What Meta’s Potential $10 Billion Data-Center Leasing Deal With Anthropic Signals About the AI Trade

Listen to the audio version of this article (generated by AI).

Meta Platforms (META) is in discussions to lease computing power from its artificial intelligence (“AI”) data centers to Anthropic, one of the leading AI model firms. While details are still being negotiated, the deal could be worth up to $10 billion over two years, according to the New York Times.

The news follows Bloomberg reports in early July that Meta is planning an AI data-center unit that could sell raw computing power or access to various homegrown AI models. The move puts Meta in clearer competition with the largest cloud infrastructure businesses: Amazon’s (AMZN) Amazon Web Services, Microsoft’s (MSFT) Azure, and Alphabet’s (GOOGL) Google Cloud.

Some investors hailed the move as a winner, as Meta looks to participate in one of the most important growth engines for these large tech companies over the past decade.

But a closer look may show something much less favorable: an oversupply of AI infrastructure that may be signaling a top to the market, as companies rush to market, heedless of returns.

Meta’s Pivot to AI Infrastructure: Should Investors Feel Bullish or Bearish?

Meta’s announcement that it was building a cloud computing business was not kind to other AI businesses, such as neo-cloud companies CoreWeave (CRWV) and Nebius (NBIS). Both stocks plunged following Meta’s announcement that a more well-financed firm was about to compete more directly with them – and another reason their stocks could hit zero if the AI bubble bursts.

Meta bulls will point to the fact that the company is monetizing its investment in AI data centers, establishing a new business line and bringing in cash as the industry expands. SpaceX (SPCX) made a similar deal in May, leasing its Memphis-based Colossus 1 data center to Anthropic.

But if AI is creating such insatiable demand and such attractive returns, how does Meta (or SpaceX, for that matter) have spare capacity to sell to others?

Meta CEO Mark Zuckerberg has an answer to this question – and it doesn’t look great for bulls.

Meta is using all its current capacity for internal projects, according to a Bloomberg interview with Zuckerberg. “That doesn’t mean Meta has already overbuilt, or has excess computing power available,” Zuckerberg said.

But that’s not inconsistent with Meta being on the verge of overbuilding, either. Other widely quoted remarks from Zuckerberg at Meta’s annual shareholders’ meeting in May suggest that Meta’s pivot into AI infrastructure is exactly the sign that it does or soon will have more capacity than it can handle.

“Almost every week there are different companies that come to us from the outside asking us to both stand up an API service or asking if we have compute that they could buy from us at some premium to what we’ve bought it at,” said Zuckerberg.

“We haven’t done that yet because we think we have a use for the compute,” Zuckerberg said. “But obviously if we get to a point where we feel that we have overbuilt, then that is an option that we have, and that is partially what gives us confidence in investing in building this out.”

Zuckerberg indicates what would get Meta to pivot to AI infrastructure: “if we get to a point where we feel that we have overbuilt.” It sounds much like the “accidental landlords” of the 2008 housing crisis who thought they could turn around and rent out their property if things got tough.

A careful reading of Zuckerberg’s words suggests that Meta is seeing AI oversupply, and Meta’s pivot to become an AI landlord follows up Zuckerberg’s words with concrete actions.

Hyperscalers Continue Investing Hundreds of Billions of Dollars

So, we’re already seeing signs that AI infrastructure may be topping, and yet hyperscalers are still pouring hundreds of billions of dollars into capital investment. The big four hyperscalers – Alphabet, Amazon, Meta, and Microsoft – are set to invest some $725 billion this year, and many analysts project this crew will make $1 trillion or more in capital investments next year.

In other words, many more AI data centers are on their way to market. Meanwhile, the AI market’s dynamics are shaping up dangerously across several key areas:

Key AI Model Companies Are Incinerating Cash

As SpaceX has done, Meta is pivoting to become an AI landlord, with its potential data-center lease to Anthropic. Anthropic and rival OpenAI have been signing huge deals to secure access to computing power for their models, locking in access today for future payments.

