AI Snake Oil—Part 2: Where Are All the Data Centers?
In AI Snake Oil—Part 1: The Skeptics, I pulled back the curtain on the massive wave of AI hype being driven by tech CEOs and the media. As the stock market treats the frontier LLMs as the undisputed future of humanity, a growing chorus of prominent computer scientists, tech reporters, and industry defectors are revealing why the AI over-investment bubble will burst soon.
I’m extremely bullish on AI as a technology, but bearish on the business strategy of the hyperscalers (i.e. Amazon, Google, Microsoft and Meta) and their debt-laden clients (i.e. OpenAI and Anthropic). The winners of the AI revolution won't be the companies burning billions to build the "Formula 1 Car" LLMs of today; they will be the nimble, specialized firms that step in to provide highly specialized AI systems to serve specific business problems and deliver actual, scalable value.
Now, in Part 2, we are going to look at the cold, hard facts behind the looming collapse. I’m not talking about the valuation bubble in the stock market—I’m talking about a physical, real-world over-build of catastrophic proportions.
The tech giants are burning cash at an unprecedented rate, promising Wall Street a digital empire, while hiding the fact that their actual construction pipelines are stalled, gridlocked, and drowning in delays.
OpenAI CEO Sam Altman said, “Building AGI that benefits humanity is perhaps the most important project in the world.” and, “Whether we burn $500 million a year or $5 billion—or $50 billion a year—I don’t care, I genuinely don’t… As long as we can figure out a way to pay the bills, we’re making AGI…breakthroughs for mankind [that] are priceless.”
Meta CEO Mark Zuckerberg stated that he would rather risk “misspending a couple of hundred billion” on AI investments than miss the opportunity to develop superintelligence, emphasizing that the greater risk lies in building too slowly.
The $750 Billion Blind Spot
The scale of capital expenditure (capex) from the major hyperscalers is staggering. Driven by the unrelenting race for AI dominance the biggest tech companies have sent their spending forecasts into overdrive. The four largest hyperscalers—Amazon, Microsoft, Google, and Meta—are on track to spend a combined $710 to $750 billion in 2026 alone, about 75% more than they spent in 2025. The vast majority of the spend will be funneled directly into AI infrastructure, including data centers, servers, custom chips, and networking equipment.
Amazon is expected to spend the most: $200 billion; Google and Microsoft about $185 billion and Meta between $125 and $145 billion.
There’s no doubt, data centers are being built, but reliable information about data center construction in-progress is very hard to find and the information that does exist is, in the words of tech reporter Ed Zitron, “continually muddied by terrible reporting.” Incomplete projects are being called “operational” when only part of them are online.
Amazingly, hyperscalers do not disclose how many data centers they’ve built, nor do they disclose how much capacity is online or available.
Amazon, Google, Meta and Microsoft have spent over $800 billion in AI capex in the last three years. They’re public companies. You’d think investors would demand to know what infrastructure they’ve built so far and what progress looks like on the investments they’ve made and are planning. Evidently not.
The 190-Gigawatt Delusion
Straightline Climate is tracking 190GW across 777 large data centers that have been announced since 2024. It’s a vast understatement to say that 190GW is a massive amount of power. The average power output of a single nuclear reactor in the U.S. is about 1GW. And there has been only one nuclear power plant constructed in the United States in the last 30 years. The largest one—the Vogtle Plant in Georgia—generates about 4.6GW. The data centers that have been announced in the last two and a half years require 41 of those.
Building a data center is difficult. In fact, no one has built a 1GW data center yet. Yet investors presumably believe that 190GW worth of them will be built in the next 6 years.
The hyperscalers have said that 16 of 190GW will come online in 2026 across some 140 projects, but only 5GW was under construction as of late February and the remaining 11GW remains in the announced stage with no visible construction progress. The typical build timeline for these projects is 12–18 months. Straightline says, “Projected delivery dates are getting harder to trust,” and they think up to half of data center capacity slated for this year is delayed.
The hyperscalers are confronting backlash from local communities and are increasingly giving up on existing power grids for AI training capacity. More on-site and hybrid approaches are likely for future projects, but for nearly half of the announced projects, the hyperscalers haven’t disclosed how they will power the planned data centers.
Ed Zitron spent a week searching for data centers that broke ground in 2023 or 2024 that have actually been finished and couldn’t find one. “Some projects are stuck in construction hell,” he said, “eternally dueling with planning departments over permitting, some are chugging along with no real substantive updates, some have done effectively nothing for the best part of a year, some are perennially adding more capacity to the order as a means of continuing raking in construction bills, and some are claiming their data centers are ‘operational’ as only a single phase has turned on.”
And after the construction is finished, the buildings themselves must be fully filled with the necessary power, cooling and compute hardware. They need to be configured to meet their client’s specifications and then, maybe, they can start generating some revenue.
Decoding the Media Hype
The hyperscalers are being deceptive—and the media is making it worse by amplifying their tricks. For example, in October of last year, Amazon announced that, “Project Rainier, one of the world's largest AI compute clusters, is now fully operational.”
