Meta’s remark about “selling computing power” crashes AI hardware! Wall Street quickly analyzes: Don’t panic, this doesn’t mean a computing power surplus, this isn’t an industry turning point.

A piece of news about Meta selling excess computing power has simultaneously put several of the most sensitive issues in AI trading on the table: **is computing power really in shortage, will Meta lower its capital expenditure, and how long can Neocloud keep making money?** [Wallstreet.cn mentions](https://wallstreetcn.com/articles/3775980?keyword=meta) that Meta is formulating plans for its cloud business, possibly offering two types of services: one is hosted models/API access, similar to AWS Bedrock; the other is “raw computing power” rental, similar to Neocloud. Following the news, star GPU cloud service provider CoreWeave’s stock price plunged 13%, Nebius fell 15%, and chip and other AI hardware sectors were heavily hit. If Meta starts selling computing power, investors will naturally ask three questions: **First, did Meta buy too much computing power?** **Second, is Meta not investing as heavily in models and AI products anymore?** **Third, is the demand curve for AI hardware and Neocloud about to change?** According to "Chasing Wind Trading Desk", on July 1, UBS, Morgan Stanley, and Bernstein, among other Wall Street investment banks, quickly dissected the incident. This may not be a collapse of AI fundamentals, but rather a pragmatic move by giants to seek balance between computing limitations and financial returns. The matter cannot simply be equated to “Meta doesn’t need computing power anymore,” but it has different implications for different assets. **For Meta, renting out computing power could be a transitional bridge for revenue and EPS.** UBS judges: "Selling cloud computing power or model access could theoretically bring quicker short-term revenue than waiting for Meta Business Agents and Meta AI chatbots to scale up, and alleviate concerns over flat or shrinking EPS in 2027." **For Neocloud companies like CoreWeave, this is potential competitive pressure.** **For the chip and server chain, the market is more concerned about whether the pace of future capital expenditures will change.** ## “Having excess to rent out” doesn’t equal “industry-wide computing power surplus” The market trades the shortest chain, **renting computing power = surplus computing power = capital expenditure downgrade**. Meta may have some stage-wise rentable computing power, but this does not automatically mean the whole industry has a surplus. Different organizations use different capacity metrics and cannot simply be added up. In Morgan Stanley’s model, Meta is expected to add around 2GW and 3.5GW of its own IT operating capacity in 2026 and 2027 respectively, starting from about 3GW at the end of 2025. By comparison, major cloud giants like Amazon and Google are expected to add about 5GW and 9GW of IT capacity in 2027. In other words, **even if Meta rents out a portion of its own capacity, it’s hard to single-handedly change the overall picture of cloud construction for the next three years.** Bernstein uses a broader measure for total data center footprint: Meta’s current global capacity is estimated at around 20GW and will add another 14GW in the coming years, including both owned and leased capacity. The number is big, but it is not "all rentable AI computing power", nor does it equal the same GPU generation, same workload, or same price curve. There is also a more aggressive back-calculation in market estimates: using contracts and capacity plans like Google-Anthropic, AWS-Anthropic/OpenAI, Microsoft-OpenAI as anchors, the total AI computing power of the top cloud firms may reach around 20GW or even higher in future; OpenAI’s own Stargate and 10GW plans related to Nvidia/Broadcom are also observed from the demand side. This measure isn’t for precise forecasts, but to make a point: **Meta’s partial renting out is not enough to prove global AI construction has entered excess.** Even more counterintuitive, Bernstein mentions weekend reports that Google limited Meta’s compute usage due to its own capacity constraints. If this is true, **Meta is still seeking external compute power while preparing to sell some in the future—it’s more like a redistribution of “different generations, different uses, and different timing windows” rather than simply “unused excess.”** ## This is not Meta’s first time putting “selling computing power” on the table On May 27, 2026, shareholders asked Meta if it would build a cloud business to compete with AWS and Azure. Zuckerberg responded: "Of course, that’s definitely in scope... We haven’t done it yet because we think we can use the computing power ourselves. But obviously, if at some point we feel we've built too much, that's an option we have, and that's part of the reason why we have confidence in continuing to invest in building.” Earlier, on October 29, 2025, Zuckerberg spoke about similar logic: “We are quite confident that any computing power we don’t need, we can absorb a large part... Of course, it’s possible to overbuild. If we really do so... We see lots of new demand internally and externally. Almost every week, people externally are asking us to build API services or asking if they can get different types of computing power from us. We haven't done this yet. But obviously, if you reach the stage of overbuilding, it could be an option.” **This explains why UBS calls it “not new news.”