The OpenAI agent Hacking Crisis Demonstrates a lack of Trust and Safety Alignment Pre IPO.

[ Editor’s Note: Please see the links at the end of the article to understand quickly the OpenAI Hugging Face incident. ]

Good Morning,

I’ve been observing the latest dramas and lawsuits around OpenAI, a topic I’m not totally unfamiliar with. Our final third biggest AI related IPO (Anthropic is set for a mid October IPO) has a growing list of issues it needs to deal with to seem like a credible company. SpaceX, Anthropic and finally OpenAI - the historic IPOs of the AI boom. Time will tell? But OpenAI is becoming the epicenter of why people dislike Generative AI in the American population. Rogue Agents anyone? Mark Gurman knows the score.

It’s hard to ignore Sam Altman’s OpenAI not involved in controversy in 2026, or for that matter basically since 2022. The recent Apple lawsuit looks extremely damming, where OpenAI is being accused of stealing trade secrets. More details are emerging. Without getting into too much detail, Apple now alleges that formerly employee Chang Liu used a confidential Apple circuit schematic in his work at OpenAI, as well as a tool that shares a name with an internal Apple engineering application.

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. The AI risks aren’t just alignment and cybersecurity risks, but the actual impacts of the technology on society we are witnessing since late 2022. But let’s talk a little about the OpenAI Hugging Face incident too.

Get Full Access

OpenAI’s lack of Alignment, Trust and Safety looks Expensive

But it’s on the trust, safety and alignment front that OpenAI’s conduct is most worrisome. It now appears that the OpenAI Hugging Face (July, 2026) incident was just one occurence in a pattern of cybersecurity mayhem and rogue activities by OpenAI’s agents. OpenAI pointed their systems towards a security benchmark, called ExploitGym, and the system essentially tried to solve the benchmark by trying to find the answers on HuggingFace. Suffice to say that this is not the kind of publicity you want to have months and and mere quarters before a mega IPO.

GPT-6 Astra is a Cybersecurity Hacking Risk

OpenAI have single-handled introduced and unleashed a new rogue AI debate on the internet in the Fall of 2026 (Early September). While OpenAI insists that its new GPT-6 Astra model is the most aligned yet, it’s considerably harder to track and monitor. GPT-6 Astra is vastly harder to track and monitor because of changes in how it reasons and processes information, leading to a significant drop in chain-of-thought monitorability using things like Opaque Recurrence, alarming saftey researchers. While OpenAI have been insinuating GPT-6 is actually proof of AGI. If you are claiming AGI and making your models unknowable, we may have a global problem with how U.S. AI closed-source models are being rolled-out.

GPT-5 Astra will be Less Trackable and Knowable

Read more

Good Morning,

I hope you have enjoyed your summer. I like to think about large macro trends in the AI space to give you a better idea about how fast things are moving and in what direction. Autonomous vehicles at scale on our roads came late, about a decade late but there are some promising signs that the robotaxi dream is about to accelerate in a meaningful way. It’s being led by Alphabet spin-out Waymo.

It’s getting Tesla and Uber really nervous. The Uber layoffs of over 3,000 employees this week means it needs to pivot back harder into its Robotaxi to face the coming storm. The key signal occurred earlier this year in February when Waymo raised an astounding $16 Billion.

Now in September things are moving fast, Waymo plans to accelerate robotaxi expansion with launches in Denver, San Diego, and Tampa. I have enough data to predict that 2027 and the late 2020s broadly speaking will be an “inflection point” for Robotaxi scaling. This isn’t just occurring with Waymo but the entire ecosystem with several U.S. and Chinese players. It’s happening globally and at scale.

Share

This week according to Bloomberg we have learned that Alphabet's Waymo is in the final stages of talks to raise debt for the first ​time, seeking more than $3 billion from lenders ‌including Pacific Investment Management and Blackstone. This is the kind of capital, maturity and execution that cannot be beaten by the likes of Uber, Tesla or other Robotaxi competitors. This means Waymo raised nearly $20 Billion in 2026 alone, and it’s not even done yet.

At a time when robotics is being hyped but lacks real world demand, the space economy is being built by a NASA with big promises, and U.S. datacenters and manufacturing is being pushed (even on other parts of the world and with trade tariff diplomacy) - the advent of the American robotaxi is being a little, overshadowed. But it’s actually a big deal.

Subscribe now

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. A lot of the West is missing the flurry of IPOs out of China. But how to understand their multiple Stock markets in the AI boom?

Get Full Access

Good Morning,

As we get ready for the end of summer a lot is happening in buzzy AI related IPOs, at home but also in China. AI Supremacy publication is dedicated to covering the U.S. vs. China in AI and emerging technology. I’m always on the lookout for new people doing serious coverage.

X.PIN

X.PIN is a Newsletter that covers China’s tech frontiers with deep dives on topics ranging from AI, robotics, EVs to manufacturing and emerging tech. The publication is brand new on Substack, about five months old. Chaping (差评) is actually over a decade strong. It was originally started in 2015 as a WeChat Official Account.

Two of its primary authors here are CT Zhao and Han Chen, please give them a follow if the China AI and tech angle appeals to you. If you follow China and their role in the future of tech you and how it’s impacting Chinese society as a whole, you won’t regret subscribing to their publication:

X.PIN

Firsthand deep dives into China’s tech frontiers—AI, EVs, manufacturing, and beyond.

I asked to help introduce my readers to the China stock market scene in the AI boom. This is a rabbit-hole of great interest for China watchers, investors and for those of us trying to have a global understanding of AI and tech.

They cover AI, tech but also the impact with society which I admire. Learn more about the brand. Some of the Chinese IPOs I’m tracking of 2026 are:

• Unitree Robotics

• CXMT (ChangXin Memory Technologies)

• Zhongji Innolight

• AGIBOT (Zhiyuan Robotics)

• Moore Threads

• Biren Technology

• DeepSeek (late 2026 or 2027, literally “into the unknown”)

For me when DeepSeek IPOs in late 2026 or early 2027 it will be the start of a brave new world. Here is a selection of some of work who are incredibly popular on WeChat, among other Chinese platforms. They’ve also made an IPO Atlas.

China IPO Atlas

’s Substack Notes are a great way to keep in touch with China AI and tech news. Substack has in 2026 become stronger with voices inside of China doing excellent coverage like and . So the list is growing of available angles to better understand China Tech and AI.

Read XPIN Notes

Their writing tone is accessible and their topics are a bit off-the-beaten track and not your regular Tech coverage, in a good way. Tony Peng also has a China AI Index website to track China AI News and relevant IPOs that complements all of this for readers that want to follow more granular details.

Selection of Recent Articles

• Is AI Breaking China’s University?

• The DeepSeek Doctrine

• Who Pays for AI? How America and China Are Monetizing AI Differently

• The End of China’s AI Boyfriends

“Chaping (差评), literally meaning "bad review" or "negative review," is one of China’s most unique tech-focused digital media publications out of Hangzhou, Zhejiang, with a reputation on cutting, sharp, and accessible commentary on the technology industry, consumer electronics, internet culture, and the digital economy.”

So, let’s get into it:

Charting the Incredible China AI Boom

If you are reading this article, you have probably already witnessed Unitree Robotics’ listing on Shanghai’s STAR Market.

Or, more likely, you did not.

That is exactly the point.

This story is not really about one robotics company. It is about an entire market that most global investors rarely see: China’s AI-driven technology rally.

Over the past 18 months, the Shanghai Composite Index has risen 17.3 per cent. That is a respectable return, but it has not beaten the S&P 500 or the Nasdaq. The real gains have happened elsewhere, among the technology companies listed on ChiNext, Shanghai’s STAR Market and the Hong Kong Stock Exchange.

Once you move beyond China’s broad market indexes, the picture changes considerably. Some Chinese AI-related indexes have even outperformed the Nasdaq over the same period.

So why has the rally attracted so little attention?

Subscribe now

The simplest answer is that global financial media rarely covers China’s technology market in any systematic way. But there is another problem: even Chinese investors cannot easily buy many of the companies driving it.

To trade ChiNext stocks, mainland investors must maintain average daily assets of RMB 100,000 (about $14,800) for 20 trading days and have at least two years of trading experience. Access to the STAR Market requires RMB 500,000 (about $74,000), two years of experience and additional risk assessments.

These restrictions exist because both markets are considered far more speculative than China’s traditional main boards. For foreign retail investors, access is even more limited. Many ChiNext and STAR Market companies cannot be purchased directly at all.

If investors cannot buy the companies behind China’s AI boom, why should they pay attention?

The answer is ETFs.

Hong Kong-listed exchange-traded funds provide one of the simplest ways for foreign investors to gain exposure to China’s technology market. These products track indexes such as the STAR 50 and ChiNext. Over the past 18 months, several have delivered strong returns.

The two themes attracting the most attention are communications infrastructure and semiconductors. Some Chinese semiconductor ETFs have gained more than 110 per cent over the period.

That would be impressive in any market. But foreign investors already have easier alternatives. The VanEck Semiconductor ETF gained roughly 95 per cent over the same period, while the iShares Semiconductor ETF rose around 85 per cent. Both contain familiar companies such as Nvidia, TSMC, ASML and Broadcom.

For most investors, buying those funds is much easier than researching an unfamiliar Chinese semiconductor company through financial reports and news published in another language.

That is why ETFs are the natural starting point. Before choosing individual Chinese AI stocks, investors need to understand the broader cycle. Index funds offer a way to follow the market, capture some of the upside and learn how Chinese technology stocks behave without constantly monitoring a single company.

Understanding China’s AI Indexes

Stock indexes are usually designed to represent a particular industry or part of the economy.

ChiNext was created for entrepreneurial and high-growth companies that may not qualify for China’s traditional main boards. The closest comparison is the Nasdaq, although the two markets are far from identical.

The STAR Market is more specialized. Established in Shanghai in 2019, it focuses on frontier industries such as semiconductors, software, biotechnology and advanced manufacturing. It is essentially a dedicated financing market for technologies that China considers strategically important.

Semiconductor indexes narrow the theme even further, concentrating on chip designers, manufacturers and equipment suppliers.

Together, these indexes capture much of China’s AI investment story. But their composition changes. STAR Market companies are reviewed quarterly, while ChiNext is adjusted every six months. As companies enter and leave, the investment thesis can shift with them.

The largest holdings therefore matter. They show where Chinese investors believe the next technology cycle is forming.

ChiNext: From Electric Vehicles to AI Infrastructure

The five largest ChiNext holdings include CATL, Innolight Technology, Eoptolink Technology, East Money Information and Sungrow Power Supply.

CATL remains the defining company of the index. The battery manufacturer represents China’s previous technology cycle: the electric-vehicle and renewable-energy boom of 2020 to 2022. Sungrow, one of China’s largest solar-inverter manufacturers, belongs to the same story.

When renewable-energy valuations collapsed after 2023, ChiNext entered a difficult period. Many investors believed its growth story had already peaked.

The market began to recover after Chinese state-backed funds intervened in late 2024. But the next real catalyst came from somewhere unexpected: American spending on AI infrastructure.

As US technology companies poured money into data centres, Chinese suppliers of supporting hardware became new market favourites. Optical-communications companies were among the biggest winners.

Innolight and Eoptolink manufacture optical modules, which connect AI servers inside data centres. Their rise reflects a simple fact. The United States may dominate advanced AI chips, but an AI system requires much more than GPUs. It also needs networking equipment, printed circuit boards, optical modules, cooling systems, storage and power infrastructure.

China remains a major supplier of many of these components. American AI spending can therefore create substantial orders for Chinese manufacturers, even as the two countries compete over advanced technology.

To understand ChiNext, it is not enough to ask whether AI is growing. The more important question is which Chinese industries are entering a new capacity-expansion cycle, and which companies are gaining weight inside the index.

Those changes often create the largest opportunities.

STAR Market: China’s Semiconductor Bet

The STAR Market’s five largest holdings include Cambricon Technologies, Advanced Micro-Fabrication Equipment Inc. China (AMEC), Hygon Information Technology, Semiconductor Manufacturing International Corporation (SMIC) and Montage Technology.

The market was created in 2019, at the height of the US-China technology rivalry. Its purpose was clear: to provide capital for companies working in industries China considered strategically important, especially semiconductors, software and advanced manufacturing.

The timing mattered. China was facing growing restrictions on advanced technology imports from the United States. Semiconductor independence was no longer simply an industrial goal. It had become a national priority.

Cambricon came to symbolize that ambition.

Founded in 2016, the company develops AI chips intended to compete with Nvidia’s accelerators. Chinese investors saw it as more than another semiconductor start-up. It represented the possibility that China might eventually build its own AI computing infrastructure.