But many analysts worry that OpenAI and Anthropic may not be able to meet their spending commitments, which total some $1.4 trillion in the case of OpenAI. Both companies are burning cash – a $20.9 billion operating loss for OpenAI in 2025 and undisclosed amounts for Anthropic.

These model companies may never become profitable, meaning they will have to keep raising cash to fund this spending. If they’re unable to raise the necessary capital, then it is unlikely they will meet all their spending commitments, including those to AI landlords such as Meta and SpaceX.

Massive Debt is Financing AI Investments

Hyperscalers and other companies have taken on massive debt to finance the breakneck pace of AI development.

• Oracle has existing debts of around $130 billion and is promising to borrow more in 2027.

• Amazon raised tens of billions already this year, and is looking to raise an additional $25 billion as part of a new bond sale in July.

• Alphabet issued $85 billion in debt in the 12 months to March 2026, then turned to issuing tens of billions more in stock to fund its AI spree.

• Meta raised $30 billion in debt in October 2025, and used a special purpose vehicle to raise another $30 billion.

With so much leverage in the system, hyperscalers need the rent payments from third parties such as OpenAI and Anthropic. Money must keep moving so that the companies can support the debts – and so much of the system relies on OpenAI and Anthropic raising more capital.

Low-Cost Chinese AI Models Are Capturing Substantial Market Share

On top of these issues, Chinese AI models are increasingly taking market share from American models, reducing the ability of OpenAI and Anthropic to increase their revenues rapidly.

For example, China’s DeepSeek models took about a 16.3% share of all tokens in June, making it the largest single-model firm, according to AI model brokerage platform OpenRouter.

One advantage of Chinese models is their low cost, with many American AI companies charging five to 10 times more, and in some cases 150 times more for high-end models. Users are responding: 80% of startups using open-source software are using Chinese AI models, says Andreessen Horowitz.

While Chinese AI has been known for low-cost models, it’s increasingly growing its capabilities, such as with the recently launched Kimi K3 model, which compares favorably with all but a few high-end American AI models.

With both low cost and high power, Chinese models can hurt the revenue growth of OpenAI and Anthropic, meaning they may not be able to meet spending commitments with Meta and others.

Is Meta Platforms the Newest Accidental Landlord?

Given the leverage involved and the ultra-high expectations built into the stock prices of AI players such as Meta, investors need to be tremendously careful. A debt-fueled investment bubble can spin out of control quickly, crushing the stocks in its wake, and we’re already seeing signs of overcapacity hit markets, even as the debt-financed buying binge continues.

Insiders at some companies are cashing out quickly, too, while they can. CoreWeave insiders sold some $200 million of their own shares in June alone, following a wave of massive selling all year long.

Meta CEO Mark Zuckerberg’s casual pivot on the company’s AI spending – his belief that he can pivot Meta’s AI data centers to cash-hemorrhaging companies such as OpenAI and Anthropic – shows that he doesn’t quite get the severity of the problem. Or at least Zuckerberg doesn’t want to admit it yet. He’s willing to turn Meta into an accidental landlord while spending as much as $145 billion in 2026 alone.

Yet, Zuckerberg has shown a remarkably tin ear when it comes to listening to markets. Remember the failed $80 billion investment in the virtual Metaverse, which sent Meta’s stock reeling into the double digits until Zuckerberg abandoned it? Let’s hope Zuckerberg can anticipate a poor investment much faster this time around, even if the losses are likely to be much higher.

Regards,

James Royal, PhD

Editor’s Note: Marc Chaikin, the founder of Chaikin Analytics, has built an award-winning system that turned “bearish” on software stocks two months before they crashed this year. Now, he’s warning that one AI lab’s breakthrough could CRASH the Nasdaq while igniting a $500 trillion wealth transfer. This 60-year Wall Street legend has found a little-known $40 “pre-IPO backdoor” into the private startup behind this economic sea change. Click here for its name and ticker symbol.