Meta's remark about "selling computing power" crashed AI hardware! Wall Street quickly analyzed: Don't panic, this does not mean excess computing power, this is not an industry turning point.

A piece of news about Meta selling excess computing capacity has brought several of the most sensitive questions in AI trading to the forefront at the same time: Is computing capacity actually in short supply? Will Meta lower its capital expenditure? And how much longer can Neocloud keep making money?

Wall Street Insights mentioned that Meta is working on a cloud business plan, and may offer two types of services to the outside world: one is managed model/API access, similar to AWS Bedrock; the other is renting out “raw computing power” like Neocloud.

As soon as the news came out, shares of CoreWeave—the new-generation GPU cloud star service provider—dropped sharply by 13%, Nebius fell by 15%, and AI hardware sectors such as chips were hit with heavy losses as well. If Meta begins selling computing capacity, investors will naturally ask three questions:

First, has Meta bought too much computing capacity?

Second, is Meta no longer putting as much effort and investment into models and AI products?

Third, is the demand curve for AI hardware and Neocloud about to change?

According to updates from Zhuifeng Trading Desk, on July 1, Wall Street investment banks including UBS, Morgan Stanley, and Bernstein quickly broke down this event. This may not be a collapse of AI fundamentals, but rather a pragmatic move by a giant to find a balance between computing capacity constraints and financial returns. This matter also can’t be simply equated with “Meta no longer needs computing capacity.” But the implications are different for different assets.

For Meta, leasing out computing capacity could be a transitional bridge for revenue and EPS. UBS assessed: “Selling cloud computing capacity or model access rights would, in theory, generate near-term revenue faster than waiting for Meta Business Agents and Meta AI chatbots to scale, and would ease concerns about 2027 EPS staying flat or shrinking.”

For Neocloud companies like CoreWeave, this is potential competitive pressure.

For the chip and server chain, what the market cares about more is whether the pace of subsequent capital expenditure will change.

“Having extra capacity available for rental” does not mean “the industry has excess computing capacity”

The shortest link the market is trading is: rental of computing capacity = excess computing capacity = downward revision of capital expenditure.

Meta may have computing capacity available for lease in phases, but this does not automatically mean there is excess computing capacity across the entire industry. Different institutions also use different definitions for “capacity,” so they can’t simply be added together.

In Morgan Stanley’s model, Meta is expected to add about 2GW and 3.5GW of owned operational IT capacity in 2026 and 2027, respectively, with a baseline of roughly 3GW by the end of 2025. By comparison, hyperscale cloud providers such as Amazon and Google could add new IT capacity on the order of 5GW and 9GW in 2027, respectively. In other words, even if Meta puts part of its own owned capacity out to rent, it will be difficult for it to single-handedly change the overall “cloud provider construction” outlook over the next three years.

Bernstein uses a broader total data center footprint metric: Meta’s estimated global capacity is currently about 20GW, and in the coming few years it will add roughly another 14GW, including both an owned-and-leased mix. The number looks large, but it is not “all rentable AI computing capacity,” nor is it equivalent to the same generation of GPUs, the same types of workloads, or the same pricing curve.

In market estimates, there is also a more aggressive bottom-up back-calculation: using contract and capacity planning as anchors—such as Google with Anthropic, AWS with Anthropic/OpenAI, and Microsoft with OpenAI—future total AI computing capacity for several major cloud providers could all be around 20GW or even a higher magnitude. OpenAI’s own Stargate, as well as arrangements in the 10GW range related to Nvidia and Broadcom, are also folded into the demand-side observation. The point of this metric is not to provide precise predictions, but to illustrate one thing: **Meta’s partial outsourcing of capacity rental is insufficient to prove that global AI buildout is entering a state of oversupply.
**

Even more counterintuitively, Bernstein also mentioned that over the weekend there was news that Google is restricting Meta’s compute usage due to Google’s own capacity constraints. If that claim is true, Meta is, on the one hand, still seeking external computing capacity, while on the other hand preparing to sell out some of its own capacity to the outside in the future. This looks more like a reallocation across “different generations, different purposes, and different time windows,” rather than a simple “it won’t have enough demand and can’t use it all.”