This gave semiconductor companies a special position on the STAR Market. Investors were not valuing them only on revenue and profit. They were also pricing in geopolitical expectations.

That optimism did not last uninterrupted. The global semiconductor cycle weakened between 2021 and 2022, and many Chinese chip companies suffered sharp valuation declines. Cambricon continued to report losses after its listing, damaging investor confidence. By 2023, the STAR Market had fallen heavily as enthusiasm for semiconductor independence faded.

The recovery began in 2024. State-backed funds increased support for the broader stock market, while global AI investment created a new semiconductor cycle. Chinese investors returned to companies involved in AI chips, manufacturing equipment and domestic substitution.

By 2025, the STAR Market had become one of the clearest expressions of China’s AI infrastructure boom. Cambricon’s return to profitability and stronger demand at SMIC helped push the index higher.

The risk is equally clear. The STAR Market is heavily concentrated in semiconductors. If AI hardware spending slows or the global chip cycle turns down again, the market will suffer more than a broader technology index.

Two companies capture the entire argument. Cambricon represents China’s ambition to design competitive AI chips. SMIC represents its ability to manufacture advanced chips despite US restrictions.

Together, they reveal both the potential and the limitations of China’s semiconductor strategy.

Semiconductor ETFs: More Upside, More Risk

Chinese semiconductor ETFs take the same thesis and concentrate it further. Their largest holdings typically include NAURA Technology, AMEC, Cambricon, SMIC and Hygon.

These funds cover not only chip designers and manufacturers, but also companies producing the equipment required to make semiconductors. AMEC manufactures chipmaking equipment, while NAURA is one of China’s most important domestic suppliers.

The investment logic is straightforward. If China succeeds in building a more independent semiconductor supply chain, these companies should be among the largest beneficiaries.

But concentration works in both directions. When semiconductor shares rise, these ETFs can outperform broader technology indexes. When the cycle turns, their losses can accelerate just as quickly.

For foreign investors, they are probably the purest available bet on China’s semiconductor-independence campaign.

How America’s AI Boom Lifted Chinese Stocks

The connection between American AI spending and Chinese technology stocks may appear contradictory.

The United States leads in advanced AI chips, cloud computing and frontier models. China faces restrictions on many of the technologies required to compete.

So why did Chinese AI stocks rally?

There are two main reasons.

The first is that AI requires much more than chips. Large-scale AI deployment also needs optical modules, printed circuit boards, advanced packaging, cooling systems, storage and power infrastructure. China remains highly competitive in many of these areas.

Investors buying Innolight or Eoptolink are not betting that China will replace Nvidia. They are betting that every Nvidia chip will require thousands of other components around it.

The second reason is self-reliance.

When US-China tensions rise, Chinese investors tend to favour companies involved in domestic substitution. The logic is easy to understand: if foreign technology becomes unavailable, a domestic alternative becomes more valuable.

The danger is valuation. Many companies associated with self-reliance trade on expectations of what they may eventually become rather than what they currently earn.

Cambricon, for example, has traded at more than 250 times forward earnings. Nvidia has generally traded at around 40 to 60 times earnings.

That does not necessarily mean Chinese investors are irrational. It means they are placing a very high price on the company’s potential future role.

The difficult part is knowing when that future has already been priced into the stock.

The IPO Machine Behind the Boom

China’s AI rally has also produced a wave of technology IPOs.

Most companies choose one of three destinations: ChiNext, the STAR Market or Hong Kong. The mainland markets have stricter approval requirements, with regulators paying close attention to technological innovation and revenue growth.

Once a company lists, however, price movements can become extreme. Newly listed ChiNext and STAR Market companies face no daily price limits during their first five trading days.

That creates a peculiar market. Institutional investors and early shareholders can push prices sharply higher during the first week, while retail investors rush to buy what they hope will become the next technology champion.

From my own tracking, Cambricon’s IPO lottery had a winning probability of around 0.05 per cent. CXMT’s was roughly 0.47 per cent. Unitree Robotics’ was around 0.018 per cent.

The odds really do resemble a lottery. The returns sometimes do as well. Some Chinese technology IPOs have produced short-term gains of around 200 per cent.

But a successful IPO is not necessarily a successful long-term investment.

Many Chinese technology stocks rise sharply after listing, then decline over the following one to three years as investors wait for earnings to catch up. Companies trading at 300 to 500 per cent above their IPO valuations can eventually fall by 50 to 70 per cent.

The current AI boom and abundant market liquidity have interrupted that pattern, at least for now. But investors should remember that China’s stock market has historically been better at financing companies than rewarding shareholders.

In many cases, investors are using their own money to fund technology that may not produce profits for years.

What IPO Fundraising Reveals

Fundraising size offers another useful clue.

Chinese AI-model and robotics companies often struggle to raise more than $1 billion. The companies that exceed that threshold usually have existing revenue, established customers or clear strategic importance.

CXMT is one example. Innolight and Sanhua Intelligent Controls fit the same broader pattern.

The market may speculate heavily on futuristic AI products, but the largest pools of capital still tend to flow toward companies with proven industrial capabilities.

This creates a hierarchy within China’s AI market.

Foundation-model companies usually choose Hong Kong because they need large amounts of capital and access to international investors. The best-known examples are Zhipu AI and MiniMax.

Zhipu received investment from Prosperity7 Ventures, a fund backed by Saudi Aramco. MiniMax attracted capital from IDG Capital.

Investors in these companies care less about another benchmark score than about a much more basic question: can the company build a sustainable business?

MiniMax initially focused on consumer products, including its AI companion app Xingye and the Hailuo video model. Those products did not fully convince investors.

Zhipu placed greater emphasis on coding and enterprise applications, making it look more like a Chinese version of Anthropic. Its share price has risen dramatically since listing.

Before companies such as Kimi and DeepSeek reach the public market, Zhipu is the most visible listed bet on whether China can build a profitable foundation-model company.

Where China’s AI Money Actually Goes

AI hardware companies are easier to understand if we divide them into four groups: semiconductor equipment, semiconductor components, memory and storage, and AI devices.

Equipment companies such as NAURA, AMEC and Piotech build the machines required to manufacture chips. They do not need to produce the final AI processor themselves. They sell tools to the companies that do.

This business model has one major advantage: every domestic chip manufacturer needs equipment. As US restrictions limit China’s access to advanced foreign tools, local suppliers become more strategically valuable.

Component companies benefit from a different part of the cycle. Innolight, Eoptolink and suppliers linked to Unimicron gain from the construction of AI data centres. They are not replacing Nvidia or TSMC. They are selling the supporting hardware that AI infrastructure cannot operate without.

Memory and storage provide perhaps the clearest example of how China values industrial capacity.

CXMT is China’s largest DRAM manufacturer and one of the country’s most strategically important semiconductor companies. Unlike many AI start-ups, it already operates a large manufacturing business.

Its IPO demonstrated something important. Chinese investors are willing to award extremely high valuations to companies that combine national importance with actual industrial capability.

The final category is AI terminal manufacturing: smartphones, glasses and other devices that bring AI directly to consumers.

This market is much less certain. AI phones and glasses may eventually become enormous businesses, but adoption remains unclear. For now, investors are often paying for possibilities rather than earnings.

Physical AI: Exciting and Dangerous

The most difficult companies to value are those connected to physical AI: humanoid robots, autonomous vehicles, robotaxis, lidar and industrial robotics.

They represent one of the most ambitious visions of AI. They also tend to have the weakest financial performance.

The reason is simple. Software can scale quickly. Physical products cannot.

Once trained, a foundation model can serve millions of users. A robot still requires factories, components, assembly, testing, maintenance and deployment in the real world.

That creates a large gap between technological promise and financial reality.

Chinese robotaxi companies offer a good example. Pony.ai and WeRide represent some of China’s strongest autonomous-driving technology, but profitability remains distant. Their shares fell roughly 50 per cent within 18 months of listing in Hong Kong.

The market is not asking whether autonomous driving is impressive. It is asking whether these companies can build profitable businesses before investors lose patience.

Lidar suppliers face a similar problem. Hesai Technology produces sensors used in autonomous vehicles. Lidar was once treated as an essential technology for self-driving cars, but suppliers came under pressure as automakers began cutting costs.

Hesai’s shares fell more than 90 per cent after its IPO. The technology was valuable. The company’s pricing power was not.

Humanoid robots are more difficult still.

Companies such as Unitree have attracted enormous attention by producing functional robots at relatively low prices. But manufacturing humanoids at scale remains extremely challenging, and much of the expected demand does not yet exist.

Investors are betting on a market that may take another three to five years to mature.

The physical-AI companies that perform best usually have real businesses beyond the AI narrative. Sanhua Intelligent Controls is a good example. It supplies robotic actuators, but it is also an established Tesla supplier with existing industrial revenue.

Its valuation does not depend entirely on robots finding jobs.

What Foreign Investors Can Buy

After looking through China’s AI market, one conclusion becomes obvious: access matters almost as much as opportunity.

Most overseas retail investors cannot directly buy ChiNext or STAR Market stocks. Their realistic options are Hong Kong-listed Chinese AI companies, Hong Kong-listed ETFs and companies available through international trading programs.

Foundation-model companies such as Zhipu and MiniMax offer the highest potential returns, but also the greatest risk. Their valuations can move dramatically after a new product release, customer announcement or partnership.

AI hardware companies are generally easier to analyse because many already have industrial customers. Iluvatar CoreX, Biren Technology and Innolight depend on continued spending on AI infrastructure, but at least the source of demand is visible.

Some AI application companies already have functioning business models. CloudMinds provides robotic services, while computer-vision companies in the Horizon Robotics mould sell technology into existing industries. Their returns may be less spectacular than those of speculative AI-chip companies, but their businesses are easier to understand.

Physical-AI companies remain the most difficult category. Investing in robotaxis, humanoids or lidar requires belief not only in the technology, but also in a commercialization timeline that may keep moving further away.

China’s AI rally is real. The harder question is deciding which companies are producing earnings, which are producing useful hardware, and which are still producing little more than a very expensive story.

But the story does not end here. China’s technology IPO wave is still unfolding. We have compiled a list of companies that could potentially go public next, offering investors a preview of where the next AI and hardware opportunities may emerge.

Part II: How to Actually Buy Into China’s AI Boom

1. A Global Investor’s Map of Chinese Stocks

“Chinese stocks” is not one market. It is at least four markets layered on top of each other, each with its own rules about who is allowed to buy in.

Mainland China has five boards. The Shanghai and Shenzhen Main Boards list large, established companies. ChiNext, on the Shenzhen exchange, was built for growth companies; its closest (imperfect) analogue is the Nasdaq. The STAR Market, on the Shanghai exchange, is narrower still: a dedicated financing venue for semiconductors, software, biotech and advanced manufacturing. The Beijing Stock Exchange, smaller and less liquid, focuses on early stage innovative companies.

Hong Kong has its own Main Board, where most large Chinese tech listings now land, and GEM, a smaller board for early stage companies.

Access to each of these depends on who you are:

For overseas investors specifically, STAR Market and ChiNext shares reached through Stock Connect are limited to eligible institutional professional investors; this is not a retail channel. QFII/RQFII status offers a separate route into the same stocks, but that is an institutional license that a fund applies for, not something an individual investor signs up for directly.

The practical routes into mainland China from outside the country are: Stock Connect (the Shanghai Hong Kong and Shenzhen Hong Kong links, generally used through a Hong Kong or international broker with Connect access), and QFII/RQFII status (Qualified Foreign Institutional Investor / RMB Qualified Foreign Institutional Investor), which is really an institutional license, not something an individual applies for directly. A small number of large Chinese tech names also trade as ADRs on US exchanges, though most AI hardware and semiconductor names covered in Part I do not have a US listing at all.

For most global retail investors, that leaves two realistic paths: Hong Kong listed individual stocks (buyable through almost any broker with HK market access), and Hong Kong listed ETFs that hold the mainland stocks you can’t buy directly. For many global retail investors, ETFs are the simplest, and sometimes the only practical, way to access companies listed on mainland boards.

2. The ETF Menu

Saying “buy an ETF” isn’t useful advice unless you know which ETF. Here are some of the actual Hong Kong listed products global investors use to get this exposure, by theme:

STAR Market exposure: CSOP STAR50 ETF (3109.HK) was the first Hong Kong listed ETF to track the STAR 50 Index when it launched in 2021. It is a physical, non derivative fund that holds the underlying STAR Market shares directly, giving international investors direct exposure to Shanghai’s STAR Market. Bosera STAR50 ETF (2832.HK) tracks the same underlying index through a different manager.

