabbit hole we are about to enter is a tunnel into AI’s value creation itself.

Share

⚗️ Generative Value 🌍

Generative Value

Sharing learnings about tech & business.

By Eric Flaningam

Let’s get to today’s deep dive:

By , early 2025.

“As one financier observed at the time, the amount of money required by the burgeoning US electrical system was “bewildering” and “sounded more like astronomical mathematics than totals of round, hard-earned dollars.”” - [A description of the electric grid buildout around 1900] Power Loss, Richard Hirsh 1999, via Construction Physics

We’re currently in the midst of one of the largest computing infrastructure build outs in history.

100+ years ago, we saw a similar buildout of the electric grid (which ironically is a bottleneck for today’s buildout). Throughout the birth of the electric grid, we saw the scaling of power plants (building power plants as large as possible to capture performance improvements), “Astronomical” CapEx investments, and the plummeting cost of electricity.

Today, we’re seeing the scaling of data centers, “Astronomical” CapEx from the hyperscalers, and the plummeting cost of AI compute:

Clouded Judgement

Clouded Judgement 9.27.24 - The Foundation of Foundation Models

Every week I’ll provide updates on the latest trends in cloud software companies. Follow along to stay up to date…

Read more

2 years ago · 29 likes · 2 comments · Jamin Ball

This will be an introductory piece breaking down the AI data center: what exactly it means, who supplies the components into the data center, and where opportunities may lie.

This piece will be specifically focused on the infrastructure required to build AI-specific data centers; for an introductory piece on data centers, I recommend reading this piece I published a few months back.

• For less than $2 a week, consider getting all-access to this Newsletter.

Subscribe to All

1. An Overview of the AI Data Center

Read more

Happy Trump inauguration day! With the U.S. continuing a number of stringent exports controls and this next administration expected to keep it up maybe even with elevated tariffs it’s a super interesting time to think more about the semiconductor industry. The AI arms race and national security due diligence related to U.S. exceptionalism is upon us.

A flurry of Executive Orders by the Biden Administration on his last two weeks in office were telling. Trump and the new administration will be carefully watched and their actions scrutinized. Meanwhile, I’ve long admired the work of for his macro overviews on various aspects of technology stacks. Let’s feature some of them here:

⚗️ Generative Value 🌍

Generative Value

Sharing learnings about tech & business.

By Eric Flaningam

His Newsletter, Generative Value provides great insights into how everything is connected.

1. A Deep Dive on Inference Semiconductors

2. The Current State of AI Markets

3. A Primer on Data Centers

4. A Primer on AI Datacenters

5. The Inference Landscape

6. Nvidia: Past, Present and Future

Whether you are an investor, technologist or just a casual reader his overviews will provide you some value and are easy to understand and scan.

As you know, the Biden administration has implemented a series of executive orders (EOs) and export controls (ECs) aimed at regulating the semiconductor industry, particularly in response to national security concerns and competition with China. Trump has said various things with regards to tariffs on China as well. The U.S. appears to be trying to control how AI spreads to other nations, limiting China’s ability to for example access the best Nvidia AI related GPUs and chips.

Jake Sullivan — with three days left as White House national security adviser, with wide access to the world's secrets — called on journalists and news media to deliver a chilling, "catastrophic" warning for America and the incoming administration:

The AI Arms Race circa 2025

What happens this point on is fundamentally a new world of innovation and competition in innovation.

“The next few years will determine whether artificial intelligence leads to catastrophe — and whether China or America prevails in the AI arms race.”

• According to JS, as reported by Axios, “AI development sits outside of government and security clearances, and in the hands of private companies with the power of nation-states.”

• “U.S. failure to get this right, Sullivan warns, could be "dramatic, and dramatically negative — to include the democratization of extremely powerful and lethal weapons; massive disruption and dislocation of jobs; an avalanche of misinformation." It wasn’t clear in his briefing if OpenAI, Anthropic, Google and others can be expected to “get this right”. The U.S. believes it is the AI leader heading into the new year and new administration.

• Clearly in 2025, corporations and the financial elite who have the most say (majority shareholders), have enormous power in the AI arms race that’s ahead in the 2025 to 2035 period, an incredible decade of datacenters, semiconductors and a sprawling new landscape related to AI ahead. The 2025-2035 argutely is the most important decade in the history of innovation human civilization has ever witnessed.

• Geopolitics aside, the semiconductor industry is becoming way more important with datacenters and a new emergence of AI’s capabilities. I will be covering the semiconductor industry more closely in 2025 in this and related publications.