** ## For Meta shareholders, selling computing power is more like an “EPS bridge,” not a new main business For Meta, **the most direct benefit of renting out computing power is turning long-term AI investment into near-term revenue.** UBS’s table shows Meta’s diluted EPS in 2026 and 2027 is around $32.6 and $33.0 respectively; the market is concerned that 2027 EPS may be essentially flat or compressed compared to 2026. **Renting out computing power or selling model access can at least provide some revenue and profit buffer before Meta Business Agents and Meta AI chatbots are truly scaled up.** Morgan Stanley’s sensitivity calculation is more intuitive: every 250MW of rented computing power, for one year, at a price of $40/Watt, could bring Meta about $2.97 in incremental EPS for 2028—about 8% upside. If capacity increases to 500MW, 750MW, or 1000MW, or prices change, EPS elasticity will further expand or shrink. **That’s also why the market doesn’t just see this as bearish news. For Meta shareholders, Zuckerberg essentially has a fallback: if internal AI products can't consume all computing power in the short term, sell it to external AI labs and recoup some investment.** The market also compares xAI renting computing power to Anthropic: 500MW corresponds to $1.25 billion/month, roughly $30 billion/GW/year. If this pricing holds, the implied return is very high, showing that high-quality computing power is still tight in some scenarios. It’s not evidence of "no demand," but "idle windows can be swept up at high prices." But this can only be called a bridge, not a main line. Morgan Stanley still puts the key Meta valuation on frontline product innovation: Meta AI, business agents, messaging, diffusion offerings, subscriptions, etc.—can they bring more lasting engagement and revenue growth? Selling computing power can supplement EPS but can’t automatically raise the valuation multiple. ## Capital expenditure may not downgrade; building full cloud may actually burn more cash The market's biggest worry is that Meta will downgrade capital expenditure in 2027, and then the whole AI hardware chain will lower expectations accordingly. But Morgan Stanley’s current model assumes Meta’s capital expenditure will rise from $145 billion in 2026 to $175 billion in 2027 and $205 billion in 2028. The model’s premise is: Meta mainly builds capacity for its own frontline products, not for becoming a super-large cloud service provider. If Meta really expands external cloud services, especially model/API platforms rather than temporary raw computing rental, capital expenditure could actually come under pressure to rise. Full cloud business requires longer-term data center capacity, more complex software platforms, and enterprise-level delivery capabilities. Bernstein also looks at this question in 2027 and beyond. Meta is one of the most important “checkbooks” in the AI market—any change in build rhythm affects the supply chain. But "temporary external rental" and "permanent cloud business expansion" have different implications for capital expenditure—they cannot be mixed. Bigger demand still lies in inference and agent applications. HY Computing & AI’s market summary takes OpenAI’s weekend article about Codex/agentic AI as a demand signal: non-developer individual users grew by 137x, organizational users by 189x, and OpenAI internal users by 12x. This perspective emphasizes that expansion of new scenarios may continue to push up inference compute demand. **So the key in this round of divergence is not “will Meta sell computing power,” but whether the AI demand curve is still steepening. If overseas ARR accelerates, inference applications grow, and cloud capital expenditure continues to trend up, Meta renting out computing power looks more like short-term asset realization. If subsequent earnings season sees capital expenditures downgraded across the board, then it becomes an industry turning point.** ## Selling raw computing power is easy, building a full AI cloud is hard Meta’s potential business has two paths, with entirely different levels of difficulty. **First is selling “raw computing power” or original chip capacity, similar to neocloud.** The client buys GPU/computing resources; Meta doesn’t need to immediately provide a complete software suite, developer tools, model platform, or sales system. **Second is hosting models/API access, similar to AWS Bedrock or Google Vertex AI.** This is not a business you can just do with racks and chips—it requires model capabilities, software stack, developer experience, enterprise client sales, and service support. Morgan Stanley is more cautious about the second path. It mentions that Meta’s Muse model family does not stand out in TerminalBench and SWE Bench Verified tests, which are tied to coding ability and third-party use cases. If Meta wants to compete with advanced models like Gemini, their subsequent models need significant improvements. That’s why the logic “Meta selling computing power = Meta exiting models” doesn’t hold. Potential plans already include model/API access. Meta AI, business agents, messengers, diffusion offerings, subscription revenue, etc. are still the long-term core for valuation. The question is not whether Meta will do models, but whether the model capabilities will be sufficient to support paid cloud services for external clients. Some market discussion sees Muse Spark, closed-source strategy, and management changes as evidence that Meta is still playing the model card. But these are better for subsequent tracking. At least from the three frameworks, the most certain current conclusion is: It’s easy to sell raw computing power, but the threshold for full-suite AI cloud is high. ## CoreWeave is the biggest “victim”? Clients become potential competitors The immediate impact hit CoreWeave and other new cloud/GPUaaS companies. Bernstein rates CoreWeave as Underperform, target price $67; Meta as Outperform, target price $850. The logic is straightforward: **If Meta offers cloud infrastructure externally, it could compete directly with CoreWeave.