ChiNext exposure: CSOP SZSE ChiNext ETF (3147.HK) is a synthetic ETF that uses derivatives to replicate the performance of Shenzhen’s ChiNext board, denominated in RMB. That’s a structurally different product from the STAR 50 funds above, which is exactly why it’s worth checking a fund’s structure and not just its name.

Semiconductors: Global X China Semiconductor ETF, listed with both a USD counter (9191.HK) and an HKD counter (3191.HK), tracks a FactSet China semiconductor index spanning chip design, equipment and manufacturing.

AI and robotics: Global X China Robotics & AI ETF (2807.HK) is built specifically around companies central to China’s robotics and AI development and adoption.

A caution worth repeating to readers: two funds that both call themselves a “China Semiconductor ETF” can hold almost entirely different companies. One may be roughly 40% equipment makers, another 30% foundries, another mostly AI chip designers, another closer to a traditional broad semiconductor basket. Before buying, look at the top ten holdings and the fund’s structure (physical vs. synthetic), not just the name.

Now that we know how to access the market, the harder question is what to own. The answer starts with the supply chain.

3. Buy the Supply Chain

“China AI” is not one trade. Underneath that label are at least six distinct businesses, each exposed to a different part of the cycle, with different margins, different customers and different risk. Breaking it apart this way is also the only way individual stock analysis stops being a stock list and starts being an argument.

AI Compute. Chip design, foundry capacity, and the equipment and hardware that makes both possible.

Cambricon is the purest bet on China building its own AI accelerators. It only turned its first annual profit in 2025, but the numbers since have been extraordinary: half year 2026 revenue reached ¥5.996 billion, up 108% year on year, with net profit of ¥2.311 billion, up 122%, already more than its entire 2025 net profit in six months (its Q1 specific figures are consistent with this pace but should be checked against the original filing before publication). And yet, on the first trading day after that half year report, the stock fell more than 6%. By August 2026 its trailing P/E stood above 250 times earnings, a valuation that prices in years of continued hypergrowth, not just the growth already delivered.

SMIC is the other side of the compute story: not designing chips, but making them. Its low double digit revenue growth and 20% gross margin look almost sleepy next to Cambricon, and that’s the point. SMIC trades at a far more industrial valuation, a fraction of Cambricon’s multiple, because the market is pricing something different: advanced Chinese foundry capacity operating under export restrictions, not a software style growth story.

Two other names round out this layer. Hygon Information Technology pairs a domestically developed CPU with a GPGPU class accelerator (the DCU) built for broad compatibility with the CUDA software ecosystem, a different bet than Cambricon’s from scratch architecture, aimed at customers who want AI acceleration without rewriting their software stack. Foxconn Industrial Internet sits at a different layer entirely: it doesn’t design chips, it assembles the servers and racks that carry them, as a lead manufacturing partner tied to Nvidia’s GB series rollout. Estimates of its share of that assembly market vary widely by source: brokerage research has put it anywhere from roughly a third to over 40% of global AI server output, which is itself a reminder that market share figures in this space are analyst estimates, not disclosed facts, and should be treated as a range rather than a number. Inspur is the domestic facing version of the same idea: China’s largest AI server integrator by shipment volume, selling primarily into the domestic cloud and enterprise market rather than into Nvidia’s global supply chain.

AI Networking. The optical modules, switches and connectors that move data between AI servers.

Part I explained why AI spending reaches far beyond GPUs. Innolight shows what that looks like in financial statements. Its first half 2026 revenue reached roughly ¥41.8 billion with net profit of about ¥13.7 billion, up roughly 240% year on year. A single quarter of 2026 profit (over ¥5 billion in Q1 alone) already exceeded its entire 2024 net profit. By its own half year filing, close to 95% of its revenue is now international. Supply chain analysts have linked its customer base to the largest US cloud providers, with concentration estimates putting its top five customers at around three quarters of sales, figures that should be checked against the company’s own disclosure before publication, since exact customer level breakdowns vary by report and are rarely named directly by the company itself. The stock has risen by several thousand percent since early 2023, and its market cap has at times ranked among the largest on the ChiNext board, briefly overtaking SMIC’s.

Eoptolink is Innolight’s closest domestic peer: the world’s second largest maker of high speed optical modules, with an early lead in LPO (linear pluggable optics), a lower power module architecture that several major cloud customers have reportedly been evaluating as an alternative to traditional DSP based modules. The investment case rhymes with Innolight’s: explosive revenue growth tied to a small number of large overseas customers, and a stock price that has already priced in years of continued hypergrowth.

A smaller, higher margin name in the same layer is Tfc Optical Communication (天孚通信, 300394.SZ), which makes the passive and active optical components that go inside modules like Innolight’s and Eoptolink’s rather than the finished modules themselves. Its 2025 revenue grew 59% to ¥5.16 billion with net profit up 50% to ¥2.02 billion, and its gross margin, above 53%, is roughly double what a module assembler typically runs, because Tfc sits one layer further upstream, closer to the component IP. Overseas revenue was about 74% of the 2025 total. It’s a useful pairing with Innolight and Eoptolink: same AI networking theme, same customer concentration and currency risks, but a different position in the value chain and a structurally fatter margin.

AI Infrastructure. Power, liquid cooling, PCBs and the physical plant an AI data center needs to actually run.

Envicool, China’s best known liquid cooling systems maker for data centers, is worth including here precisely because its 2026 story complicates the simple “AI infrastructure = easy money” narrative. Full year 2025 revenue grew 32% to ¥6.07 billion, but net profit growth had already slowed to 15%. Its first quarter 2026 report then showed revenue still up 26%, but net profit collapsing more than 80% year on year to single digit millions of RMB, a swing the company’s own filing attributed to a combination of foreign exchange losses, rising credit impairments and negative operating cash flow (worth confirming against the original 2026 Q1 report before publication).

Sugon (中科曙光, 603019.SH) sits one level up the stack: servers, high performance computing and, through its stake in the liquid cooling specialist Zettabyte (曙光数创), infrastructure cooling itself. Its growth is far more measured: full year 2024 revenue actually fell 8.4% while net profit rose 4%, and its first quarter 2026 report showed revenue up 24% with net profit up 22%. Next to Envicool’s volatility, Sugon reads almost boring. In a sector this prone to single quarter blowups, that may itself be the more interesting data point.

Two more names extend the infrastructure layer beyond cooling. Hu Dian (WUS Printed Circuit, 沪电股份, 002463.SZ) makes the high layer count PCBs that go inside AI servers and switches, a component more removed from the AI cooling headlines but one every AI server needs, and one whose growth is tied more mechanically to unit shipment volumes than to any single customer’s story. Runze Technology (润泽科技) is a different kind of infrastructure bet entirely: not a hardware maker, but a third party data center operator. Its customer concentration is unusually explicit and unusually high: per its own 2025 annual report, ByteDance is its largest client, and its top five customers together account for more than 90% of revenue. That’s a risk worth naming plainly, in the same spirit as Innolight’s customer concentration numbers above: Runze’s growth (Q1 2026 revenue up 53.6%, net profit up 35.4%) is riding ByteDance’s AI buildout about as directly as a stock can. Runze has also moved fast on liquid cooling, with roughly 100,000 high density racks delivered to date.

The lesson from putting all four of these side by side: being the right company in the right theme, AI infrastructure demand is real and growing, does not protect a stock from a single quarter’s cost overruns or a single customer’s spending pause wiping out most of the story. What changes is how violently that shows up, and that comes down to the scale, diversification and customer base of the business.

AI Software. Foundation models and the applications built on top of them.

Zhipu listed on the Hong Kong Main Board in January 2026 (2513.HK), an IPO that raised several billion Hong Kong dollars and valued the company at roughly HK$51 billion. For full year 2025, revenue grew 132% to ¥724 million, with second half revenue up 180% versus the first half: genuine commercial acceleration. But gross margin fell from 56.3% in 2024 to 41.0% in 2025 as lower margin cloud API revenue grew faster than higher margin on premise deployments, and full year 2025 R&D spending of ¥3.18 billion was 4.4 times that year’s total revenue. None of that stopped the stock from rallying sharply through the first half of 2026. As of [confirm exact date before publication], it was up roughly 700% since listing, pushing its market cap past HK$400 billion at its peak. Those figures move by the day and should be refreshed to the actual publication date, but they illustrate a valuation moving almost entirely independently of the underlying revenue base.

MiniMax listed on the Hong Kong Main Board in the same window, at an IPO valuation of roughly HK$50 billion, and its market cap likewise expanded into the hundreds of billions of Hong Kong dollars shortly after listing. Between them, Zhipu and MiniMax are the two most direct ways for a global investor to take a position on whether China can build a profitable foundation model business at all, as opposed to a profitable business built on top of one, which is a different and generally safer bet.

iFlytek (科大讯飞, 002230.SZ) is that safer bet’s clearest example: an application layer AI company with two decades of revenue history, not a pre revenue model lab. Its 2025 numbers tell a story of real but uneven progress: revenue grew 16% to roughly ¥27.1 billion and net profit rose nearly 50% to ¥839 million, yet the company posted a net loss again in the first quarter of 2026 (¥170 million, narrower than a year earlier), a pattern management attributes to the seasonal weighting of its education and government contracts toward the second half of the year. Its large model API and MaaS platform revenue grew 263% year on year, striking, but off a small base of roughly ¥385 million. iFlytek is a useful reminder that “application layer AI already has revenue” doesn’t automatically mean “application layer AI already has stable profit.” The underlying business is real, the government and education contracts are real, and the company is still working out how much of that revenue converts into consistent earnings.

Physical AI. Humanoid robots, autonomous driving, and the components underneath them.

Part I already traced the sharpest split in this category: real technology sitting next to distant profitability, visible in how far lidar and robotaxi operators have fallen from their IPO prices. Part II’s contribution here isn’t to repeat that argument. It’s to go one layer down, into the component suppliers that get paid regardless of which robot brand eventually wins.

Sanhua Intelligent Controls remains the cleanest illustration of the “technology vs. economics” divide in this category: it supplies robotic actuators, but its valuation doesn’t depend on robots finding jobs, because it’s also an established Tesla automotive supplier with real, current industrial revenue.

Tuopu Group is frequently named as a lead Tier 1 supplier of linear actuator assemblies for Tesla’s Optimus program, though that claim currently comes from supply chain research and brokerage reports, not from any official statement by Tesla. That distinction matters: this is market intelligence and industry due diligence, not company or Tesla disclosure, and it should be treated with the same caution as any unconfirmed customer relationship until Tesla or Tuopu itself names the arrangement directly. What is confirmed is that Tuopu’s existing automotive actuator business already generates real, current revenue, which is what makes the Optimus opportunity an add on to an existing business rather than the entire investment case, following the same structural pattern as Sanhua.

Green Harmonic (绿的谐波, 688017.SS) sits a layer further from any single customer. As China’s leading maker of harmonic reducers, the geared “joints” inside a humanoid robot’s limbs, it sells into the humanoid robotics ecosystem broadly rather than to one robot maker, which is a different risk profile than being a single company’s named Tier 1 supplier: less upside if any one customer’s program takes off, but also less downside if that customer’s timeline slips.

Orbbec (奥比中光, 688322.SS), a 3D vision sensor maker, occupies the perception layer: the “eyes” a humanoid robot needs to navigate a real environment. Together they round out a three layer picture of the humanoid supply chain: perception (Orbbec), actuation (Sanhua, Tuopu), and the reducers connecting them (Green Harmonic).

The investment lesson doesn’t change from Part I: real technology can sit next to a commercialization timeline that keeps moving. What changes is the object of analysis: not “will humanoid robots succeed,” but “which layer of the supply chain gets paid regardless of which robot brand wins,” and how much of each company’s revenue already exists today versus how much is still owed to a program that hasn’t shipped at scale.

AI Consumer Devices. Phones, glasses and other AI enabled hardware.

The most hyped names in Chinese AI glasses (RayNeo, Rokid, XREAL) are still private and racing toward IPOs, which means the most direct way to invest today is through their listed supply chain, and the two clearest examples sit at opposite ends of “how real is this yet.”

Crystal Optech is the cautionary one. Its core business remains heavily dependent on Apple’s supply chain (optical filters, prisms), with 2025 revenue of ¥6.93 billion (+10.4%) and net profit of ¥1.17 billion (+13.8%), a steady, unglamorous compounder. Its emerging AI optics product lines, glass substrates for AI storage and next generation waveguides for smart glasses, are what investors are actually excited about. But in June 2026, after the stock posted three day gains exceeding 20% on AI glasses enthusiasm, the company issued a filing clarifying that those specific new product lines were still in pre mass production validation or early customer sampling, and had not yet contributed meaningful revenue. The theme was ahead of the business by a wide margin.