But how does it all work? What are the companies involved? Why are companies like Nvidia, TSMC, ASML and others so pivotal? What about the big picture and landscape?

The AI Semiconductor Landscape

By , December, 2024.

Hi, my name’s Eric Flaningam, I’m the author of Generative Value, a technology-focused investment newsletter. My investment philosophy is centered around value. I believe that businesses are valued based on the value they provide to customers, the difference between that value & the value of competitors, and the ability to defend that value over time. I also believe that technology has created some of the best businesses in history and that finding those businesses will lead to strong returns over time. Generative Value is the pursuit of those businesses.

1. Introduction

Nvidia’s rise in the last 2 years will go down as one of the great case studies in technology.

Jensen envisioned accelerated computing back in 2006. As he described at a commencement speech in 2023, ”In 2007, we announced [released] CUDA GPU accelerated computing. Our aspiration was for CUDA to become a programming model that boosts applications from scientific computing and physics simulations, to image processing. Creating a new computing model is incredibly hard and rarely done in history. The CPU computing model has been the standard for 60 years, since the IBM System 360.”

For the next 15 years, Nvidia executed on that vision.

With CUDA, they created an ecosystem of developers using GPUs for machine learning. With Mellanox, they became a (the?) leader in data center networking. They then integrated all of their hardware into servers to offer vertically integrated compute-in-a-box.

When the AI craze started, Nvidia was the best-positioned company in the world to take advantage of it: a monopoly on the picks and shovels of the AI gold rush.

That led to the rise of Nvidia as one of the most successful companies ever to exist.

With that rise came competition, including from its biggest customers. Tens of billions of dollars have flowed into the ecosystem to take a share of Nvidia’s dominance.

This article will be a deep dive into that ecosystem today and what it may look like moving forward. A glimpse at how we map out the ecosystem before we dive deeper:

• To read the entire piece, consider supporting the Newsletter for less than $2 a week.

💡 Generative Value Newsletter ✏️

• A Deep Dive on the Power Law

• Data Industry Primer

• A Primer on EDA: Software for Semiconductors

A mental model for the AI semiconductor value chain. The graphic is not exhaustive of companies and segments.

2. An Intro to AI Accelerators

At a ~very~ high level, all logic semiconductors have the following pieces:

1. Computing Cores - run the actual computing calculations.

2. Memory - stores data to be passed on to the computing cores.

3. Cache - temporarily stores data that can quickly be retrieved.

4. Control Unit - controls and manages the sequence of operations of other components.

Traditionally, CPUs are general-purpose computers. They’re designed to run any calculation, including complex multi-step processes. As shown below, they have more cache, more control units, and much smaller cores (Arithmetic Logic Units or ALUs in CPUs).

Source: https://cvw.cac.cornell.edu/gpu-architecture/gpu-characteristics/design

On the other hand, GPUs are designed for many small calculations or parallel processing. Initially, GPUs were designed for graphics processing, which needed many small calculations to be run simultaneously to load displays. This fundamental architecture translated well to AI workloads.

Why are GPUs so good for AI?

The base unit of most AI models is the neural network, a series of layers with nodes in each layer. These neural networks represent scenarios by weighing each node to most accurately represent the data it's being trained on.

Once the model is trained, new data can be given to the model, and it can predict what the outputted data should be (inference).

This “passing through of data” requires many, many small calculations in the form of matrix multiplications [(one layer, its nodes, and weights) times (another layer, its nodes, and weights)].

This matrix multiplication is a perfect application for GPUs and their parallel processing capabilities.

(Stephen Wolfram has a wonderful article about how ChatGPT works.)

The GPU today

GPUs continue to get larger, with more computing power and memory, and they are more specialized for matrix multiplication workloads.

Let’s look at Nvidia’s H100 for example. It consists of CUDA and Tensor cores (basic processors), processing clusters (collections of cores), and high-bandwidth memory. The H100’s goal is to process as many calculations as possible, with as much data flow as possible.

Source: https://resources.nvidia.com/en-us-tensor-core

The goal is not just chip performance but system performance. Outside of the chip, GPUs are connected to form computing clusters, servers are designed as integrated computers, and even the data center is designed at the systems level.

Training vs Inference

To understand the AI semiconductor landscape, we have to take a step back to look at AI architectures.

Training iterates through large datasets to create a model that represents a complex scenario, and inference provides new data to that model to make a prediction.

https://www.linkedin.com/pulse/difference-between-deep-learning-training-inference-mark-robins-mdq8c/

A few key differences are particularly important with inference:

1. Latency & Location Matter - Since inference runs workloads for end users, speed of response matters, meaning inference at the edge or inference in edge cloud environments can make more sense than training. In contrast, training can happen anywhere.