This is not the first time Meta has put “selling computing capacity” on the table

On May 27, 2026, a shareholder asked Meta whether it would build a cloud business to compete with AWS, Azure, and others. Zuckerberg replied:

“Of course, that’s definitely something within our consideration… We haven’t done it yet because we believe we can use that compute ourselves. But obviously, if we get to some stage and feel that we’ve built too much, that becomes an option we have—and that’s also part of the reason we’re confident about continuing to invest in the portions we’re building.”

Earlier, on October 29, 2025, Zuckerberg had also discussed a similar logic:

“Any computing that we don’t need, we’re extremely confident we can absorb a very large portion of… Of course, it’s possible that we build too much. If we really do… We see a lot of new demand both internally and externally. Almost every week, companies from outside approach us, asking whether we can build API services for them, or whether they can get different types of compute from us. We haven’t done that yet. But obviously, if you reach a stage where you’ve built too much, this can become an option.”

This explains why UBS calls it “not new news.”

For Meta shareholders, selling computing capacity is more like an “EPS bridge,” not a new main business

For Meta, the most direct benefit of leasing out computing capacity is turning long-term AI investment into near-term revenue.

In UBS’s table, Meta’s diluted EPS for 2026 and 2027 is about $32.6 and $33.0, respectively. The market concern is that EPS in 2027 will be roughly flat compared with 2026, or even compressed. Leasing out computing capacity or selling model access rights can provide a period of revenue and profit buffering—at least before Meta Business Agents and Meta AI chatbots truly scale.

Morgan Stanley’s sensitivity analysis is even more straightforward: leasing out 250MW of computing capacity, with a one-year lease term, priced at $40 per Watt, could add about $2.97 in incremental Meta 2028 EPS—roughly equal to around 8% upside. If the capacity expands to 500MW, 750MW, or 1000MW, or if the price differs, the EPS elasticity would continue to amplify or shrink accordingly.

This is also why the market didn’t interpret it purely as a negative. From the perspective of Meta shareholders, Zuckerberg effectively has another fallback: if internal AI products cannot consume all the compute in the short term, Meta can first sell it to external AI labs and recoup part of the investment.

The market also draws an analogy to xAI leasing computing capacity to Anthropic: 500MW corresponds to $1.25 billion per month, or roughly $30 billion per GW per year. If that pricing holds, the implied returns are extremely high—on the contrary, it suggests that high-quality computing capacity is still tight in certain scenarios. It isn’t evidence that “nobody wants computing power,” but rather evidence that “idle windows can still be snapped up at high prices.”

But this can only be called a bridge; it can’t be called the main storyline. Morgan Stanley still places the key to Meta’s valuation on front-line product innovation: whether Meta AI, business agents, the messaging business, diffusion offerings, subscriptions, and so on can deliver more durable engagement and revenue growth. Selling computing capacity can supplement EPS, but it cannot automatically lift valuation multiples.

Capital expenditure may not necessarily be lowered; doing it as a full cloud could actually burn more money

What the market is most worried about is that Meta will lower its 2027 capital expenditure, and then the entire AI hardware chain will see expectation cuts as well.

However, Morgan Stanley’s current model assumes Meta’s capital expenditure rises from $145 billion in 2026 to $175 billion in 2027 and $205 billion in 2028. The premise of this model is that Meta mainly builds capacity for its own first-line products, rather than building a complete hyper-scale cloud service provider.

If Meta really grows the external cloud service business—especially building a model/API platform rather than temporarily renting out bare compute capacity—capital expenditure could face upward pressure. Because a full cloud business requires longer-term data center capacity, more complex software platforms, and the ability to deliver to enterprise customers.

