Meta enters the computing power market, changing the cloud computing landscape.

Author: Chaoxiang Research

On July 1, Bloomberg broke a major story: Meta is forming a cloud computing business unit, preparing to sell its excess AI computing power to external customers.

After the news broke, the market reaction was immediate but sharply divided. Meta itself surged over 8% in pre-market trading, while the two flagships of the "new cloud computing" track, CoreWeave and Nebius, plunged 6% and 10% respectively. Amazon also turned lower in pre-market trading. On one side, celebration; on the other, panic. The news cut through the AI computing power industry chain like a scalpel, precisely splitting the lines of interest.

The most ironic part: Not long ago, Meta signed a $21 billion computing power procurement agreement with CoreWeave and a collaboration worth up to $27 billion with Nebius. Now, it turns around and is about to compete with its own suppliers.

What exactly is going on?

According to Bloomberg, citing sources familiar with the matter, a department within Meta called "Meta Compute" is leading this initiative. The department is co-led by three senior executives: Meta's infrastructure head Santosh Janardhan, Meta's super intelligence lab's Daniel Gross, and Meta's president Dina Powell McCormick.

Currently, two business models are under consideration:

The first is "Model-as-a-Service," allowing external developers to pay for access to AI models hosted on Meta's infrastructure, including Meta's self-developed Muse Spark model. This path is analogous to AWS's Bedrock service, essentially opening up Meta's expensive AI model inference capabilities as a paid API.

The second is more aggressive, directly renting out bare GPU computing power. This is exactly what CoreWeave and Nebius are doing, and it is this path that caused the stocks of these two Neocloud companies to crash instantly. When your largest customer announces it will do exactly the same business as you, your investors will likely run for the hills first.

Meta has not yet made an official comment on this matter.

A $145 Billion Computing Power Bet Needs an "Insurance Policy"

To understand the core logic of this news, you first need to look at a set of numbers.

In 2026, Meta's AI-related capital expenditure guidance is $125 billion to $145 billion, an increase from the previous guidance of $115 billion to $135 billion. What does this number mean? It is only slightly lower than Google's parent company Alphabet's $175 billion to $185 billion, Microsoft's $190 billion, and Amazon's $200 billion. The four tech giants will collectively spend over $700 billion on AI infrastructure this year.

More importantly is Meta's unique situation. Among these four companies, Amazon has AWS, Microsoft has Azure, and Google has Google Cloud. They all have mature cloud computing businesses to absorb AI infrastructure investments and can directly sell computing power to customers. Only Meta does not. All its data centers and GPU clusters, in theory, serve only its own social platforms, advertising systems, and AI research and development.

This creates a huge risk exposure: If Meta's internal demand for AI computing power grows less than expected, then the $145 billion in capital expenditure becomes sunk costs. Data centers built, GPUs purchased, long-term power contracts signed—these are all "rigid" investments that cannot be scaled back like adjusting a marketing budget.

Zuckerberg personally addressed this concern at the shareholder meeting on May 27. He said that cloud computing business is "definitely on the table" and revealed that "almost every week, external companies come to us asking either if we can open up API services or if we can sell them computing power at a premium."

Translating Zuckerberg's subtext: We're not afraid of spending this much money. If all the AI computing power is used, the return on this investment will be reflected through our products; if there is excess capacity, we can sell it to make money. Either way, we can't lose.

Wall Street's biggest concern has always been that "Meta is spending too much on AI with no visible returns." The cloud computing business effectively gives investors a safety cushion: the $145 billion capital expenditure is no longer purely a risky bet but becomes a dual-sided wager with options to advance or retreat.

The Survival Crisis of Neocloud

But the celebration of Meta's investors is a nightmare for CoreWeave and Nebius.

To understand this relationship, first look at a key fact: The entire business model of Neocloud companies is to manage GPU computing power on behalf of tech companies that do not build their own clouds. The reason CoreWeave and Nebius could land blockbuster contracts is the core logic that "Meta has huge AI computing power demand but no cloud business to absorb excess capacity, so it needs to rent externally."

Now Meta says, I'm ready to do it myself.

The impact is structural. CoreWeave currently has nearly $100 billion in revenue backlog orders, a significant portion of which comes from Meta and other large AI companies. Nebius's $27 billion contract with Meta includes reserving $12 billion in GPU capacity for Meta starting in early 2027. If Meta's self-built capacity begins to replace external leasing, the conversion rate of these orders becomes questionable.

The deeper issue is that Neocloud companies already hold a fragile position in the AI industry chain. They essentially provide a "computing power intermediary" service: buy GPUs from Nvidia, build data centers, then mark up and sell to AI companies. This model could generate huge profits during the GPU shortage phase, but when supply bottlenecks ease and large customers start building their own, the intermediary's value quickly shrinks.