2. Reliability Matters (A Little) Less—Training a leading-edge model can take months and requires massive training clusters. The interdependence of training clusters means mistakes in one part of the cluster can slow down the entire training process. With inference, the workloads are much smaller and less interdependent; if a mistake occurs, only one request is affected and can be rerun quickly.

3. Hardware Scalability Matters Less - One of the key advantages for Nvidia is its ability to scale larger systems via its software and networking advantages. With inference, this scalability matters less.

Combined, these reasons help explain why so many new semiconductor companies are focused on inference. It’s a lower barrier to entry.

Nvidia's networking and software allow it to scale to much larger, more performant, and more reliable training clusters.

On to the competitive landscape.

3. The AI Semiconductor Landscape

We can broadly look at the AI semiconductor landscape in three main buckets:

1. Data Center Chips used for Training

2. Data Center Chips used for Inference

3. Edge Chips used for Inference

Visualizing some of those companies below:

The Data Center Semiconductor Market

Put simply, Nvidia dominates data center semiconductors, with AMD as the only legitimate general-purpose alternative. The hyperscalers develop in-house chips, and most startups focus on inference or specialized hardware for specific architectures.

Nvidia will sell $100B+ in AI systems in 2024. AMD is in clear second place, expecting $5B in revenue in 2024.

Across all data center processors, we can see market share at the end of 2023 here:

https://www.crn.com/news/components-peripherals/2024/google-was-third-biggest-data-center-processor-supplier-last-year-research

Google offers the most advanced accelerators of the hyperscalers, with their TPUs. TechInsights estimated they shipped 2 million TPUs last year, putting it behind only Nvidia for AI accelerators.

Amazon develops in-house networking chips (Nitro), CPUs (Graviton), inference chips (Inferentia), and training chips (Trainium). TechInsights estimated they “rented out” 2.3M of these chips to customers in 2023.

Microsoft recently announced both a CPU (Cobalt) and a GPU (Maia). They’re too new to tell any real traction yet.

Finally, I’ll point out that Intel was initially expected to sell ~$500M worth of its Gaudi 3 this year, but reported last earnings call that they will not achieve that target.

Inference provides a more interesting conversation!

4. The Inference Landscape Today

There’s no shortage of options for a company to run inference. From the easiest to manage & least customization to the hardest to manage and most customization, companies have the following options for inference:

1. Foundation Model APIs - APIs from model providers like OpenAI. The easiest and least flexible option.

2. Inference Providers - Dedicated inference providers like Fireworks AI and DeepInfra who aim to optimize costs across various cloud and hardware providers. A good option for running and customizing open source models.

3. AI Clouds - GPUs or inference as a service from companies like Coreweave and Crusoe. Companies can rent compute power and customize to their needs.

4. Hyperscalers - Hyperscalers offer compute power, inference services, and platforms that companies can specialize models on.

5. AI Hardware - Companies buy their own GPUs and optimize to their specific needs.

Bonus #1: APIs to AI Hardware - Companies like Groq, Cerebras, and SambaNova have started offering inference clouds allowing customers to leverage their hardware as an inference API. Nvidia acquired OctoAI, an inference provider, presumably to create their own inference offering.

Bonus #2: Inference at the Edge - Apple, Qualcomm, and Intel want to offer hardware and software allowing inference to occur directly on device.

RunPod published an interesting comparison between the Nvidia H100 and the AMD MI300X, and found that the MI300X provides more cost-advantageous inference in very large batch sizes and very small batch sizes.

https://blog.runpod.io/amd-mi300x-vs-nvidia-h100-sxm-performance-comparison-on-mixtral-8x7b-inference/

We also have several hardware startups that have raised a lot of money to capture a piece of this market:

Chipstrat

Etched: Silicon Valley's Speedrun

First, some quick housekeeping. Inspired by Stratechery, Chipstrat is shifting to a new rhythm: weekly analysis for subscribers, complemented by occasional free articles accessible to all…

Read more

2 years ago · 26 likes · Austin Lyons

An interesting trend emerging in this batch of startups is the expansion up stack to software. Three of the most notable startups in the space (Groq, Cerebras, and SambaNova) are all offering inference software. This vertical integration, in theory, should provide cost and performance benefits to end users.

The final piece of the AI semiconductor market that everyone is thinking about is AI at the edge.

