Author: Chao Xiang Research

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

Following the announcement, the market reacted immediately, but the reaction was extremely polarized. Meta's stock soared over 8% in pre-market trading, while the two flags of the "new cloud computing" sector, CoreWeave and Nebius, plummeted 6% and 10%, respectively. Amazon's stock also turned down in response. On one side was celebration, and on the other was panic; this news acted like a scalpel, accurately slicing through the profit line on the AI computing power supply chain.

The most ironic part is that Meta had just signed a $21 billion computing power procurement agreement with CoreWeave and a partnership worth up to $27 billion with Nebius. Now, it is turning around to compete for business with its own suppliers.

What is going on here?

According to reports from Bloomberg citing insiders, an internal department called "Meta Compute" is leading this initiative. This department is co-led by three executives: Santosh Janardhan, head of Meta Infrastructure; Daniel Gross from Meta's Super Intelligence Lab; and Dina Powell McCormick, president of Meta.

Currently, two business model paths are being considered:

The first is "Model-as-a-Service," allowing external developers to pay to access AI models hosted on Meta's infrastructure, including the Muse Spark model developed by Meta. This path is roughly comparable to AWS's Bedrock service, which essentially opens up the expensive AI model inference capabilities built by Meta as a paid API.

The second, more aggressive path involves directly renting out bare GPU computing power. This is what CoreWeave and Nebius are currently doing, and it's this path that led to the immediate collapse of these two Neocloud companies' stocks. When your largest customer announces it will do exactly the same business as you, your investors are likely to flee.

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

$145 billion bet on computing power needs an "insurance policy"

To understand the core logic of this news, one must first look at a set of numbers.

In 2026, Meta’s capital expenditure guidance related to AI is set at $125 billion to $145 billion, up from a previous guidance of $115 billion to $135 billion. What does this number mean? It is slightly lower than Google’s parent company Alphabet’s $175 billion to $185 billion, Microsoft’s $190 billion, and Amazon’s $200 billion. Together, the four tech giants will 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 of its data centers and GPU clusters are, in theory, solely for its social platform, advertising system, and AI research and development.

This creates a massive risk exposure: if Meta’s internal demand for AI computing power grows less than expected, then the $145 billion capital expenditure becomes a sunk cost. The data centers are built, GPUs are purchased, and long-term power contracts are signed; these are all "rigid" investments that cannot be easily scaled down like adjusting marketing budgets.

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

To translate Zuckerberg's subtext: We are not afraid of spending this much money. If AI computing power is fully utilized, the return on this investment will be reflected through our products; if there is surplus capacity, we can sell it to make money. Either way, it's not a loss.

Wall Street's primary concern has always been that "Meta spends too much on AI without seeing returns," and the cloud computing business provides investors with a safety net: the $145 billion capital expenditure is no longer purely a risk investment, but rather a double-sided bet that can turn offensive or defensive.

The survival crisis of Neocloud

However, the celebration of Meta's investors represents a nightmare for CoreWeave and Nebius.

To understand this relationship, consider a key fact: the entire business model of Neocloud companies is to manage GPU computing power for tech companies that do not engage in cloud computing. CoreWeave and Nebius have secured exorbitant contracts based on the logic that "Meta has huge AI computing power needs but lacks its own cloud business to absorb excess capacity, so it needs to rent from outside."

Now, Meta is saying, I’m ready to do it myself.

This impact is structural. CoreWeave currently has nearly $100 billion in 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 worth of GPU capacity starting from early 2027. If Meta's self-built capacity begins to replace external rentals, the conversion rate of these orders will be in question.

A deeper issue is that Neocloud companies have a weak position in the AI supply chain. They essentially provide a "computing power brokerage" service, buying GPUs from Nvidia, building data centers, and then reselling them to AI companies at a markup. This model can yield huge profits during times of GPU shortages, but when supply bottlenecks ease and major clients start building themselves, the value of middlemen will rapidly diminish.