CoreWeave reported Q1 revenue of $2.08B this year, up 168% year-over-year, but a net loss of $740 million, doubling the loss from the previous year. Its total debt has exceeded $25 billion. Nebius reported Q1 revenue of $399 million, surging 684% year-over-year, a standout performance, but also lost over $100 million, with total debt exceeding $9.5 billion. Both companies are still trading high leverage for high growth. If the largest customer becomes a competitor, the "borrow to expand" model becomes particularly dangerous.

The market is already voting with its feet. In June, CoreWeave's short ratio reached 14%, and Nebius's was even higher at 20%. Investor confidence in the Neocloud track is wavering.

The Traditional Three Cloud Giants Can't Rest Easy Either

Meta's entry into cloud computing is also bad news for AWS, Azure, and Google Cloud.

The global cloud infrastructure market reached a quarterly scale of $129 billion in Q1 2026, annualizing growth of 35%, heading towards an annual revenue of $500 billion. This market has long been shared by three players: AWS, Azure, and Google Cloud, which together hold over 60% market share.

If Meta officially enters, it will break this triopoly. Moreover, Meta has several unique advantages: it operates one of the world's largest social network platforms, with extensive practical experience in AI models and application scenarios; it has built a strong developer ecosystem in open-source AI (Llama series models); and its AI infrastructure investment scale is already close to the three major cloud giants.

Of course, cloud computing is not just hardware. AWS dominates the market not only because of its data centers but also because it has spent nearly 20 years building a complete product suite: from computing, storage, and databases to machine learning, security, and IoT, with over 200 services. For Meta to build a comparable product matrix from scratch would require enormous engineering resources and time.

But Meta's strategy may not be to fully replicate AWS. A more realistic path is to focus on the hottest and fastest-growing segment: AI computing power. If Meta only does "AI computing + model services," its entry barriers would be much lower, and it would cut into the most profitable part of the cloud market.

Seeking Alpha noted that after the Bloomberg report, Amazon's stock turned from a pre-market gain to a loss. AWS is Amazon's most profitable business, with Q1 2026 cloud revenue up 28% year-over-year, the fastest growth in 15 quarters. Any new entrant that takes a slice of the cloud market will make AWS investors nervous.

Spending $600 Billion on "Excess" That Can Also Be Sold: The Deep Logic of Meta's Computing Power Strategy

One detail is worth repeated contemplation.

In January this year, Meta announced a "Meta Compute" plan, aiming to accumulate "tens of GW" of computing capacity within this decade, with a long-term vision of "hundreds of GW or even more." Meta currently operates over 30 data centers, with AI-optimized facilities under construction ranging from 1 GW to 5 GW. In June, it also signed a 1.6 GW computing power procurement contract with data center company Crusoe.

What do these numbers add up to? Meta is building AI infrastructure on the scale of a "super computing nation."

And this pulls in another broader context: In 2026, the real bottleneck for the AI industry is neither chips nor capital, but electricity. Just days ago, reports indicated that Google, unable to provide sufficient Gemini computing power, had to limit Meta's access to its models. Google Cloud itself has over $460 billion in signed but undelivered contract backlog. Even the richest tech company on Earth can't buy enough computing power—not due to lack of money or chips, but due to lack of electricity.

Against this backdrop, Meta acting as both buyer and seller takes on another strategic meaning: Whoever locks in electricity and data center capacity first will have a structural advantage in the AI race. The infrastructure Meta is building with $145 billion is both a weapon to catch up in AI superintelligence and a "strategic reserve" that can be monetized externally.

Several Key Judgments

The market impact of this news will continue to ferment in the coming months, with several dimensions worth watching:

For Meta itself, if the cloud computing business materializes, it will open a new revenue stream, reducing its over-reliance on advertising. Currently, over 99% of Meta's revenue comes from advertising. Even if the cloud business accounts for only a few percentage points of revenue initially, its symbolic significance and valuation effect should not be underestimated.

The uncertainty facing the Neocloud track has significantly increased. The investment thesis for CoreWeave and Nebius is built on the premise that "big tech companies need to rent external GPU computing power." If Meta's self-built cloud succeeds, other tech giants may follow suit, compressing the long-term survival space for Neocloud. Of course, in the short term, the supply-demand gap for AI computing power remains huge, and the signed contracts of Neocloud companies provide some revenue certainty. But valuations will require more safety margin.

The bigger picture: The AI industry is transitioning from a phase of "crazy spending on infrastructure" to a phase of "how to make infrastructure investments generate returns." Meta selling computing power, Open Standard issuing OUSD, major banks deploying stablecoins—these seemingly unrelated events point to the same logic: When investment scales become large enough, capital itself will seek every possible path to monetization.

For the AI arms race, Meta's move is essentially telling the world: $145 billion is not a gamble; it's infrastructure investment. The characteristic of infrastructure is that once built, you can charge everyone for its use.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. Tech stocks and related investments carry high risk; readers should make their own judgments.