If you don’t know the difference between "Fully Operational" and "Fully Built," you’re obviously going to think the project is complete and fully online. Nope. Amazon finished construction on 7 of 30 planned buildings in the project and brought online a cluster 500,000 chips. Fully operational means that the data center is officially online, not that the project has been completed. In fact, Project Rainier is a multi-phase, $11 billion project that is supposed to draw up to 2.2GW across its 30 buildings.
Stargate Abilene, a data center campus which is supposed to reach 1.2GW has two buildings online generating about 200MW (that’s 0.2GW). When the first of eight buildings of Stargate opened, CNBC declared that “OpenAI’s first data center in the $500 billion Stargate project is open in Texas,” burying the comment that only one was operational with another nearly complete several hundred words in their report.
Also in that report, CNBC said that OpenAI together with CoreWeave and other partners, said “they are ahead of schedule to meet their full 10-gigawatt commitment by the end of 2025.”
Again, the tech companies are using word tricks that the media refuses to decode for the public. What the companies mean is that by the end of the year they aimed to secure the commitments and locations for the 10GW—not to have construction planned or started let alone have the power turned on.
Building data centers and procuring that much energy takes years. 10GW is enough to power roughly 7.5 million homes. The scale of what has been announced—dreamed about—is what you read in the headlines. What is actually being built is a small fraction of that and what has actually been brought online is a small fraction of that.
Examples of the hyperscalers and their clients, like OpenAI and Anthropic, obviously trying to mislead investors are endless. The media is complicit because in order to maintain their access to these companies, they know they need to amplify their deceptive announcements.
Microsoft says it’s Fairwater data centers are operational, but they’re still unfinished. In September 2025, Microsoft CEO Satya Nadella claimed that Microsoft had added 2GW of capacity “in the last year.” He acted as if the Fairwater data center project—comprising two actively constructed data centers, one in Wisconsin that broke ground in September 2023 and another in Atlanta that broke ground in July 2024—was something to be “announced,” rather than a very expensive project plagued by delays. To make matters worse, Nadella’s announcement was in direct conflict with reporting from Ricardo Torres of the Milwaukee Journal-Sentinel who said that the data center Microsoft said was online wasn’t.
“Fairwater” sounds like a single data center site, but it actually refers to a project where multiple data centers are linked with high-speed networking to make one larger cluster, a massively ambitious project that remains under construction—and may never be finished. In reality, Microsoft wasn’t anywhere close to adding 2GW of capacity. In fact, Ed Zitron reported last week that none of Microsoft’s announced data center capacity since 2024 has been completed.
Data centers are being built, but they are massive projects that are extremely difficult to complete and they’re taking much longer and facing many more challenges than the hyperscalers, their clients and the media want you to believe.
THE ALPHA ROCK DIFFERENCE - Our investment strategies are a compelling alternative to traditional buy-and-hold investing. By focusing on liquid alternative investments and active risk-management, we target absolute returns, not benchmarks. To see how we’re invested and why, download our most recent Quarterly Investment Outlook, The End of Easy Money.
REVIEW YOUR INVESTMENTS - When market volatility increases, it’s a good time to review your investments, especially if you’re using a passive, buy-and-hold strategy. Please schedule a call or meeting)
We’d love to hear from you!
Your thoughts are important to me, so don’t hesitate to share them. They give me great motivation and encouragement.
IMPORTANT DISCLOSURE INFORMATION
This commentary reflects the personal opinions, viewpoints and analyses of the Alpha Rock Investments, LLC employees providing such comments, and should not be regarded as a description of advisory services provided by Alpha Rock Investments, LLC or performance returns of any Alpha Rock Investments, LLC client. The views reflected in the commentary are subject to change at any time without notice. Nothing in this commentary constitutes investment advice, performance data or any recommendation that any particular security, portfolio of securities, transaction or investment strategy is suitable for any specific person. Any mention of a particular security and related performance data is not a recommendation to buy or sell that security. Alpha Rock Investments, LLC manages its clients’ accounts using a variety of investment techniques and strategies, which are not necessarily discussed in the commentary. Investments in securities involve the risk of loss. Past performance is no guarantee of future results.
The S&P 500 Index or the Standard & Poor's 500 Index is a market-capitalization-weighted index of the 500 largest U.S. publicly traded companies. The S&P 500 is a float-weighted index, meaning company market capitalizations are adjusted by the number of shares available for public trading. Note: Investors cannot invest directly in an index. These unmanaged indices do not reflect management fees and transaction costs that are associated with most investments.
Alpha Rock Investments, LLC provides links for your convenience to websites produced by other providers of industry related material. Accessing websites through links directs you away from our website. Alpha Rock Investments, LLC is not responsible for errors or omissions in the material on third party websites, and does not necessarily approve of or endorse the information provided. Users who gain access to third party websites may be subject to the copyright and other restrictions on use imposed by those providers and assume responsibility and risk from use of those websites.