** More troubling, **Meta itself is a major CoreWeave client. In Bernstein’s measure, Meta now has $35.2 billion in CoreWeave contracts, making up over one-third of CoreWeave’s order backlog.** Add Microsoft’s ~$14 billion contract, and almost half of CoreWeave’s orders are from clients who may become competitors when renewing. Short-term risks are not so direct. Existing contract constraints are strong and hard to exit immediately, so CoreWeave’s short-term revenue and debt pressure may not deteriorate right away. Long-term problems are harder to solve. If clients build their own cloud and sell computing power themselves, new cloud companies lose bargaining power. Especially at renewal, CoreWeave faces not just buyers but potential suppliers who have money, tech, and data center experience. J.P. Morgan’s trading desk notes that the market’s reaction—CoreWeave (CRWV) down 13%, Nebius (NBIS) down 15%—is understandable: Meta overnight became a potential competitor instead of just a client. For chip hardware, the impact is more indirect; for GPUaaS, it’s more like a business model stress test. ## Why hardware fell first: beyond fundamentals, overcrowded positions matter On the short-term trading level, the market isn’t only trading the fundamentals. J.P. Morgan’s trading desk debates split into two sides: one is whether Meta news signals a story shift in CSP capex and AI compute demand; the other is the impact from overcrowded positions, deleveraging, and profit-taking amplifying the drop. They lean toward the latter weighing more; whether fundamentals really shift depends on upcoming earnings calls. Background positions are not light. Major index rebalancing just happened, overall flows and leverage are starting high, in the past 4 weeks both long and short positions increased by +2 standard deviations; in the past 5 years, July is often when hedge funds deleverage, usually at -1 to -3 standard deviation change. Semiconductor and memory positions are near the 100th percentile. This explains why one piece of Meta news could hit the entire AI hardware chain. **Overcrowded trades meeting the narrative that “computing power may not be scarce” results in selling first. On the day, software, crowded shorts, and China ADRs rose by over 1.4 standard deviations—also matching the short squeeze pattern during deleveraging.** For reversal signals, the market mainly watches: **whether Meta clarifies; whether overseas AI ARR accelerates; whether cloud capital expenditure continues to upgrade; whether Q2 results beat expectations. The timing is centered on July to August. Currently it’s more of an observation period than a period of consensus.** ## One tail risk: the higher the stock price, the harder it is to ignore rumors of equity financing If Meta’s stock price is pushed up by the “computing power monetization” narrative, it could actually increase the probability of equity financing rumors. The logic: below 17x 2027 EPS, Meta is unwilling to do dilutive financing; if this news and strong Q2 earnings push valuation above 20x, the market should not be surprised by possible equity offerings. This is not a main line in the three external frameworks, nor is it confirmed by the company. But it explains why Meta’s stock reaction may not be simple. Selling computing power can ease ROI anxiety, but equity financing rumors bring dilution concerns—both forces will affect trading. ## Three firms don’t price Meta as a “computing power sales company” UBS maintains a Buy rating on Meta, target price $865, valuation based on diluted GAAP EPS of $33.26 for full year up to Q1 2028, with a 26x P/E; since the company hasn’t confirmed potential sales of computing power, forecast is not revised for now. Morgan Stanley maintains Meta as Overweight and Top Pick, target price $775. The base scenario implies about 23x 2027 P/E; the core remains ad revenue, Reels monetization, engagement uplift from AI, improved efficiency, and new product optionality. Bernstein maintains Meta Outperform, target price $850, and CoreWeave Underperform, target $67. This combo illustrates market divergence: Meta gets more optionality, CoreWeave faces more competition risk. But the risk doesn’t disappear. Downside factors include: ad cycle downturn, regulatory pressure, uncertain ROI for Reality Labs, execution errors in data center build-up leading to higher long-term capex intensity, etc. ~~~~~~~~~~~~~~~~~~~~~~~~ The above content is from [Chasing Wind Trading Desk](https://mp.weixin.qq.com/s/uua05g5qk-N2J7h91pyqxQ). For more detailed interpretation, including real-time insights and frontline research, please join [Chasing Wind Trading Desk Annual Membership](https://wallstreetcn.com/shop/item/1000309).  Risk Disclosure and Disclaimer: There are risks in the market; investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their specific situations. Investing based on this article is at their own risk.