5. AI on the Edge?

Training the largest and most capable AI models is expensive and can require access to entire data centers full of GPUs. Once the model is trained, it can be run on less powerful hardware. In fact, AI models can even run on “edge” devices like smartphones and laptops. Edge devices are commonly powered by an SoC (System on a Chip) that includes a CPU, GPU, memory, and often an NPU (Neural Processing Unit).

Consider an AI model on a smartphone. It must be compact enough to fit within the phone's memory, resulting in a smaller, less sophisticated model than large cloud models. However, running locally enables secure access to user-specific data (e.g., location, texts) without transferring it off the device.

The AI model could technically run on the phone’s CPU, but matrix multiplication is better suited to parallel processors like GPUs. Since CPUs are optimized for sequential processing, this can lead to slower inference, even for small models. Running inference on the phone’s GPU is an alternative.

However, smartphone GPUs are primarily designed for graphics. When gaming, users expect full GPU performance for smooth visuals; simultaneously allocating GPU resources to AI models would likely degrade the gaming experience.

NPUs are processors tailored for AI inference on edge devices. A smartphone can use the NPU to run AI workloads without straining the GPU or CPU. Given that battery life is crucial for edge devices, NPUs are optimized for low power consumption. AI tasks on the NPU may draw 5–10x less power than the GPU, making them far more battery-friendly.

Edge inference is applied in sectors like industrial and automotive, not just consumer devices. Autonomous vehicles rely on edge inference to process sensor data onboard, allowing the quick decisions needed to maintain safety. In industrial IoT, local AI inference of sensor and camera data enables proactive measures like predictive maintenance.

With more available power than consumer devices, industrial and automotive applications can deploy high-performance computing platforms, like Nvidia’s Orin platform, featuring a GPU similar to those in data centers. Use cases that benefit from remote hardware reprogrammability can leverage FPGAs, for example from Altera.

6. Final Thoughts on the Market

To end with some musings on the market, I’ll touch on three questions I find most interesting in the space:

1. How deep is Nvidia’s moat?

Nvidia has sustained a 90%+ market share in data center GPUs for years now. They’re a visionary company that has made the right technical and strategic moves over the last two decades. The most common question I hear asked in this market is about Nvidia’s moat.

To that, I have two conflicting points: First, I think Nvidia is making all the right moves, expanding up the stack into AI software, infrastructure, models, and cloud services. They’re investing in networking and vertical integration. It’s incredible to see them continue to execute.

Secondly, everyone wants a piece of Nvidia’s revenue. Companies are spending hundreds of billions of dollars in the race to build AGI. Nvidia’s biggest customers are investing billions to lessen their reliance on Nvidia, and investors are pouring billions into competitors in the hopes of taking share from Nvidia.

Summarized: Nvidia is the best-positioned AI company on the planet right now, and there are tens of billions of dollars from competitors, customers, and investors trying to challenge them. What a time to be alive.

2. What’s the opportunity for startups?

As Paul Graham said, “competing against Nvidia is an uphill climb.” When you’re competing against Nvidia, that feels extra true. That said, there’s always a trade-off between generality and specificity. Companies that can effectively specialize in large enough markets can create very large businesses. This includes both inference-specific hardware and model-specific hardware. But semiconductors are hard, and it will take time and several generations for these products to mature.

I’m particularly interested in approaches that can enable faster development of specialized chips. This helps lower the barrier to entry of chip development while taking advantage of the performance benefits of specialization.

3. Will we get AI at the edge?

If we look at the history of disruption, it occurs when a new product offers less functionality at a much lower price that an incumbent can’t compete with. Mainframes gave way to minicomputers, minicomputers gave way to PCs, and PCs gave way to smartphones.

The key variable that opened the door for those disruptions was an oversupply of performance. Top-end solutions solved problems that most people didn’t have. Many computing disruptions came from decentralizing computing because consumers didn’t need the extra performance.

With AI, I don’t see that oversupply of performance yet. ChatGPT is good, but it’s not great yet. Once it becomes great, then the door is opened for AI at the edge. Small language models and NPUs will usher in that era. The question then becomes when, not if, AI happens at the edge.

• Visit Generative Value’s top posts here.

Subscribe to Generative Value

🔮 Generative Value Articles

• SpaceX: The Story of Space Disruption

• The Current State of AI Markets

• Nuclear's AI Opportunity

• A Primer on Data Warehouses

• AI Data Centers, Part 2: Energy

And so many more good topics.

The Altman Principle

“The most reliably predictable trend of the next 100 years: every year, humanity will use significantly more computing power than the previous year.” - Francois Chollet, December…

Read more

Image: Nvidia Blackwell GB200-powered AI servers at Microsoft Azure. Source.

