AI Compute Warehouses will Disrupt Meritocracy and Civilization's Energy Grids
Capex in 2026 is set to surpass $400 Billion not counting private companies like OpenAI, xAI, Anthropic and others.
945 terawatt-hours. The IEA's models project that data centres will use 945 terawatt-hours (TWh) in 2030, roughly equivalent to the current annual electricity consumption of Japan. By comparison, data centres consumed 415 TWh in 2024, roughly 1.5% of the world's total electricity consumption (see 'Global electricity growth').
~ Power used by Datacenters by 2030. Read their Report.
Part I: “The Demand for Compute”
I’m a bit obsessed with the great datacenter boom. It’s the most tangible thing in the real world that’s correlated with this obsession of the market and Silicon Valley with AI. There’s mounting evidence American BigTech’s investment in capex will directly correlate with technological automation which Anthropic has warned in its most recent 2025 Economic Index report. If you understand the impact of AI coding on the career ladder of software engineers or the impact on entry level jobs, you know what I’m getting at. In terms of geopolitics too, Generative AI is leading to a new era of colonialism that is fairly worrisome if you aren’t a developed nation as automation will hit poorer regions first and harder. But if you are a consumer at home in a developed country, you don’t just need to pay for AI in multiple subscriptions, you will see mounting energy bills as well. More young people will be unemployed due to AI in the coming decade.
This is not even considering the serious ecological considerations of all of these new datacenters and how they keep getting bigger. The depiction of xAI’s Colossus 2 fits that bill by Semianalysis recently (September 16th). In an AI market bubble where companies are being rewarded for outrageous bets on capital expenditures on AI Infrastructure, the mounting discrepancy between the cost of compute and ROI is going to be a thing to watch. As of late 2025, building 1 gigawatt datacenters went from fantastical back in 2023, to the new normal if you are a major player just two years later, now in 2025. One gigawatt (GW) is a unit of electrical power equal to one billion watts, or 1,000 megawatts (MW) and it’s roughly equal to the energy consumption of an entire city.
“I do guess a lot of the world gets covered in data centers over time” -
- Sam Altman (OpenAI CEO Sam Altman told podcaster Theo Vonn).
BigTech hyperscalers spent $364 Billion on capex in 2025, but this number is set to eclipse $400 Billion in 2026, but this is not counting firms that are still private like BigAI startups such as OpenAI, xAI, Anthropic and others. This means half a Trillion will be spent on AI Infrastructure in 2026 and the demand for compute will keep rising exponentially. Where does this all lead to? It could lead to a certain automation of jobs, rising wealth inequality and companies more powerful than nations.
Even projections of Capex in the recent past turned out to be wrong, in reality BigTech ended up spending a lot more on datacenters, these compute warehouses that seem to get bigger and bigger are becoming a new normal. Capex each quarter is revised up by major MAGA players. Oracle’s stock popped 85% on this trend with a new lucrative deal with OpenAI to begin in 2027. Today, Nvidia, Broadcom and TSCM might as well be called BigTech as well.
OpenAI per the Information expects to spend half of its total revenue on Cloud computing by 2030. OpenAI have a $10 Billion dollar deal with Broadcom to make their own custom AI chips.
For hyperscalers Capex in AI Infrastructure correlates to some degree with Cloud computing revenue growth. Everyone is at capacity with deals spilling out to CoreWeave, Nebius and others. One thing is becoming obvious, the demand for compute is going to be insatiable. Reasoning models, AI agents and greater adoption globally is going to make sure of that.
Nvidia expects a sizable amount of its revenue to come from Sovereigns, other regions like the Middle East and allies of the U.S. signing on to the frenzy, as if this was the expected way to pay tribute to America. Our fair imperial leader in AI. Sovereign AI as far as Nvidia is concerned, means foreign nations buying more its own AI chips, where it is a sizable monopoly. Power in the UK is already very expensive before all of these big datacenters even come into being.
Pretty soon human civilization will have to start putting its datacenters into space, because the current trend is obviously not sustainable. We know now that Data centers in the U.S. could consume as much electricity by 2030 as some entire industrialized economies. But I think we are underestimating the demands on our power grids given the incentives of American shareholder capitalism and recent moves by companies like xAI, Meta, OpenAI and Amazon.
The International Energy Agency works with countries around the world to shape energy policies for a secure and sustainable future. In mid April they released a report that is fairly credible. The IEA’s special report Energy and AI, (304 slides, offers the most comprehensive, data-driven global analysis to date on the growing connections between energy and AI.
In February, 2025 Goldman Sachs said AI would lead to a 165% surge in Energy demand by 2030. I actually think the surge will be a lot more than that and it will test a badly planned and designed U.S. energy grid in the process. The Trump Administration is downplaying the role of clean and sustainable energy sources even when America is years to a decade behind parts of Europe and China. Now these massive data centers are literally coming online faster than power plants can be built and connected to grids. And it’s going to get a lot worse, because the demand for compute will rise faster than anyone can imagine, a bit like global warming two decades ago.
“In the United States, power consumption by data centres is on course to account for almost half of the growth in electricity demand between now and 2030. Driven by AI use, the US economy is set to consume more electricity in 2030 for processing data than for manufacturing all energy-intensive goods combined, including aluminium, steel, cement and chemicals” - the IAE. (International Energy Agency).
The scale of these datacenters enters the realm of science-fiction in the fastly approaching 2030s, where we cannot call them the belittling term of “data centers”, but instead compute warehouses that will stretch on and on, essentially bigger and bigger. Until they will be so large we’ll need to build them in the ocean or in space itself. The hyperscaler Cloud providers or BigAI companies that dominate this AI arms race of LLMs will become way more powerful monopolies than those that we have today in 2025. The Trump Administration is protecting these behemoths as a projection of American power from antitrust scrutiny, whether that’s from the EU or China. America’s bet on itself leading AI Supremacy is not just a bull-market but a National strategy of civilization leading importance.
If you think datacenters are overwhelming the energy grids today in 2025, you haven’t seen anything yet. The cost of AI won’t just be some roles becoming obsolete and some white collar professionals having their careers disrupted. That demand for compute which is an insatiable hunger for electricity that threatens to overwhelm the grid will become more and more powerful politically and in its urgency reshaping nearly every U.S. state (and many entire countries) into a new normal. I believe data center investments may exceed $1 trillion not by 2029 as some have suggested, but by 2027.
What I’m noticing is estimates that are even more than six months old are already outdated. This hunger for compute will break many things in society and the impact of the AI that results from it is more or less uncertain and unpredictable in terms of the economic footprint, wealth distribution and amount of technological disruption and automation that will occur. While it might benefit the United States, it might also harm other regions of the world considerably. States like Virginia are on the front lines of how to negotiate this demand for compute and power. And it’s nothing compared to what’s coming.
2026 Capex: The demand for compute is insatiable
Keep reading with a 7-day free trial
Subscribe to AI Supremacy to keep reading this post and get 7 days of free access to the full post archives.