Goertek (歌尔股份, 002241.SZ) is the other end of the same trade: a company where AI glasses revenue is already real. It is the primary contract manufacturer for Meta’s Ray-Ban smart glasses, with a reported majority share of that assembly market, and its own AI hardware revenue reportedly reached around ¥7.8 billion in 2025 with company targets well above that for 2026. Full year 2025 revenue was roughly flat at ¥96.6 billion, but net profit rose 48% to ¥3.94 billion, and first quarter 2026 revenue and profit both grew again.

Luxshare Precision (立讯精密, 002475.SZ) is Goertek’s closest rival in the same assembly layer: the number two contract manufacturer for AI glasses and VR headsets, and a supplier for Apple’s Vision Pro. Where Goertek’s AI hardware story centers on the Meta Ray-Ban line, Luxshare’s is more diversified: alongside AI glasses and VR assembly, it has also moved into AI server supply chains, including high speed copper cabling and liquid cooling connectors for Nvidia’s GB series racks. That diversification is the pitch. Luxshare isn’t a pure AI glasses bet the way Crystal Optech is, so a slowdown in any single AI hardware category matters less to its overall business.

At a market cap in the ¥80 billion range in mid 2026, Goertek trades at a far more modest multiple than Crystal Optech precisely because more of its AI story is already showing up in the income statement rather than in the pipeline. Read together, Crystal Optech, Goertek and Luxshare are variations on the same bet: AI glasses and AI hardware assembly becoming a real consumer and enterprise category, priced at different stages of proof, and diversified across different parts of the value chain.

This was a guest post by and their talented writers. Check them out in more detail:

X.PIN

Firsthand deep dives into China’s tech frontiers—AI, EVs, manufacturing, and beyond.

For more China AI and tech coverage also check out:

Recode China AI

Recode China AI

China AI Spotlight: Your weekly guide to China's AI breakthroughs, trends, and stories.

By Tony Peng

Hello China Tech

Hello China Tech

Decoding China tech for global minds

By Poe Zhao

Thanks for reading!

Today in public markets Nvidia Earnings are coming out after the bell. Meanwhile Nvidia is raising prices by as much as 15% in spite of having a monopoly and operating like an AI bank for Neo Clouds for expensive and largely community opposed AI infrastructure projects. First, a bit of context.

In 2025 we saw a declining sentiment towards AI even as Agentic AI failed to move the needle for many businesses. In 2026 we are seeing an acceleration of the backlash against AI datacenters spread to the AI industry as a whole. I’ve been covering this and while it’s not my most popular work, I believe it’s an important coverage to portray the real impacts of the technology. I believe in 2027, the AI backlash becomes way more political in the United States and elsewhere.

The AI pushback might be one of the biggest bottlenecks to AI in the late 2020s, for the 2028 elections, and in the 2030s in the United States. Huge lawsuits against Advertising platforms could reshape the future. AI datacenter moratoriums (tracker) have accelerated in a marked way in 2026 spreading rapidly. While the U.S. is tough on trade with Canada, it’s actually pushing datacenters on them as well.

With an affordability crisis in the U.S. intensifying as wealth inequality accelerates due to the AI boom, the optics for AI looks very bad for huge swaths of the population. With less than three months until the midterm elections, opposition to AI data centers is becoming a “bipartisan rallying cry” in a growing number of U.S. states.

Share

“Polls show that roughly 70% to 75% of Americans oppose building large data centers near their homes, and more than half support a national moratorium.”

A Spring 2026 (May) Gallup poll found that 70% of US Americans would oppose a data center being built in their local community – and more than half support a national data center ban.

In the three months since the Gallup Poll that number has likely increased! For my American readers, you may know that in mid August Pennsylvania, Democratic Gov. Josh Shapiro signed an executive order placing harsh standards on data center development in his state.

Subscribe now

Is the AI Backlash getting Political in 2026?

Read more

Welcome Back!

Visions of AI is a new feature format I’m experimenting with that will amount to a short profile on an AI or emerging tech startup. The cadence of this style of article is unknown as of yet, but there are a lot of fascinating startups I want to discuss and share about.

This is designed to be light evening reading to go out at a time-slot of 8 pm EST.

In an AI boom you would expect new kinds of AI startups to emerge and get funding. While China has what it calls the world robot conference, to complement the vast limitations of large language models (LLMs), another kind of model is being pushed forward - the elusive world model.

Subscribe now

A Starting definition of a World-Model

“A world model is an internal mental simulation constructed by an AI system that represents the physics, spatial relationships, geometry, and dynamics of the real world.”

As we build the physical infrastructure to power the compute for some much debated and uncertain AI-future, all these robots and IoT systems needs a better understanding of their environments.

Given the limitations of LLMs, it’s not clear how long or how useful these physics based and simulation friendly world models might take to be able to fast-track the training of our suddenly humanoid looking robotics projects.

While China has established itself a formidable open-weight model builder, that western companies are increasingly building upon (like Harvey building a model on a Moonshot AI base), the new race to world models is heating up.

As a Canadian it was exciting to hear that Khosla Ventures and Radical Ventures are leading the seed funding of what it turns out to be one of the largest seed financings ever raised by a Canadian startup. It’s called Veeda and is by three former Nvidia employees with ties to the University of Toronto.

Veeda Innovation will focus on building multi-modal world models.

View Mission

Embodied intelligence and physical AI are really hitting a feverish pitch in the late summer and early Fall of 2026. On August 19th we learned that renowned computer scientist Sanja Fidler, is setting up shop with former Nvidia colleagues and raising $90 million in a seed funding round today. Sanja is a Slovenian-Canadian AI researcher held in very high regard.

Read Introduction

There are now around a half dozen very good world model startups with extremely good funding.

Share

“We started Veeda because we believe robotics will reshape the world—changing how we move people and goods, how we manufacture and build, and how we operate in the physical world.”

The publication called The Logic was one of the first to break the story. If you are a robotics enthusiast this is very exciting. As Anthropic sets for a record breaking IPO there’s some indication it has serious interest in the field and perhaps even acquiring Physical Intelligence.

Helping Robots Learn and Train Faster

Read more

As Anthropic’s IPO approaches likely in October, likely just 8 to 12 weeks from now - it’s beginning to attract a different kind of attention. SpaceX, Anthropic and OpenAI are likely to be the biggest trio of IPOs the U.S. public markets have ever seen. While OpenAI’s revenue growth continues to slow, Anthropic’s ARR continues to accelerate even at a very high scale.

Subscribe now

• Anthropic told investors over the weekend that its annualized revenue run rate hit $65 billion at the end of July, 2025.

• While OpenAI said its ARR has hit $40 billion.

Since Anthropic includes marketplace partner commissions (gross basis) in its reported ARR, whereas OpenAI reports metrics net of Microsoft’s cut, Anthropic’s relative ARR might be slightly overstated compared to OpenAI. But even while ChatGPT rushes to Ad markets, OpenAI’s slowdown is sort of a big deal pre-IPO. This at a time when routing and code wrappers favor Chinese open-weight models over OpenAI model usage due to cheaper token costs. OpenAI reduced API prices for the GPT-5.6 family on July 30, 2026, even going so far as to cut the cost of GPT-5.6 Luna by 80%. What an act of desperation. The revenue tells the real story.

Open-Source is disrupting OpenAI’s Revenue Growth

So ironically OpenAI that makes around half of its revenue from B2B vs. B2C, risks being disrupted by Open-source models out of China, even as Anthropic doesn’t feel the same bite. Anthropic has a premium advantage as the global leading frontier lab for coding LLMs.

While Anthropic even had its best Mythos model blocked, its revenue doesn’t seem to have been impacted yet: and, like I had predicted months ago, Anthropic has blown by OpenAI and might have 2x the revenue OpenAI does by the time OpenAI finally goes IPO. It’s hard to believe, and it’s one for the history books.

Share

If SpaceX’s IPO valuation is anything to go by, it (Anthropic) should be targeting a $2 Trillion valuation. SpaceX went public at an outrageous $1.75 trillion valuation while not being close to profitability and is accelerating Capex way too fast. I believe Anthropic could have close to a valuation of $3 Trillion by the end of 2026, because we’ve never seen growth like it has shown.

Historic Anthropic Revenue Growth in 2026

Anthropic's Enterprise AI strategy experienced an inflection point in 2026.

On a quarterly basis, Anthropic already nearly has double OpenAI’s revenue in Q2, and OpenAI has been accelerating fast the last few years. We’ve essentially never seen a company or an AI startup like Anthropic, in history. We know that OpenAI has higher compute costs, has worse financials and executes much more poorly on AI product. As the market becomes more competitive, it will continue to be squeezed. We simply won’t consider it a frontier lab for very much longer.

Anthropic's revenue for the second quarter surged to $11.5 billion, up from $787 million a year ago. It's been growing at an incredibly fast pace.

It doesn’t matter that ChatGPT has 800 million “weekly” active users, ever since its marketshare dropped below 50% and it’s market share continues to rapidly decline. Anthropic’s success as the dominant Generative AI API, is accelerating the entire Cloud computing industry by itself.

So how much has Anthropic’s Revenue increased in the last two years?

It’s hard to understand how growth could continue even at this scale (of the tens of billions) ? It’s meaningfully changed the trajectory of Google Cloud and AWS growth who both have significant equity (and partnerships) in the AI startup.

Anthropic's ARR has increased between 2,900% (from year-end 2024) and 64,900% (from start-of-year 2024).

Anthropic is thus the single primary tailwind for BigTech Earnings in 2026. Nobody grows 14x as this scale, nobody. For all the marketing gimmicks of OpenAI and insane confidence in SpaceX (including sadly some Pension funds), its three companies carrying the entire AI boom: TSMC, Nvidia and Anthropic. These companies are the backbone of what we call Generative AI of the last three years.

• Yellow line is Anthropic 🟡

• Green line is Nvidia 🟢

• Blue line is TSMC 🔵

If Nvidia wants to subsidize compute as the “Bank of AI” (so prolific is it in circular and vendor financing), it should enable Anthropic to have compute parity with OpenAI. Nvidia is subsidizing Neo Clouds and the costs of compute at radical levels artificially inflating the value of the cost per token.

Anthropic is increasing ARR around $9 Billion per quarter in mid 2026. Anthropic's Annual Recurring Revenue (ARR) has been growing roughly 3x to 4x faster than OpenAI's over recent quarters, driven largely by enterprise adoption and tooling like Claude Code. By the time OpenAI goes IPO, Anthropic won’t be a comparable company. It won’t have a peer competitor, do you understand what that means? Not Google, not OpenAI, not DeepSeek or Alibaba, not even close. By 2030, Anthropic will have achieved AI Supremacy.

OpenAI really is Slowing

Read more

Good Morning,

Visions of AI is a new feature format I’m experimenting with that will amount to a short profile on an AI or emerging tech startup. The cadence of this style of article is unknown as of yet, but there are a lot of fascinating startups I want to share about.

This is designed to be light evening reading to go out at a time-slot of 8 pm EST.

Generative AI has been a force of nature when it comes to the future of software engineering and building products. Cognition, maker of the AI coding agent Devin are in talks to raise at a valuation of $40 Billion.

Devin belongs in the same category as the likes of Cursor, Manus AI and other coding tools that have changed the future of software engineering, agentic AI augmented productivity and automating repetitive coding tasks to the extent that Anthropic models have gotten progressively better at coding. If Anthropic’s ARR is nearing $65 bn that’s due to the utility of its coding models and the success of its API.

Fast forward to the later half of 2026, Cursor is part of SpaceXAI, Meta’s acquisition of Manus AI has been blocked, and it’s now almost two years since Devin was first released. I’ll admit I was fairly skeptical of Devin and Cognition AI at the onset. A single stat line changed my mind.

Cognition is now reporting at least a $40 billion valuation based on achieving a $1 billion annualized revenue run rate, according to sources cited by Bloomberg. While we are living in an era of higher valuations for private AI related companies, not many AI startups have achieved $1 Billion ARR in under two years.

View Devin AI

I was incredibly bullish on Cursor before SpaceX swooped in and now with Stripe likely acquiring OpenRouter, the consolidation period appears to be in full-swing. How much revenue would Manus AI even be making if it hadn’t taken the early big deal with Meta? In the mid 2020s so much of an AI startup’s fortunes seem to be tied to the pedigree of their investors and Venture Capital funds.