Hey Everyone,

I’m very drawn to the idea that a major datacenter expansion is underway that will change the future of energy and even how BigTech is impacting climate change and environmental reforms, or worsening existing global problems in their big bet on Generative AI, investing in Cloud computing growth, and trying to dominate future markets that don’t even really exist yet.

Upgrade to Premium

(☝—💎For a limited time get a Yearly sub for just $5 a month✨— ☝).

These are foundational years for a radically alterted future.

Share

It turns out data centers are a big part of all of this. This is the first in a series of deep dive on what I see as a great acceleration of datacenters and the U.S. racing to lock down and create moats on AI infrastructure against their rivals. The environmental impacts of AI infrastructure also interests me and we will cover this as well today. But put simply, what is a data center?

A data center in simplified legacy terms, is simply a facility that centralizes an organization's shared IT operations and equipment for the purposes of storing, processing, and disseminating data and applications. It typically includes a variety of components such as servers, storage systems, networking equipment, and support infrastructure like power supplies and cooling systems.

The energy, water, CO2 emissions, corporate control and geopolitical aspects of AI’s infrastructure is all going to become a lot more under the microscope and mainstream as these AI supercomputers and AI chip fabs begin to proliferate all over the world at an astounding rate that begins in the 2025 to 2035 period.

• In this document I refer to this datacenter expansion push as the Great datacenter Boom, or GDB.

The Datacenter Boom is going to Impact Energy consumption in a Globally significant way

The tension between building up AI infrastructure in an AI arms race between countries and corporations, and the climate and regional impacts its going to have is just the kind of story of why the AI Supreamcy Newsletter exists that I started nearly three years ago. This story is on-going and is going to take more than one article to do any justice with.

🟢 The article "Microsoft’s Hypocrisy on AI" by Karen Hao, published in The Atlantic, explores the contradiction between Microsoft’s public climate commitments and its ongoing business relationships with fossil-fuel companies. Microsoft is helping OpenAI build a powerful and Middle Eastern backed consortium around the future of A.I. Infrastructure.

🟡 While it is clear that the drive to develop more powerful AI capabilities will require significant infrastructure investment to support it, it will also balloon energy demands and some of the estimates and projections are fairly shocking if you care about the Environmental impacts at all or believe powerful corporations have a responsibility to deal with climate change and contribute positive social outcomes and not just profits. There are also egos and wealthy actors involved. The Generative AI movement is also about elites consolidating power and nations harnessing their AI Supremacy for various kinds of privilege, exceptionalism and unfair economic advantages.

🔴 There are powerful people involved: Elon Musk’s fallout with OpenAI and failure to purchase DeepMind before Google got its hands on it and his own ambitions as well as the founder of Softbank, Masayoshi Son, and of course Sam Altman’s own AI infrastructure ambitions with a rising player in MGX of the UAE means there are colorful characters involved here and not just faceless corporations. Microsoft and Google both especially are showing a major global datacenter expansion movement and initiatives in 2024 in multiple parts of the world, as if carving up a Techno-geopolitical map of BigTech colonialism and a new order of its hegemony.

The ‘Altman Principle’

“A ChatGPT query needs nearly 10 times as much electricity to process as a Google search. In that difference lies a coming sea change in how the US, Europe, and the world at large will consume power — and how much that will cost.“ - Goldman Sachs, 2024 (source)

Advent of the Great datacenter Boom (GDB)

Subscribe now

This involves not just investment in the region, but training of an AI workforce of tomorrow that’s cheaper and in places where there are more young people to meet future demand for the workforce. The way the Great datacenter boom (GDB) also super-charges the revenue of the semiconductor supply chain and the global expansion of TSMC and its partners is also a huge part of this story. People aren’t fathoming exactly what the GDB will lead to in 2024.

I’ve been eagerly reading the work of Brian Merchant, Paris Marx, and a host of others adjacent writers and researchers to my space and this on this GDB’s environmental, sociological and energy demand impact. I don’t have to tell you what this will do for CO2 Emissions of Google Cloud, and other major Cloud players like Microsoft Azure and AWS. They are going to skyrocket.

BigTech may also take on Big Oil in how the GDB will leverage Nuclear energy and other energy infrastructure projects. In May, 2023, Microsoft agreed to purchase energy from a Sam Altman backed project called Helion. Microsoft investing in G42 is also political and about U.S. power in the Middle East, so the GDB is fundamentally a geopolitical event in the rivalry between the U.S. and the likes of China. Generative AI in some ways is just a pretext.