Bernstein also looks at this issue beyond 2027. Meta is one of the most important “checkbooks” in the AI market, and any changes to construction pacing will affect the supply chain. But “temporary outsourcing of capacity” and “permanent expansion of cloud business” have different implications for capital expenditure, and they shouldn’t be mixed together.

The bigger demand-side drivers still come from inference and agent applications. HY Computer & AI Computing Power’s market roundup used OpenAI’s weekend article on Codex/agentic AI as a demand signal: the number of individual non-developer users grew 137 times, the number of organizational users grew 189 times, and OpenAI’s internal user count grew 12 times. This perspective emphasizes that the expansion into new scenarios may continue to drive up demand for inference computing capacity.

So the key to this divergence isn’t whether Meta will sell computing capacity; it’s whether the AI demand curve is still steepening. If overseas ARR accelerates, inference applications grow, and cloud providers’ capital expenditure continues to be revised upward, then Meta’s leased-out computing capacity looks more like phase-based monetization of assets. If, in subsequent earnings seasons, capital expenditure is collectively revised downward, then this becomes an industry turning-point signal.

Selling bare computing capacity is easy; building a complete AI cloud is hard

Meta’s potential business has two paths, with completely different difficulty levels.

The first is selling “bare computing capacity” or raw chip capacity, similar to neocloud. Customers are buying GPU/compute resources, and Meta doesn’t need to immediately complete a full set of enterprise software, developer tools, model platforms, and sales systems.

The second is providing managed model/API access, similar to AWS Bedrock or Google Vertex AI. This isn’t a business you can run just by having “a data center, with chips.” It requires that model capabilities, the software stack, developer experience, enterprise customer sales, and service support all keep up.

Morgan Stanley is more cautious about the second path. It noted that Meta’s Muse model family does not perform outstandingly on TerminalBench and SWE Bench Verified—tests that are related to coding capabilities and third-party usage scenarios. If Meta wants to compete with frontier models like Gemini, subsequent models need significant improvement.

This is also where the idea that “Meta selling computing capacity = Meta exiting models” is shaky. The potential plan already includes model/API access. First-line products such as Meta AI, business agents, messengers, diffusion offerings, subscription revenue, and so on remain the core of long-term valuation. The issue isn’t whether Meta will do models; it’s whether it can package model capabilities into a cloud service that external customers will pay for.

Some people in market discussions cite Muse Spark, closed-source strategy, and management adjustments as evidence that Meta is still at the table for models. But these are more appropriate as follow-up tracking items. At least from the three frameworks, the more certain conclusion for now is: the execution threshold for selling bare computing capacity is low, while the threshold for building a full AI cloud is high.

Is CoreWeave the biggest “victim”? Customers become potential competitors

This shock lands most directly on new cloud/GPUaaS companies such as CoreWeave.

Bernstein’s rating for CoreWeave is Underperform, with a target price of $67; Meta is rated Outperform, with a target price of $850. Its logic is very straightforward: if Meta provides cloud infrastructure externally, it could compete directly with CoreWeave.

What’s more troublesome is that Meta itself is CoreWeave’s major customer. Per Bernstein’s figures, Meta currently has $35.2 billion in CoreWeave contracts, accounting for more than one-third of CoreWeave’s order backlog. Adding Microsoft’s roughly $14 billion contract on top, nearly half of CoreWeave’s orders come from customers that could become competitors when contracts come up for renewal.

In the short term, the risks are not as immediate. Existing contract constraints are strong, so CoreWeave likely won’t be able to exit quickly; therefore, CoreWeave’s short-term revenue and debt pressures may not worsen right away.

The long-term problems are harder to deal with. If customers build their own clouds and sell computing capacity themselves, the bargaining power of new cloud companies will weaken. Especially during renewals, CoreWeave will no longer face only demand-side players—it will also face potential supply-side players with money, technology, and data center experience.