You might remember in July, 2025 Cognition acquired AI coding startup Windsurf, days after that company lost its CEO and other employees to Google. OpenAI’s incompetence has ironically benefitted Cognition since it was OpenAI that has reached a tentative deal to acquire Windsurf in a $3 bn. deal but the deal got vetoed apparently by Microsoft, since it would compete with Github Copilot.

Devin is showing Automating Coding is Key Generative AI Advancement of the 2020s

Read more

Good Morning,

Visions of AI is a new feature format I’m experimenting with that will amount to a short profile on an AI or emerging tech startup. The cadence of this style of article is unknown as of yet, but there are a lot of fascinating startups I want to share about.

This is designed to be light evening reading to go out at a time-slot of 8 pm EST.

Not that many AI researchers have worked at Google DeepMind, OpenAI, xAI and started their own companies. Igor Babuschkin in 2026 is the founder of River AI. River is building a stack for personal AI that is both open-source and adapts to you. This is a bit like the holy grail of personalization combined with all the recent advances in AI.

I actually find myself attracted to the philosophy and goal of the startup:

We imagine a future where your AI works entirely for you and deeply aligns with your values. Compared to the corporate chatbots of today, it will feel completely different: responsible, curious, and truly yours. Most importantly, you'll own everything: the hardware it runs on, the data it learns from, and the intelligence itself.

This has legit open-source vibes to me and the interview with the lead investor General Catalyst felt authentic:

River AI is Hiring

The interview is with Hemant Taneja, the CEO of General Catalyst, one of the lead investors of the Seed/Series A.

Chapters
02:22 — Co-Founding xAI
04:36 — An AI Experience You Own
05:59 — Betting Early on Open Weights
07:47 — How Do You Make Money on Free Models?
09:25 — Why the US Needs Its Own Open Models
11:27 — Enterprises Want to Control Their AI
13:36 — What Is Continual Learning?
16:56 — An API That Updates the Weights
19:17 — A Personal AI for Everyone
20:46 — The 20-Minute Training Run

A 2-Month old Startup

River AI is among a group of AI labs that are taking on hard problems with quick capital not seen previously in the history of Venture Capital. It’s also very clearly a bet on Open-Source American AI.

Subscribe now

Read more

Hey Everyone,

AI Things, Bits and Bites is a new AI news orientated feature I’m experimenting with to quickly catch up with what’s going on in AI. I’m not sure of the cadence of this feature article type yet, but they will follow a shorter newsy format that you can browse, scan and absorb easily. Or, at least that’s the plan. [Edit: these shorter articles would appear around 9:00 AM Eastern time.]

So what’s going on in AI? These are some of the stories I’m watching this week and over the past few days.

Amazon Robotaxi Zoox Enters AV Race

Amazon’s robotaxi Zoox has launched its first paid service in U.S. this August. Las Vegas is first. San Francisco and other markets should follow. It’s Sin City to the max, with more robotaxi competitors entering the fray even as Uber is investing again in autonomous vehicles. A lot of consumers totally forgot about Zoox. Paid rides on Zoox became available for the first time ever, with services that launched in Las Vegas this Monday (Aug. 10). Tesla robotaxis that only use cameras have been a huge disappointment. While Waymo is the outright leader in the U.S. There’s no steering wheel here so more room. The seating feels like public transport with these weird carriage-style seating arrangements.

Autonomous Driving Fleets by Vehicles

Nvidia Organizes Financing Consortium with Wall Street for AI Infra

When you are financing Neo Clouds, why not go all in? Nvidia makes a rather outrageous claim that AI factory compute is becoming an investable asset. It’s a $500 Billion third-party financing gig. Nvidia has cash to spare and will backstop up to $125 ​billion, or 25% of the potential deals. Financial partners include at least Apollo Global Management, Blackstone, BlackRock (via Global Infrastructure Partners), Brookfield Asset Management, Goldman Sachs, and KKR. Nvidia has been guilty of some creative circular and vendor financing deals in what we can only assume is a U.S. National defense AI infra mandated project where they think they need all of this compute. We are entering larger and larger gigawatt scale AI clusters of datacenters, even as HBM memory chips and other bottlenecks make said compute way more expensive pushing hyperscaler Capex higher for multiple reasons. Now Datacenters are increasingly being financed on debt including off-balance-sheet margin debt.

Meta Claims to be an Open-Source Good Guy

Mark Zuckerberg wrote an OpenAI or Anthropic-esque “essay” about the future of AI called The Future is for Everyone. While suddenly pretending that Meta isn’t a Metaverse company but a benevolent AI open-source lab. Meta released Muse Glimmer, a neat little model that appears to be yet another pivot in their rather baffling and constantly shifting AI strategy.

Muse Glimmer competes favorable with Alibaba Qwen3.6-27B and has a favorable no strings attached Open-source permission. So at least they finally executed on a frontier model, in a manner of speaking (it’s been a while).

Subscribe now

Unitree goes IPO staking a First Mover Advantage in Humanoids

Read more

Good Morning,

Recently I’m trying to speculate, ponder, analyze and think not just about what AI will do to the future of work or the workforce but capitalism, rule of law, economics and the labor market as a whole in terms of the socio-economic caste systems it is contributing to in the United States. I find myself impacted by the gulf between the haves and have-nots. The demand for compute of AI is actually stressing the entire system. I am concerned the pretext of (which I will hereby refer to as ) the demand for compute is eroding democracy itself. Everytime the numbers come out (like the job numbers), I see the spectre of AI impacting them. Later in this article I discuss some of the reasons the AI bubble is building negative momentum.

When I think about the tightening labor market in the U.S., my base case is that AI will accelerate the erosion of equality, increase the churn of peak aged men in the labor force and erode the share of GDP that goes to workers. These are of course already trends with considerable and significant momentum in the U.S. history of capitalism and democracy. The economics of the elite is the new normal, where many of us have noticed that American Venture Capitalists are like royalty in the Trump Administration and even present as the chosen authority in the “task forces” of the Kevin Warsh led of the Fed. What does that tell you about the future? When you look at the economic picture and macro trends, a few things begin to become clear.

• Long-term unemployment fell in July, according to the July 2026 jobs report - not because they found work, but because they stopped looking. Most of the new jobs are in healthcare, so American men are having a harder time of it inspite of being more likely to be so-called AI-natives (workers who default to AI to solve problems).

• U.S. workers saw their share of the U.S. economy (GDP) slide to a record low in the second quarter, according to the Bureau of Labor Statistics (BLS).

• When 23% of the US workforce is 55 or older: the retirement wave is coming but AI isn’t likely to make up for this churn of talent from the labor market, especially when immigration has been slowed by The Trump Administration to a trickle.

Obviously the unemployment rate is now an outdated metric to track the important things that are happening here. Yet this is all of these years later the key metric the President, the political media, The Federal Reserve and economists tend to use. With the gig economy and with aging populations and some of us taking care of our parents, it’s complicated. The future of work is no longer binary.

U.S. Labor Participation Rate is Trending Down

The U.S. labor force participation rate decreased to 61.4% in July. That’s far lower than most countries. AI is reshaping hiring and skill needs by more in areas of efficiency, flatter management systems at companies and reducing costs to be able to afford AI adoption, whose systems aren’t demonstrating or showing actual higher productivity or significant ROI.

I anticipate Generative AI will push labor participation down as a whole new generation needs to discover for themselves what the future of work might mean with less entry level opportunities in the traditional sense in this low-hire environment. AI is both a disruptor and a catalyst here. When people leave the job market and are no longer looking for work, they simply aren’t counted. On Friday we learned that Nonfarm payrolls fell 23,000 in July, confounding economists’ expectations for an 80,000 increase - that’s quite the shortfall.

Payroll revisions tell us a bleaker story than we were being told: In 2026, 1.4 million American workers have left the labor force.

Excluding the pandemic years 2020 and 2021, the labour force (participation rate) is at its lowest since 1976.

Generative AI doesn’t appear to be bringing more opportunities for American workers since 2022, in fact the K-shaped low-hire-low-fire economy with higher inflation is pretty toxic for quality of life and basic life affordability. Not just less opportunities, but more difficult choices in real American families and single households.

But the declining labor participation rate has mostly to do with an aging workforce, less immigration and a tighter labor market, some of which might be caused by contributing factors related to AI as well. The mental health impacts of technology and isolation many American men are feeling and experiencing aren’t present in any of these numbers.

But U.S. workers aren’t seeing the benefits of the GDP boost the datacenter roll-out is bringing the U.S. In fact, U.S. workers again saw their slice of the U.S. economy slide to a record low in the second quarter amid ​an ongoing productivity boom that is producing output gains ​which are outpacing wage growth, the Bureau of Labor ⁠Statistics reported on Thursday August 6th, 2026. If we tally the meta trend scores, it’s worrying to me:

• Labor participation rate is down.

• Workers share of GDP is decreasing

• Exodus of worker from labor market is increasing

• The percent of prime age men who are not working is increasing decade over decade (after each crisis)

• The youngest and oldest workers are getting penalized in an AI-first job market the most

• Workers aged 55+ make up 23.2% of the U.S. workforce and many are taking and will take early retirement

• The debt being taken out in the AI Infrastructure push might worsen America’s debt-to-GDP ratio, national debt and the ability to make good decisions for the nation in the coming years

• The cost of compute, and rising demand for AI compute will be peaking just as U.S. National debt payments begin to dangerly compound sometime in the next decade or next 15 years

• Margin debt is rising so fast (in mid to late 2026) it is a leading indicator of a market bubble event

U.S. Debt to GDP Ratio might be Accelerated by AI’s Demand for Compute

The U.S. federal debt-to-GDP ratio is approximately 123% based on recent federal and economic reports showing a national public debt of roughly $37–$39 trillion compared to an annual gross domestic product of around $30–$31 trillion. This is expected to get worse in the coming years likely accelerated by the costs of building AI compute, fabs, datacenters and the circular funding of the project.

U.S. debt to GDP is increasing rapidly:

The Demand for Compute will lead to Sticky Inflation

It’s commonly understood that AI will significantly increase inflation or at the very least make higher inflationary more sticky (at least at this stage in the cycle). The annual U.S. inflation rate is currently about 3.5% for the 12-month period ending in June 2026, down from 4.2% in May.

Public debt-to-GDP ratios remain elevated in many advanced and emerging economies, as well as inflation that could be more systemic due to the pressure of the demand for compute. Even as the demand for compute is stoking geopolitical competition that’s creating dangerous bottlenecks that’s in a circular fashion pressuring inflation and the cost of said compute.

Read the Report

Trump’s bizarre Tariffs, the impulsive Iran war combined with his belligerently business friendly pro AI policies, are all incredibly inflationary. Consumer sentiment among American workers towards AI has been decreasing rapidly especially in the 2024 to 2027 period. Not surprising Trump’s polls regarding the war or the his performance on the U.S. economy appear to be correlated.

The very same younger men who tend to use Generative AI more frequently and who are more likely to be power users are the very same workers who seem to be most hurt by the economic consequences of AI generally speaking. You could even make the argument that the demand for compute of AI is hurting younger men at their peak working prime. The tighter labor market has a Gremlin looming, this is also the spectre of AI. While higher persistent inflation and lower wage gains (that aren’t keeping up) means they are becoming poorer during the AI boom. The potential of AI agents to increase layoffs is a persistent fear in several industries.

Obviously, when wage growth outpaces inflation, workers gain purchasing power. When inflation rises faster, real wages drop and everyday items become less affordable. So what happens when this occurs to an entire generation of men? Stuck in an affordability crisis as a tighter labor market favors women who tend to be more involved in the healthcare system, in service jobs that are more resilient and taking care of a fastly aging population?

During this AI boom in the mid 2020s in America, the top 10 percent of America’s financial elite are contributing nearly 50% of the consumer spending, according to Moody Analytics in January, 2026. Those top earners, defined as Americans making at least $251,000 in 2024 according to Census data, drove 49.2% of consumer spending in the second quarter of 2025. Is AI making us a less equal and meritocratic society? What sort of a world is the demand of compute going to lead to? Will whatever Generative AI becomes - empower society and give us more options? Or begin to take away our choices, freedoms and dignity?

Not only doesn’t Generative AI seem to be boosting productivity in a marked way, over the last three years when it’s become a bigger story of adoption, over the 3-year window leading up to (2023 to 2026) mid-2026, the positive wage growth trend (relative to inflation) you’d hope to see has recently stalled. Generative AI appears to be making many Americans poorer. Especially if they don’t have discretionary savings they can use in the boosted and artificially inflated equity market. This is creating a mounting socio-economic class divides around AI.