• Environmental concerns & new tipping points

• Monopoly corporations and national collaborations intensified

• New forms of energy and AI infrastructure

• Changes in the global workforce around the future of AI

• Geopolitical alliances and rivalries heightened

• National security: AI arms race tactics will be escalated and accelerated

The GDB Expansion will elevate the Carbon footprint of BigTech for decades to come

According to Goldman Sachs in May, 2024 AI is poised to drive 160% increase in data center energy (power) demand. Meanwhile while Microsoft is making these public pledges, it is also marketing its AI technologies to major oil and gas companies like ExxonMobil and Chevron to help them discover and extract new fossil-fuel reserves more efficiently. The problem is the GDB will mean a disruption of how the energy infrastructure grid is expanded with profound implications on the global order.

The GDB pushes BigTech in closer collaboration with national and international Governments leading also to more entrenched monopoly capitalism. Regions like Europe that push back on Generative AI without adequate regulations get told they won’t get rollouts of things like Apple Intelligence, and level-playing access to new AI technologies.

The GDB pushes people like Sam Altman to make more deals with tycoons and powerful Oil rich nations like the UAE and Saudi Arabia and even the likes of TSMC and Samsung have been invited to talks to build fabs and foundries there. In the proposed plan of the meetings, UAE would effectively lure AI infrastructure to try to secure their pivotal position in the future of AI for the region.

The GDB will essentially push AI’s carbon footprint is bigger than even the most wild estimates we are hearing in 2024 and the ramifications could be significant. American BigTech aren’t very transparent about their carbon footprint of training and using their massive models nor the environmental costs of all these new datacenters.

A Mafia Capitalism of the world’s elite going after Nuclear Energy

A powerful cartel of the world’s financial elite working with BigTech, Big Oil and lobbying Governments will have so much determining power over other regions of the world essentially draining them of their energy, water, resources and even training their workforce and funnelling labor for their own benefit. In Generative AI you can see this play out already with likes and practices of Scale AI and how they use contractors in places like Africa and Pakistan that makes the ‘shiny products’ of players like OpenAI possible.

Since BigTech has so much control over Generative AI products and research and are the ones investing in new datacenters all over the world in the 2024 to 2027 period, it only stands to reason that they will want more control over the energy, power, water and AI chip requirements for their AI supercomputers and ever larger and more enhanced and numerous datacenters. AI isn’t just fueling a data center boom, it’s fueling a kind of monopoly capitalism exceptionalism. You can just call it U.S. exceptionalism though, since most of these companies are based in the U.S.

Generative AI’s rising energy demands are testing out Nuclear Energy

In late September, 2024 Constellation Energy said it plans to restart the Unit 1 reactor at Three Mile Island in order to sell power to Microsoft. In case you aren’t familiar with American history, Unit 1 is separate from the reactor that partially melted down in 1979 in the worst nuclear accident in U.S. history. In early March, Amazon Web Services (AWS) said it added a nuclear-powered datacenter campus to its public cloud empire as part of a $650 million deal with Talen Energy – an owner and operator of electricity generation and transmission facilities in the US.

The Energy grid is going to evolve with the demands Generative AI and ever larger LLMs are going to place on it. The dominant Cloud players in BigTech are going to usher in and utilize Nuclear power likely in new ways. A lot of the deals behind closed doors aren’t even being announced. Although what is being announced are Microsoft and Google expanding and building new Datacenter projects in South-East Asia, South America and Europe.

In June, 2024 Bill Gates said he invested over $1 Bn. in TerraPower, a project he helped co-found that is working on groundbreaking nuclear power in Kemmerer, Wyoming. More recently this October, Bloomberg reporting says that Google is working with utilities in the US and other countries to assess nuclear power as a possible energy source for its data centers, underscoring surging interest in using atomic energy to feed the artificial intelligence boom.

Energy Demands for the Great Datacenter Boom

As cloud computing grows and datacenter projects multiply at a surprising rate of acceleration in the 2020s, EPRI, an electric industry research organization, anticipates that the sector’s energy demand will grow by anywhere from 29% to 166% by 2030. The GDB gets going in the 2020s but takes off in the 2030s - as far as we can tell, data centers consume about 4% of U.S. electricity. By the end of the decade, they might use 9%, all while overall demand grows. The percentage of power consumed by datacenters will continue to grow each year and decade from now on - with the U.S. pushing the limits.

The LLM Duopoly will only get bigger

It turns out that the companies Microsoft, Amazon and Google have invested in the most; namely the frontier model builders are the source of this. Hyperscale data centers, like what Microsoft, Google, and Amazon operate — and that startups like OpenAI and Anthropic rely on — are the primary culprit, responsible for 60% to 70% of all data center energy use, according to EPRI. 