CoreWeave reported Q1 revenue of $2.078 billion, up 168% year-on-year, but had a net loss of $740 million, which doubled compared to the previous year. Its total debt has exceeded $25 billion. Nebius had Q1 revenue of $399 million, an astronomical increase of 684% year-on-year, but also lost over $100 million, with total debt exceeding $9.5 billion. Both companies are still using high leverage to achieve high growth. If the largest customer becomes a competitor, the "borrowing money to expand production" model becomes particularly dangerous.

The market is already voting with its feet. In June, CoreWeave's short ratio reached 14%, while Nebius was as high as 20%. Investor confidence in the Neocloud sector is being shaken.

The traditional three cloud giants can't rest easy either

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

The global cloud infrastructure market reached a quarterly scale of $129 billion in Q1 2026, growing at an annual rate of 35%, and is moving towards an annual revenue of $500 billion. This market has long been divided among AWS, Azure, and Google Cloud, together capturing over 60% of the market share.

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

Of course, cloud computing is not just about hardware. AWS dominates the market not just because it has data centers, but because it has spent nearly 20 years building a complete product system: from computing, storage, databases to machine learning, security, and the Internet of Things, with over 200 services. For Meta to start from scratch and establish a product matrix of equivalent level requires a massive investment of engineering resources and time.

However, Meta's strategy may not be to completely replicate AWS. A more realistic path is to focus on the most popular and fastest-growing submarket of AI computing power. If Meta only focuses on "AI computing + model services," the barriers to entry are much lower, and it targets the most profitable part of the cloud market.

Seeking Alpha noted that after Bloomberg's news broke, Amazon's stock price turned from rising to falling in pre-market trading. AWS is Amazon's most profitable business, with Q1 2026 cloud revenue growing by 28% year-on-year, the fastest growth rate in 15 quarters. Any new entrant into the cloud market will make AWS investors nervous.

Spending $600 billion on "surplus" can still sell: the deeper logic of Meta's computing power strategy

There is one detail worth pondering repeatedly.

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

What do these figures mean when added together? Meta is building its AI infrastructure on the scale of a "supercomputing nation."

This brings up another more macro background: the real bottleneck in the AI industry in 2026 is neither chips nor funds, but electricity. Just a few days ago, reports indicated that Google had to limit Meta's access to its models because it could not provide enough Gemini computing power. Google Cloud itself has over $460 billion in contracts signed but not yet delivered. Not even the richest tech companies on Earth can purchase enough computing power, not due to lack of money or lack of chips, but due to lack of electricity.

In this context, Meta acting as both buyer and seller has another layer of strategic significance: whoever locks in power and data center capacity first will have a structural advantage in the AI race. The infrastructure built with Meta's $145 billion is both a weapon for catching up with 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, and several dimensions are worth paying attention to:

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

The uncertainty faced by the Neocloud sector has significantly increased. The investment logic of CoreWeave and Nebius is based on the premise that "large tech companies need to rent external GPU computing power." If Meta successfully builds its own cloud business, other tech giants may follow suit, compressing Neocloud's long-term survival space. Of course, in the short term, the supply and demand gap for AI computing power remains enormous, and the existing contracts of Neocloud companies provide a degree of revenue certainty. But their valuations require more margin of safety.

The bigger picture is that the AI industry is transitioning from a phase of "madly burning money to build infrastructure" to a phase of "how to generate returns on infrastructure investment." Meta selling computing power, Open Standard issuing OUSD, and various banks laying out stablecoins, these seemingly unrelated events point to the same logic: when investment scales reach a certain level, capital itself will seek all possible paths for monetization.

For the AI arms race, Meta's move is actually telling the world: the $145 billion is not gambling but infrastructure investment. The characteristic of infrastructure is that once built, it can charge everyone.

Disclaimer: This article is for informational reference only and does not constitute investment advice. Tech stocks and related investments carry high risks, and readers should judge for themselves.