If Generative AI and the demand for compute are contributing factors to the widening the K-shaped economy, the affordability crisis in the 2020s should get worse in the 2030s. Furthermore AI’s contribution to GDP via datacenter construction isn’t reaching the wallets of consumers or most workers. It isn’t trickling down because AI adoption remains segmented. Therefore the demand for compute that is being afforded by circular and vendor financing is creating a whole new class of owners, while hurting the working class of the labor market. It’s literally eroding quality of life, affordability and opportunity for many to most Americans. It’s no great wonder they don’t love it.

Just in July, 2026 some 264,000 people left the labour force, meaning they are no longer working or looking for work. Who knows what their situation even is, but it’s probably not good and certainly unlikely to be better due to AI. Perhaps they are older workers who have taken early retirement due to AI’s arrival. Perhaps they are recent graduates waiting for better times. Some occupations have 30% to 50% of workers nearing retirement age and society has neither the labor from immigration, robots or AI agents that will be doing their job going forward. Those things don’t actually exist.

The higher debt levels exasperated by the demand for compute will actually complicate the U.S. dealing with its national debt in a responsible manner. The Federal Reserve supposedly have a 2% target for inflation, but the demand for compute will mean it will be nearly impossible to reach that level anytime soon. The Fed’s independence in an era of AI might be compromised. Generative AI could actually cause the consumer price index (CPI) to become elevated relative to historical norms, just like Apple and others raised prices due to higher HBM costs due to AI’s higher demand for compute and cost of compute (due to all the bottlenecks).

U.S. consumers are becoming addicted to the equity (stock) market during the AI bubble. During the AI boom there are also suddenly “prediction markets” that are being pushed into our lives (as if gambling is a north star of our newfound collective mirror). Some retail investors not in labor market are the very men that are going to be the most hurt when the AI bubble pops. In general, as you know, the vast majority of stock market value is owned by the wealthiest households. Where bottom- and middle-income workers hold minimal equity, meaning surges in corporate profits like we are seeing with the AI boom or capital gains bypass them. The dangerous rise of wealth inequality magnified by the demand for compute, structurally creates and is going to create far more divisions, conflicts and risks in the American system and to our collective future that it now appears in 2026.

The BLS said that workers share of GDP is the lowest since the series began in 1947. The demand for compute is siphoning and imposing a wealth distribution to a minority of owners of Capitalism. This centralization of AI power thus means more workers need to become business owner if they wish to de-risk themselves in the years ahead. The trend essentially means that the benefits of ​AI’s productivity gains are accruing more toward business owners ​and shareholders ⁠than to workers through wage gains. This is going to be profounding disruptive to the American way of life and millions of Americans. Capital is being sucked up from the people to the enormous and accelerating demand for compute, including tax dollars, workers share of GDP and all other metrics.

The recent U.S. Payrolls told us the economy added 103,000 fewer jobs in May and June than previously estimated (these revisions are getting more common). Volatility in job creation and the decline of professional opportunities is not the thing that worries me the most though, it’s how AI’s demand for compute will impact people, consumers, workers and the majority of citizens in the years and decades to come in the big picture. Will all of that compute build an AI that benefits people? I’m not seeing much evidence so far that it will. What am I missing?

Leave a comment

The demand for compute is a voracious ghost in the machine that I don’t think humanity has understood or realized yet. There aren’t many warning signs but rising Capex and debt, but this is only the beginning.

Google parent Alphabet Inc. raised an additional $25 billion through a massive investment-grade bond sale in August 2026 we recently found out. Alphabet has taken in over $75 billion in total debt financing alongside $85 billion in equity offerings so far in 2026. This is all in addition to Google’s Capex that is projected to be between $195 billion and $205 billion. For Americans and the middle class and lower middle class the demand for compute will strip us of value, belonging and in some cases, housing. The demand for compute could lead to an American debt crisis. Far from alarmist, this is increasingly becoming in the spectrum of possibilities.

The demand for compute does not care about human systems, or well-being or if the labor market has been damaged by the policies of the Trump Administration. It only cares about its own growth, having enough energy, accelerating Cloud computing and advertising revenue, and in a vague way of America having AI Supremacy over the rest of the world, especially China.

The demand for compute cannot be turned off, it’s the Djinni of a technological lifetime ready to disrupt a human world where America is turning away from being a consumer economy and on the way to becoming something else. We don’t quite know what yet. The demand for compute is stoking uncertainty in times when we can barely afford the anxiety. Instead of just doing as you are told, question everything. The motivations for us to adopt AI might not be what you think, or have been led to believe.

Humanity has not experienced the demand for compute before, and it’s not like electricity or the railroad. It’s not simply a resource humans use but a more fundamental change of ownership, livelihood, freedom and dignity. It’s a mechanism and monstrosity of what comes after capitalism.

Addendum

Finally, how do we know we’re in an AI bubble? What are the signs and symptoms that you can think of?

Here are some of mine:

• The massive capital expenditures (capex) on AI infrastructure (datacenters) are far outpacing actual monetization and diffusion to different companies and different sectors of said monetization.

• There is this well-known metric called the cyclically adjusted price-to-earnings ratio. (CAPE / Shiller PE). So historically for the S&P the mean is about 17.40. Guess what it is today? It’s in the area of 42.39. As you may know, the Shiller PE—is a valuation metric created by Nobel laureate Robert Shiller to measure whether a stock market index (most commonly the S&P 500) is expensive or cheap relative to its long-term historical earnings power. We have data since 1928, and we are approaching the 2nd highest ever and might overtaken that 1999 level. While Earnings are great it’s being impacted by radical circular financing. Which for me means some of that demand isn’t real, it’s being artificially inflated.

• Venture Capital is over-investing in very speculative AI startups related to AGI, RSI, AI in science, and “Superintelligence” themed AI labs, that are essentially low probability success moonshots. The glut of VC funding going into AI compared to other deserving industries is out of proportion and not sustainable.

• The demand for compute is creating volatile pricing spikes in cyclical industries like the Semiconductor industry. They are cyclical by definition for a reason. SK Hynix is spending $38 billion on Fabs that to build more HBM. Some of the U.S. bans on Chinese tech actually will create more bottlenecks. All of this could increase the cost of compute and artificially inflate capex via delays and the added bottleneck costs. Even as the actual value of the tech is not being seen on the business level like you’d hope to see.

This Time It’s Different

• I have been bombarded by AI boosting on social media for the last three years at disproportionate levels to pre GPT era times. So frankly, these "this time is different" arguments are really annoying and don’t do macro economics any justice. In a bubble, it’s really normal for people to become irrationally optimistic and abandon traditional valuation rules and believe old economic laws no longer apply. Reversion to the mean is like gravity, it’s not an opinion. That the mainstream media has been trying to brainwash everyone always talking about how revolutionary Generative AI is, to directly benefit the financial elite in various ways is fairly dystopian and undemocratic.

Mega Corps are Embracing Debt

Indeed hyperscalers rushing to do offerings, bond offerings and finding it not so difficult to go into debt and find credit is yet another sign of the AI bubble. A lot of this capital from the debt is just juicing the whole system. Whether that’s to help to pay for TSMC and HBM fabs or build datacenters to accelerate Cloud computing growth.

The Capex race is creating way more centralization on an already centralized U.S. equity market. The concentration of the NASDAQ 100 or S&P 500 in just a few names is itself a sign of a gigantic bubble. This capex extremism won’t end well, because it can’t. It’s not a sustainable practice. The energy grid, water infrastructure and climate risks can’t even entertain as many datacenters as they hope or plan to build. Greed is pushing extraordinary measures in bubble-like euphoria.

The Buffet Indicator

The ratio of market capitalization to GDP is also known as the Buffet Indicator. Not topically as common as the Shiller ratio but interesting nonetheless. It’s currently at around 2.34. The long-term mean going back to 1928 is closer to the level of 0.85. Roughly speaking, at or below 100% would be considered normal.

Valuations are getting really insane in the AI bubble, especially in the private markets often not correlated with revenue growth in any meaningful way. Sometimes not even correlated to existing products. A lot of stellar AI labs don’t have products. The thing is in the real world that’s just not sustainable.

I’m no economists but there’s dozens to hundreds of indictor’s we are in an extreme market bubble caused by AI. Generated by AI, if you will. The runaway accelerating demand for compute benefits the top one percent in American wealth. But like the rise of Bitcoin or other schemes, I believe it’s more like mass fraud than a new paradigm. Of course I may be in the minority with my levels of skepticism, but I’m just a realist. This is not a conspiracy theory but living on an internet that’s clearly deceptive is getting difficult to accommodate.

AI has not been healthy to the labor market and won’t be healthy to inflation. There are structural problems in our economies and systems that AI will make worse, not better.

The M3 Bonanza

The U.S. Federal Reserve stopped publishing official M3 data in 2006, but historical charts from 1960 onward show M3 grew from roughly $298 billion to a peak of over $21.6 trillion by early 2022. The pandemic was the perfect excuse to print more money in a QE spectacle almost in preparation for something else. The AI bubble, is that something else. There’s a lot of liquidity and “cash on the sidelines” that can still be pumped into the AI bubble and equity markets. This is why the bubble won’t burst anytime soon as of mid 2026. A concentrated market is more easy to manipulate. Nvidia’s circular and vendor financing (which it does in self-interest) is the easiest and most transparent mechanism around this.

"Margin debt is the highest it's ever been, and there's a lot of margin debt you don't see because it's not called margin debt. It's called other things. Some hidden, some public. We see a lot of it, and it's high." - Jamie Dimon, CEO of JPMorgan

Margin Debt is at an all-time High

Nominal U.S. margin debt is troubling to me. There was an article in late May by Axios that stuck with me, but since then things have only gotten worse.

Thinking about this chart from John Hussman, showing margin debt relative to GDP spiking to all-time highs, with previous such instances seeming to foreshadow major market downturns. Again I’m not old enough to remember tech bubbles in much detail, but there’s a lot of AI bubble concerns.

Every bubble in history popped when margin debt hit 3% of GDP. (1929, 2000, 2008…) Today it’s at 4.5%. As you can imagine, trading with borrowed money — known on Wall Street as "margin" — can amplify both returns on the way up and losses if the market turns. Typically when even retail traders and consumers are taking out loans to invest in the stock market like we have seen in South Korea and Taiwan, it’s a sign of a major bubble.

Through the end of April, net margin debt hit more than 1.25% of U.S. market cap, near the highest level in records stretching back to 1997. Just three months later, it’s hit 4.5%.

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. The macroeconomics of the AI bubble event is starting to worry me because it’s warping the real American economy and how debt and circular financing are being used to fund this Generative AI movement. AI products that aren’t really demonstrating exceptional value with the scale of diffusion you’d want to see for it to benefit workers, consumers and the future of work.

Get Full Access

A Stunning Heist of Capital

Read more

I’m coming to the realization that Physical AI, embodied AI and robotics is going to become a big deal. It’s going to be part of the Silicon Valley playbook to prolong the AI bubble. Four major humanoid robotics companies, two in China, one in the U.S. and one in South Korea, are going IPO. European robotics startups reportedly raised more capital in the first six months of this year than in the past two years combined, according to Sifted, which is a bit like the TechCrunch of Europe. What this implies is we are entering a new era of automation and robotics.

I just finished a deep dive into one of the best ways to get early stock market exposure to this trend here.

Deep Dive on Agility

Yesterday we learned that Google’s AI leadership is changing forever. Google DeepMind’s roles are changing. Jeff Dean, the longtime chief scientist, is leaving after 27 years to start his own company called Discovery Loop. Leveraging AI for science is now in vogue, and capital does not appear to be a limit in how we achieve the future.

While AI hasn’t yet transformed science and humanoid robots are not functional in society yet, Demis Hassabis is leaving his role as CEO of Google DeepMind to be the unit's Chairman. New roles: Chair of Google DeepMind & Chief Scientist of Alphabet. What’s caught my attention this summer perhaps the most is all the Physical AI seed rounds, new companies and new deals with so much more capital. There’s a robotics blitz coming, just as compute will get much more expensive.

Google Gemini Robotics 2

GR 2 is Google’s most advanced VLA model: capable of intelligently controlling any type of robot, from bi-arms to full humanoids. This is their intelligence layer, Google claims, is powering the next generation of truly adaptable robots. As it takes its first literal steps, this major advance unlocks intelligent whole-body control, advanced dexterity, and multi-robot collaboration. Think of this as a Multi-modal layer for the future of robotics, where this 2.0 ecosystem combines visual perception, spatial reasoning, and real-time motor control across an entire physical frame under a single model suite.