J.P. Morgan’s trading desk noted that the market’s relatively easy-to-understand reaction to CRWV falling 13% and NBIS falling 15% is because overnight, Meta went from being a customer to a potential competitor. For chip hardware, the impact is more indirect; for GPUaaS, the impact is more like a stress test of the business model.

Why hardware fell first: beyond fundamentals, there are also crowded positions

On the short-term trading front, the market isn’t only trading fundamentals.

J.P. Morgan’s trading desk split the debate into two sides: one side is whether the Meta news represents a shift in the narrative around CSP capital expenditure and AI compute demand; the other side is that positioning is too crowded, and deleveraging and profit-taking amplified the drop. Their view is that the latter has a higher weight. Whether the fundamentals have truly turned depends on how the upcoming earnings season is discussed.

The positioning backdrop isn’t light. The major index rebalancing just passed, and total flows and leverage start points were already high. Over the past four weeks, increases in both longs and shorts were around the +2 standard deviation level. Over the past five years, hedge funds often deleveraged in July, with changes typically in the -1 to -3 standard deviation range. Semiconductor and memory positions are close to the 100th percentile.

This explains why a single Meta news item could hit the entire AI hardware chain. When crowded trades meet a narrative that “computing capacity might not be scarce,” it’s easy to sell first and ask questions later. The same day, software, crowded shorts, and Chinese ADRs rose by more than 1.4 standard deviations—consistent with short-covering behavior during the deleveraging process.

For the follow-up reversal signals, the market is mainly watching a few things: whether Meta clarifies the situation; whether overseas AI application ARR accelerates; whether cloud providers’ capital expenditure continues to be revised upward; and whether Q2 performance exceeds expectations. The timeline is concentrated in July to August. Right now it looks more like an observation period rather than something that has already formed a consensus conclusion.

There is also a tail risk: the higher the stock price, the harder it is to ignore equity financing rumors.

If Meta’s stock price is pushed up by this “compute can be monetized” narrative, it may actually increase the probability of equity financing rumors.

The logic is that when it’s below 17x FY2027 EPS, Meta is unwilling to do dilutive financing. If this news and strong Q2 performance push the valuation above 20x, the market should not be surprised by potential equity financing.

This is not the main line across the three foreign frameworks, and no company has confirmed it. But it explains why Meta’s stock price reaction may not be straightforward. Selling computing capacity can ease anxiety about returns on investment, while equity financing rumors bring dilution concerns. Both forces influence trading at the same time.

The three valuation frameworks didn’t price Meta as a “compute-selling company”

UBS maintains a Buy rating on Meta, with a target price of $865. Its valuation is based on full-year diluted GAAP EPS of $33.26 as of the entire period ending Q1 2028, using a 26x P/E multiple. Since the company has not confirmed the potential compute selling news, it has not adjusted its forecasts yet.

Morgan Stanley maintains an Overweight and Top Pick rating on Meta, with a target price of $775. Its base case implies approximately a 23x 2027 P/E ratio. The core drivers remain ad revenue, Reels monetization, improved engagement driven by AI, efficiency improvements, and optionality from new products.

Bernstein maintains an Outperform rating on Meta, with a target price of $850, and at the same time maintains an Underperform rating on CoreWeave, with a target price of $67. This pairing very clearly reflects market divergence: Meta’s options have increased, while competitive pressure on CoreWeave has grown.

But risks have not disappeared. Downside factors include: a decline in the advertising cycle, regulatory pressure, uncertainty around the returns from Reality Labs spending, and higher long-term capital intensity due to execution mistakes in data center construction.

The excellent content above comes from Zhuifeng Trading Desk.

For more detailed interpretations, including real-time analysis and frontline research, please join **【Zhuifeng Trading Desk ▪ Annual Membership】**

![](https://img-cdn.gateio.im/social/moments-daffa420b9-5d84ec8f66-8b7abd-7649e1)

Risk warning and disclaimer

          

            The market involves risk; investing requires caution. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial conditions, or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. If you invest based on this, you do so at your own risk and bear responsibility.