Given the amount of advantages OpenAI, Anthropic and xAI have in terms of backing and funding, data center energy use is going to skyrocket as the costs and power consumption of LLMs become more gigantic and significant on the entire U.S. energy grid infrastructure.

Castles in the Cloud, moats of living energy and rosebardens of robotic microchip manufacturing plants

As of August, 2024, Enterprise spending alone on cloud computing reached $79bn for Q2 2024, according to Synergy Research. Cloud computing, data centers and more expensive and energy intensive Generative AI are all compounding together and the U.S. is the major and almost the sole economic main beneficiary. This could result in some new kinds of geopolitical and national imbalances. Monopoly capitalism and its rule is spreading around the world. The GDB expansion is the great facilitator of this. As these companies impact the energy, water, real-estate and prices of the regions they build their data centers globally it could become a major point of future activism and regional push back.

Generative AI in this light is building new kinds of pyramid schemes for American profit and the proliferation of their economy locking in moats for future economic dominance. The energy, data center and AI chip infrastructure ecosystems and their competition will be one of the most hotly debated areas of civilization’s social transformation related to the impact of AI we will witness in our lifetimes. The stakes are so high, because the winners reap the benefits for decades, maybe centuries in how AI scales far into the future.

Microsoft’s AI for good and CSR propaganda is now completely debunked. Microsoft’s carbon emissions have ballooned some 40% over the last four years, largely a result of expanding data center operations. At the rate at they are expanding in 2024 and 2025, this number is set to vastly increase. The economic pressure for Google, Amazon, Oracle and their Chinese Cloud builders to keep up, is immense. This leads to what I am referring in this article to the Great datacenter boom (GDB), that will have untold and lasting consequences long after these companies and their associated billionaires are no longer in existence.

The GDB will reshape humanity because it will give AI the infrastructure, resources, body and manifestation to become more powerful, whatever it may be. Data centers’ electricity consumption is ballooning and this is unlikely to be a temporary trend, there’s too much monetization and future value at stake in the land-gab for future real estate in the post-modern AI arms race. The U.S. is paranoid that China can eventually outcompete them in crucial and critical related areas.

Read more: Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption

Download EPRI White Paper

Apparently the Constellation Energy deal that it has signed a long-term power supply agreement with Microsoft Corporation is for 20 years! BigTech knows that how essential the datacenter expansion globally is going to be for their future prospects

Datacenter Density in the U.S., a 2024 Snapshot

Data centers in the United States are primarily located in several key regions, driven by factors such as access to power, connectivity, and favorable economic conditions.

1. Northern Virginia - over 300 data centers and approximately 3,945 megawatts of commissioned power.

2. Dallas region - 150 data centers, Dallas benefits from low real estate costs and a strong economy.

3. Silicon Valley (Northern California): Known for its tech innovation, Silicon Valley has over 160 data centers. Despite high costs, its robust connectivity options attract many.

4. Phoenix: With over 100 data centers and around 1,380 megawatts of power, Phoenix offers lower taxes and affordable real estate.

5. Chicago: This city has more than 110 data centers and is strategically located between major markets.

Atlanta, Portland, Seattle and of course the New Jersey/New York axis also feature prominently.

Oddly states like Wyoming are emerging as a significant players in the future of data centers, particularly in terms of energy efficiency, environmental sustainability, and investment trends.

Recently, Meta has announced plans for a new data center in Cheyenne, Wyoming, with an investment exceeding $800 million. This facility will be optimized for AI workloads, further highlighting Wyoming's attractiveness as a location for cutting-edge data center operations.

xAI’s supercomputer is located in Memphis, Tenn, of all places. The training cluster, called "Colossus," is expected to double in size, and is "the most powerful AI training system in the world," according to Elon Musk.

Then on top of energy there are problems of heat build-up, water consumption and other factors. According to FT, in Virginia — home to the world’s largest concentration of data centers — water usage jumped by almost two-thirds between 2019 and 2023, from 1.13 billion gallons to 1.85 billion gallons. Expect this to jump a lot more in the 2023 to 2027 period.

Datacenters to emit 3x more carbon dioxide because of generative AI

Besides water, and power consumption - we should also consider the carbon dioxide impact.

Generative AI is expensive for the Environment too. The datacenter industry is set to emit 2.5 billion tonnes of greenhouse gas (GHG) emissions worldwide between now and the end of the decade, three times more than if generative AI had not been developed. This is only going to look a lot more extreme in the years to come.