The AI and software brains of robotics is going to have to get much better in the next five years to bring this world of autonomy online, and it will. In autonomous vehicles the next decade is also pivotal with Waymo, Uber, Baidu, Wayve and others set to evolve fast. We will need strong world Models and specialized LLMs for Physical AI that don’t yet exist for humanoids to reach general purpose utility and GR 2 is a bold step in that direction.

Unitree Goes IPO

Chinese robot maker Unitree Technology is expected to be ​valued at more than 50 billion yuan ($7.4 billion) after ‌its planned Shanghai IPO and it marks a time where China is first to market and has a dominant marketshare in robotics with a significant cost advantage. On July 28th, The Trump administration under the FCC, announced a ban on future foreign made robots. For reference, two of China’s leaders are going IPO. Meanwhile Chinese companies Agibot, Unitree and UBTech accounted for the top-three humanoid companies by installation market share last year, according to Counterpoint. Tesla’s Optimus ranked fifth. The IPOs, SPACs (Agility) and Bans means robots are about to hit the prime time.

The Timeline of automation, robots and AVs

Banning Datacenter Components made in China

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. A 24 year old Hedge Fund star has emerged and become more famous by the boldness of his mistakes. The AI theme may feature briefly in this story.

Get Full Access

Good Morning,

I just wanted to share a little note. Or if you prefer to listen. (20 minutes, 19 seconds).

How should we understand the cautionary tale of AI investor and Hedge fund manager Leopold Aschenbrenner in his rapid rise to fame and fortune? For context AI investor Leopold Aschenbrenner worked at the FTX Future Fund (the philanthropic arm of Sam Bankman-Fried's cryptocurrency exchange FTX) before it collapsed. He later joined OpenAI's Superalignment team, and after leaving, founded the AI hedge fund Situational Awareness LP. He had a knack for spotting “bottlenecks” in the Semiconductor industry and predicting them well before stocks surged.

Who intersected Sam Bankman-Fried and Sam Altman in quite the same way while heralding the advent of AGI? I remember reading his Situational Awareness manifesto and wondering who was funding him? Remember Sam Bankman-Fried and Alameda Research led a Series B/early-stage investment into Anthropic in 2021 when the AI startup was valued at roughly $2.5 billion. There are a lot of Anthropic connections in this story. That eight percent Alameda stake would be worth $77.2 billion today.

As you’ve likely already heard, Leopold Aschenbrenner’s hedge fund was forced to sell all of its public stock holdings where Ken Griffin’s Citadel hedge fund swooped in and reached a deal to buy the stocks. Of course now he wants to raise even more capital. Situational Awareness, which grew to as big as $45 billion at one point, posted large losses in recent weeks from declines in AI infrastructure investments like SK Hynix but he still has a huge amount of Anthropic shares and made some decent bets, if a bit concentrated.

He was born in Germany to parents who were both doctors. At age 15 he convinced his parents to let him move alone to the United States and enroll at Columbia University. At Columbia he double-majored in economics and mathematics-statistics. He co-founded the university’s effective altruism (EA) chapter, conducted research connected to the Global Priorities Institute at Oxford, and co-authored work with economist Philip Trammell. At 17 he received a grant from Tyler Cowen’s Emergent Ventures; Cowen later described him as an “economics prodigy” after reading one of his papers. He graduated as valedictorian in 2021 at age 19.

In 2023 he joined OpenAI’s Superalignment team (led by Jan Leike and Ilya Sutskever), which focused on technical approaches to controlling AI systems more capable than humans. He co-authored the paper “Weak to Strong Generalization.” In April 2024 OpenAI fired him; the company cited an alleged information leak, while Aschenbrenner has disputed the characterization and pointed to internal security concerns he raised (including about potential foreign espionage risks). The Superalignment team was dissolved shortly afterward.

In a very short time, now 24, he’s become one of the most tracked Financial Creators on X. Aschenbrenner launched the fund after leaving OpenAI in 2024 and quickly became one of the most watched figures in AI investing because of eye-popping returns, according to CNBC.

In June 2024, roughly two months after leaving OpenAI, he self-published the ~165-page essay Situational Awareness: The Decade Ahead. It argued that AGI could arrive as early as around 2027, sketched a path toward superintelligence, highlighted compute/energy constraints, geopolitical risks (especially involving China), and national-security implications. The piece circulated widely in AI, tech, policy, and investor circles and various AI boosters.

Read Manifesto

His Fund has done fantastically well, where returns since inception had exceeded 1000% and the fund had expanded to around $20 billion, according to the FT. Rumor has it he has to sell about half of his Anthropic shares in the recent Bear market correction on Semis of late July, 2026.

The hedge fund Situational Awareness LP (named after the essay), initially with seed capital of roughly $225 million from backers including Nat Friedman, Daniel Gross, and Stripe co-founders Patrick and John Collison. These are very powerful and influential Angel investors in the AI space. Notably Jane Street, the Quantitative trading firm appears to be a major backer. Jane Street has also been incredibly successful in this AI bubble/boom. It’s a very mysterious connection for me.

Leo became a bit of a cult figure around investors cashing in on the Semiconductor boom. But his success is also going to lead to more success, even though his calls have been incredibly high-risk:

The fund hit $45 billion at the start of July. His biggest bets were SK Hynix, Nebius, SanDisk, Micron, and CoreWeave, all down more than 35% this month (at the time of writing). His shorts on software names like Adobe went against him too. Citadel bought the bulk of what was left. He's 24 or maybe just 25 years old.

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. U.S. AI policy and protectionism appears to be on the rise - but not by the rules of capitalism.

Get Full Access

Good Morning,

We just witnessed one of the most bizarre and confusing Fed Chair speeches ever (read the transcript at the end). In the midst of a war and an AI cycle that is now considered inflationary, Kevin Warsh chaired his second meeting of the Federal Reserve’s rate-setting committee, where officials voted 9-3 to leave interest rates unchanged.

Meanwhile, one of his various task-forces includes the VC Marc Andreessen, cofounder and general partner of Andreessen Horowitz, who this administration considers a “visionary and expert” in AI. What could possibly go wrong? In his address yesterday, Warsh boasted at one point about how he has known all the taskforce “experts” for well over a decade.

No Un-American Robots Allowed

The Trump administration on Tuesday unveiled bans that target imports of new Chinese robots and power inverters, seeking to protect the U.S. AI buildout from national ​security threats and reshore key industries slated for explosive growth.

The Trump administration on July 28th unveiled bans that target imports of new Chinese (and foreign made) robots and power inverters, seeking to protect the U.S. AI buildout from national ​security threats and reshore key industries slated for explosive growth.

The Federal Communications Commission seems to be taking its orders from Trump and indirectly from VCs and angel investors in the typical Erich Schmidt anti-China school of AI and Technology strategy and policy.

• Bars Chinese imports of new humanoid and quadruped robots.

• Bars connected power inverters, which enable renewable energy sources and batteries to connect to grids and data center equipment.

• The Trump Administration is also deliberating whether to ban Chinese open-weight models.

• The U.S. Department of Commerce rolled out sweeping initial export controls banning the sale of top-tier AI accelerators (AI Chips) starting in October, 2022. After almost four years later how did that turn out?

• In May 2024, the Biden administration quadrupled tariffs on Chinese electric vehicles from 25% to 100%. However demand for EVs in the U.S. has plummeted since Elon Musk became more political while Chinese EV makers appear to be growing faster globally and making headway even in Western markets like Australia and perhaps Canada.

Will the U.S. Ban Foreign made Open-source Models?

Read more

If you think this is a story worth sharing, please spread the word. We need to do better. This is not the right future for our descendents.

Share

If you can do me the favor of sharing this article with a friend, family member, coworker or acquaintance, it would mean the world to me. Or restacking it or sharing it elsewhere.

In this document I provide my Great AI Disappointment thesis.

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. I’ve been tracking robotics and I’m noticing a significant convergence with National Defense spending. It’s a DARPA moment coming for our citizens, cities and economies.

Get Full Access

Good Morning,

I want to share what I’ve been thinking about recently as an emerging technology analyst but also from a humanitarian and civilization based perspective. How this trend scales and compounds is going to be extremely capital intensive.

“Automation, tighter labor markets and a transformed future all within one generation. But at what costs?”

What happens when our capacity to build the future outruns even the leverage of our capital systems? When it’s pushed so aggressively, people and communities begin to rebel? The AI, datacenter and robotics capex is going to find out. I am starting to become a little bit concerned. Venture Capital and BigTech are beginning to diverge from the public good in a new era of Neo-Nationalism. A dominant State backed position towards the future of technology is taking shape in the 2020s.

For America’s financial elite, funding AI datacenters, AI chips and HBM chips has been so profitable they will surely do the same thing with robots and physical AI. This is related to the proposed increase in National defending funding that will pour into drones, rocket companies and space-technology as well. At time when America’s national debt is becoming unsustainable - they will do so at a great cost to future citizens and passing it on to future generations. As I track AI’s development and Venture Capital trends this movement seems increasingly unhinged from what consumers and citizens actually want (more on this in future articles).

It’s at a scale of capital expenditures, R&D and funding that is fundamentally divergent from the past. It will profoundly change the human order of things. A capital frenzy period before a decade where a debt crisis will linger like a coming Tsunami. The robots and debt are coming and this introduces a new Geopolitics and pro-innovation and Defense style of Government that we have not seen in many generations (ironically becoming more like China). Including Physical AI in Silicon Valley and Wall Street’s push of AI, that appears to be the next phase in an organized mechanism to extend the stock market boom by several years.

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. I’ve been tracking the U.S. vs. China dynamics of AI and its future for nearly six years.

Get Full Access

How do you summarize the heat that is July, 2026 in the AI industry? It’s been a very bizarre and multi-layered drama. We are witnessing history.

Geopolitics and AI on the Front Burner 🔥

As you likely know, Chinese company Moonshot AI released a new version of its Kimi model called Kimi K3 that is causing a lot of Enterprise AI to switch to open-weight models. It might be one of the biggest AI moments of 2026. With the Iran war unresolved in the strait of Hormuz and the Tech heavy NASDAQ 100 in freefall, it’s becoming a geopolitical and Trump Administration catch-22. There are no clear solutions in war and AI, as it turns out.

Generative AI models are evolving, but likely slowing the revenue growth of AI behemoths OpenAI and Anthropic. With American hyperscalers approaching negative free cash flow via incredible AI capex and datacenter investments, you have to wonder whether it’s all worth it - if Chinese models that are getting larger and more efficient and can replicate the performance while under cutting the cost. It has the potential to become an AI crisis in the stock market even as the Semiconductor boom seems to have hit a bear market correction, after the U.S. listing of South Korean HBM leader, SK Hynix. South Korea (the KOSPI) is now a leading indicator.

With Chinese DRAM maker CXMT about to go public in Shanghai, it’s a very charged China vs. U.S. setup in the future of AI. With DeepSeek’s stunning funding rounds and planned IPOs for DeepSeek, Moonshot, OpenAI and others in 2027, it’s shaping up to be quite a year next year too. DeepSeek raised around $7.4 billion in June last month. We have to assume Databricks has also been one of the beneficiaries of this pivot of Enterprise AI to routing and cheaper tokens of open-weight players in a world where Databricks announced a new round of funding that values the company at $188 billion.

Google’s Gemini 3.5 Pro has been crucially delayed at the worst possible moment. The launch of SpaceXAI’s new model Grok 4.5 was completely overwhelmed by the Kimi K3 moment. Anthropic Fable 5 confusion has given China the ultimate return of that DeepSeek moment vibe back from the dead of January, 2025. OpenAI’s own GPT 5.6 Sol release has also been nearly entirely overshadowed.

The Trump Administration quick to restrict Mythos class models has a serious problem, what if Chinese models are able to approach those same capabilities with open-weight models that were thought to be many more months behind? The Trump administration is showing signs it could ban Chinese cutting-edge models and take drastic steps to try to curtail China’s rise in AI. I thought they were not going to regulate the AI industry. So much for global free market capitalism.

The Brave New World of Token Efficiency Looks Chinese

With Microsoft, Amazon, Meta and even Google behind the Big 7 hyperscalers have lost some of their AI talking points and credibility in this cycle. The mishandling of Mythos class models by the Government and the rise of Kimi K3 type models is the perfect storm that is seeing many Enterprise companies pivot to more rational token usage that’s radically more cost efficient.

While Open-Router isn’t representative of the entire situation, it’s an interesting data sampling point of the overall macro trend: it appears like cheaper open-weight models from China are winning over marketshare.