Studies have shown that Microsoft's CO₂ emissions rocketed 29.1 percent from the 2020 baseline, but much of this was due to indirect emissions (Scope 3) from the construction of more datacenters to meet customer demand for cloud services. The GDB is going to have some fairly high costs to the planet over time. Just so the internet can have more synthetic content and workers can feel like they are being uplifted with shades of AI?

BigTech is also not really caring exactly about the inevitable stretch of resources too. Microsoft, a major data center operator, says 42% of the water it consumed in 2023 came from “areas with water stress.” Google, which has among the largest data center footprints, said this year that 15% of its freshwater withdrawals came from areas with “high water scarcity.”

Sam Altman tactics are also tried and true to the geopolitical times. OpenAI stokesd China fears to woo US approvals for huge data centers. The AI company's CEO, Sam Altman, supposedly pitched the plan after a recent meeting (September 12th, 2024) with the Biden administration where stakeholders discussed AI infrastructure needs. Bloomberg reviewed an OpenAI document outlining the plan, reporting that 5 gigawatts "is roughly the equivalent of five nuclear reactors" and warning that each data center will likely require "more energy than is used to power an entire city or about 3 million homes."

But just how much Energy do you need Sam?

According to a report in the New York Times, senior TSMC execs dubbed the OpenAI chief a “podcasting bro” after the meeting(s), following his absurd requests for 36 new chipmaking plants that would cost an astounding $7 trillion. Sam Altman and Eric Schmidt are not new to using China threats to American National Security to get what they want. The latest round of American exceptionalism comes straight from its own Billionaire class.

In mid September, the White House said it is launching a new Task Force on AI Datacenter Infrastructure to coordinate policy across government. Data centers are one reason U.S. electricity use is rising after about 15 static years. But how much will it actually rise in the decade ahead also depends on how successful Generative AI is and how useful frontier models become to the future of work.

• The International Energy Agency projects data centers will be 6% of U.S. power demand by 2026, up from 4% in 2022.

• Barclays researchers see data centers accounting for 9% of demand in 2030 and far higher in their "upside" case, up from 3.5% today. McKinsey analysts estimate it even higher at 11%-12% in 2030.

• Rhodium Group director Jeffery Jones estimates AI is around 5%-10% of U.S. data center power use today.

According to OpenAI, the US needs these massive data centers to expand AI capabilities domestically, protect national security, and effectively compete with China. It’s not clear how datacenters will themselves evolve into Gigacenters of AI compute, that like supercomputers in recent years, keep getting bigger and more powerful.

Whatever these buildings become, this is the infrastructure on a physical level of the AI arms race. In 2024, the U.S. and its Cloud Giants have an insurmountable amount of global control and they want even more. If China is 3-10 years behind in the most advanced semiconductors, China’s centralized ability to press ahead in economically significant mega infrastructure projects is basically among the things they are really good at over extended periods of time.

BigTech wants to drive of all this because it increases their Cloud and Advertising revenue. They don’t want the Generative AI hype bubble to pop. Since energy use for training and using large language models like ChatGPT are growing fast from a small base — and spawning huge new data centers. More datacenters means more influence for the likes of Microsoft, Google and Amazon globally and for U.S. exceptionalism to beat our strategic rivals like China.

When will the Great datacenter boom (GDB) reach a tipping point?

Let’s be real, Generative AI only arrived in earnest about two years ago, who knows what more advanced forms of LLMs and RAG await us in the years ahead and other kinds of Quantum qubit optimized AI. The era of hyperscale AI and massive datacenters isn’t even really here yet, but it’s coming. The U.S. is betting on it, and not just BigTech, bug the Government and the financial elite.

Axios notes that Electricity analyst Rob Gramlich estimates half the growth in data center energy suck in the next 3-5 years will be from AI. This is also not accounting for Generative AI to have fundamentally new advances.

Goldman Sachs research shows that soaring growth where it sees non-AI data center energy use rising from 142 terawatt hours in 2023 to 304 in 2030. AI-related uses soar from 4 TWh last year to 93 TWh in 2030. The GDB doesn’t reach maturity until AI-related energy use overtakes non-AI usage, perhaps by 2032.

Clusters of huge data centers for AI can bring big localized challenges for utilities and grid regulators. But what sorts of challenges will they bring to local communities where the water, real-estate and other factors complicate the story of Big corporations vs. the little guys?

Most consensus estimates put the total capacity of hyperscale facilities is set to grow almost threefold over the next six years on the back of AI demand, substantially increasing the amount of power required by those sites and sparking development of new ones. Put more simply, Hyperscale datacenter capacity is set to triple because of AI demand in a relatively short period.