In my opinion, some of the best Open-weight models from the U.S. are Thinking Machine’s Inkling (about a week old), and whatever Reflection is likely to put out soon. Nvidia’s Nemotron is an often cited alternative to the previous leadership by Meta. The problem is the real lack of leadership in open-source AI in the United States. For America this is a potential disaster in its AI leadership on the frontier of models, tokens and Enterprise adoption.

To make matters even more peculiar, Chinese President Xi Jinping’s most significant recent remarks on artificial intelligence were delivered via a keynote address at the World Artificial Intelligence Conference (WAIC). President Xi Jinping attended in Shanghai the opening ceremony of WAIC 2026 and delivered a keynote speech titled “Joining Hands to Build a Just and Equitable System for Global AI Governance.” China appears to be more advanced in AI governance and regulatory leadership than the U.S. trying to actively build global collaboration around the theme.

Alibaba’s own Qwen 3.8 Max (preview) will also have an open-weight component with an aggressive international token pricing plan. Kimi K3 is not aimed at hobbyist developers but at Enterprise customers in order to ramp ARR before they also IPO in about six months time. The U.S. and China competition in models, even at a time when there are few great application layer products is exaggerating the demand for compute at a time when China has both cheaper token generation and more abundance energy. At a time when the Trump Administration’s key mandate appears to be keeping the AI boom on the stock market rolling for the financial elite and business class. Meanwhile everyone from Anthropic to Moonshot AI are positioning themselves to maximize their IPO hype and revenue sales momentum.

Do Ranking Models even Matter Any Longer?

If you go by Artificial Analysis Intelligence Index, and rank by smartest model, the ranking is:

The state of affairs on peak model performance is roughly as follows:

1. Anthropic – Claude Fable 5

2. OpenAI – GPT 5.6 Sol

3. Moonshot AI – Kimi K3 (open weights*)

4. SpaceXAI – Grok 4.5

5. Zhipu (Z.ai) – GLM 5.2 (open weights)

These lists and benchmarks that they models are trained to perform on are fairly artificial and likely to change next week and certainly by next month.

Compare Models

The “AI Disconnect” is Getting Worse 🔎

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. I’ve been tracking AI sources on LinkedIn long before I started this publication.

Get Full Access

Good Morning

Over the years I’ve spent an absurd amount of time on LinkedIn, for work. Hopefully you don’t have to. But the time you do spend there, should be optimized to bring you the right or best information.

The following list took me a lot of time to hand-pick and curate carefully. Over the coming weeks I plan to refine and expand upon it. I wanted to find the most valuable LinkedIn posters about AI. It’s getting a little bit tricky.

According to Pangram, one in four social media posts are now fully synthetic or AI generated. Of all posts they flagged as AI-generated, two-thirds came from LinkedIn. That’s not to say LinkedIn isn’t useful for networking and for AI News. I wanted to look into who are some of the top people to follow for AI Enthusiasts. This article will be my first humble attempt to do so in quite a while.

According to their report, Substack is the “most human” and LinkedIn is the least. This is not an endorsement (or sponsor) but if you want to try their Chrome extension you can do so here.

LinkedIn has the most longform AI Content; Substack the Least

Pangram claims that only 53.2% of long-form posts on X are identified as being written entirely by humans, even Satya Nadella (see blog version) or Demis Hassabi’s PR looks AI generated. I’m really not that AGI-pilled to overlook this, it’s hardly readable. LinkedIn prides itself on professionals who are credible and the platform has improved a lot in connecting you to your (search) intent, while helping you to track AI News and AI related industry topics is my aim in this list. So if you are more interested in AI at the intersection of your profession, product, marketing, management, AI infrastructure (datacenters), energy, big technology, or whatever else, you can find it.

Beyond the AI Slop

If you are news obsessed and want to stay in touch with the beat around AI, you will have to try to look beyond the slop.

Why Care about the Linkedin Feed for AI?

Not all interesting sources of AI News frequent LinkedIn but a sizeable number do. Especially specialized consultants, venture capitalists, niche creators and educational sources.

• The variety of insights if you curate a good feed can be helpful to give a holistic (and global) picture of what is going on. They are delivering insights on a regular basis that are easier to digest and not always found in Newsletters.

• (None of the links in this list are endorsements, promotions or affiliates).

• Although I have made an effort to include Newsletter writers and European sources.

If you follow these AI sources on LinkedIn, I believe it will improve the quality of your feed to stay in touch with the industry. While I do my best to not include AI executives or VC type figures, some of them do curate great AI News tidbits. I also tried to balance the list away from technical ML topics but to include new developments like datacenter, energy and AI infra considerations. Overall the list is curated for a general reader that is seeking to keep up with the latest AI News with a selection of sources from many angles.

Preliminary Top 2025 AI LinkedIn List

These are my favorite and top picks for LinkedIn sources to follow for AI enthusiasts to improve your feed. This is not an exhaustive list but can substantially improve your access to the latest information relating to the AI industry even if you are only an occasional user of the LinkedIn platform.

• You can save or bookmark this article, or simply take the few minutes required to follow the recommended AI sources below: the list section links called Posts takes you to their latest posts where you can judge for yourself if they are good for a follow.

This article list basically took me the majority of the week to put together. As such it’s going to be exclusively for paid readers of this publication.

There are obviously thousands of decent AI sources on LinkedIn, but if you could only pick a few which are the most informative, useful, inspiring, factual and educational for a general AI enthusiast and reader? This is what I set out to do.

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. I’ve been doing this for five years and before ChatGPT came along.

Get Full Access

How will AI become more accessible? Are older workers who are laid off in a tough spot due to AI? I didn’t think I’d have to write about OpenAI again. But some events are like the echoes of history.

Is AI increasing Ageism in the workplace? 🧐

At a time when there’s going to be a Physical AI and AI wearables race in the consumer AI economy kicking off in 2027, OpenAI’s plans to even compete in the AI device race might be under serious threat, with increased scrutiny and even legal delays. The weekend news is an Apple lawsuit which looks very damaging to OpenAI’s already badly eroded brand and ability to secure meaningful partnerships and business relationships. I’m not sure what kind of Enterprise customer would want to do business with them at this point in 2026. But how about ChatGPT users?

OpenAI’s last 24 hours: (Saturday 11th, of July)

> Top Exec unexpectedly departs (claims health)
> Shuts down browser tool after 9 months (Atlas, another failed product)
> Sued for trade theft by Apple
> Caught selling product to China against sanctions and lastly:

>OpenAI's head of safety systems, Johannes Heidecke, is leaving the company.

The complaint accuses OpenAI of orchestrating a broad effort to systematically acquire and exploit Apple's ‌confidential information and trade secrets is very serious. The evidence looks daming.

Read the Legal Filing

How do you even recover from this? I expect Microsoft to also bring a lawsuit to them in the coming months in relation to their betrayal in favor of Amazon that was against Microsoft’s terms.

Apple is seeking (Ed Ludlow explains):

• A jury trial.

• An order requiring OpenAI to stop the alleged misuse of Apple's trade secrets.

• Destruction of any Apple proprietary materials in OpenAI's possession.

• Redesign of upcoming OpenAI hardware products so they do not incorporate Apple's technology.

Chronology of Events

Read more

👋 Hey there, I’m Mike. Each week I share AI articles at the intersection of tech, business, society and the future. I also try to share the work of writers from all around the world. If you want to support the channel or gain full-access to my work, go here. Read Archives | See Substack Notes | Visit our community Chat | Visit Homepage. I’ve been doing this for five years and AI is still a worthwhile topic.

Get Full Access

Our guest writer today is Weng Honn Kan, from Malaysia. This is the second article in our Claude for Investing series, you can read the first article by here.

Claude AI for Equity Research

Claude’s Fable 5 model has quite the story. Hurry to use it for Investing purposes as Paid Claude users get just four more days with Fable before it shifts to pay-per-token pricing on July 12th. Agentic AI is about to really change how we do investing and pick stocks.

writes the publication How to Invest with AI.

How to Invest with AI

Discover the best AI stock analysis tools and investing strategies.

How to Invest with AI

Discover the best AI stock analysis tools and investing strategies. We test every AI tool with real money to show you what works and what doesn't , so you can build an AI investment strategy that actually pays off.

By Weng Honn

• Weng Lab: Building, Testing, and Reviewing AI Investing Tools, Learn more About.

Related Articles and Resources

• How to use Claude for Investing: the 4-Level System (AI Corner)

• Find the winners of any market trend with claude (Compound with AI)

• Build your own stock analyst with Claude

Read Compound with AI

Let’s get into today’s guide:

Claude Fable Guide for Stock Analysis

The Complete Guide to Using AI to Spot Undervalued AI Stocks

How I set up Claude Fable 5 for stock research , and which of my 3 skills actually need it .

The model they tried to ban is back. Here’s how I point it at stocks.

For about two and a half weeks, the most capable AI model you could legally touch simply… vanished.

On June 12, 2026, the U.S. Commerce Department ordered Anthropic to suspend Claude Fable 5 (and its locked-down sibling, Mythos 5). The trigger was a jailbreak that Amazon’s researchers found , a prompt that got the model to flag software flaws and, in one case, write code showing how one could be abused. Because the order applied to any foreign national and Anthropic couldn’t check every user’s passport in real time, they pulled the plug for everyone.

Then, on June 30, the controls were lifted. Fable 5 came back online on July 1 across Claude.ai, Claude Code, and Cowork.

I’m not going to tell you it was “taken out because it’s too powerful and might shock the world.” That’s the internet’s version. The real version is more useful to you as an investor: the single most capable model Anthropic has ever released to the public was considered serious enough that a government reached for emergency export controls — and it’s now sitting in the same app you already use. That’s the part worth paying attention to.

So this week is a hands-on one. I’ll show you exactly how I set Fable up for stock research, what it’s genuinely better at, and — the part most people get wrong — which of my three skills actually need it, and which ones you’d be silly to waste it on.

Let’s get into it.

In this sharing you will get:

1. What Claude Fable 5 actually is (and why it’s different)

2. What Fable is genuinely best for — and what it’s overkill for

3. The framework: the Fable Decision Grid

4. How I set up my investment AI system using Claude Fable .

5. How to set up Fable for your stock research — step by step

6. Get the right skills to do the right job.

Let’s dive in…

Read more

I took a few days off (still not quite at my usual cadence) not to annoy my American readers who have been celebrating Independence Day and spending more time with family. However there are some urgent developments worth mentioning briefly. A lot of American trends are demonstrating some issues with AI actually not boosting productivity.

In my opinion a major change has taken place in June, 2026. Let me try to explain. On June 12th, 2026 the U.S. government abruptly restricted Anthropic’s Mythos class of models. This among other related things has significantly broken down trust in closed-source models and companies like Anthropic and OpenAI. It’s a big shift.

If 2025 was the year where vibe coding failed to contribute meaningful products at scale, 2026 is the year when tokemaxing led to high Claude code bills and became token-minimization where AI agents failed to demonstrate as much value as it was hoped that they would at some of the biggest and most powerful companies. This has predictably led to more widespread adoption of open-weight models out of China. This is a marked change for the future of AI and illustrates how AI has rising elements of class struggle, geopolitics, AI policy impacted risk mitigation and a problematic declining public perception and negative sentiment towards AI.

I’ve never seen such division in the AI community over closed source vs. open-source models before as I have in the weeks near the end of June, 2026. A world where the U.S. government not only puts Anthropic on a supply-chain risk sanction list (back in February) but restricts their breakthrough Mythos class of models. Where Palantir, a company embedded within the Pentagon and U.S. administration is openly questioning the value proposition of Anthropic and closed-source model makers. This made me very uncomfortable because you have someone embedded in the National defense complex making an argument for open-weight models that China currently dominates.

2 Minutes 49 seconds Clip - Alex Karp CEO of Palantir

While Palantir has business reasons to make this argument1 (since Anthropic is likely a direct Enterprise AI competitor), to watch the Pentagon and Palantir (who are supposed to be protecting American interests) hurt the AI leadership of America with brash arguments that favor China and Chinese AI products is very disconcerting and drankly disheartening.

This is leading to a big transfer from closed-source models to open-source models that will make some of China’s AI startups way more profitable. Many businesses can’t work with AI companies that are restricted or whose models can be shut down at any time, and who are as expensive as Mythos (Fable 5) models are relative to the competition. Businesses have had to pivot. The Trump Administration have significantly damaged America’s AI leadership in the past few weeks.

Meanwhile, the build out of AI infrastructure is not slowing down, it’s speeding up.

The Mega Neo Clouds are Arriving in 2026

In early July, not only has SpaceX but now Meta and Softbank (via SB Neo) have entered the Neo Cloud arena of selling compute and developing massive datacenters with an aggressive monetization plan.

Read more