Author | Chen Jia, Editor | Yun Peng Going head-to-head with the three major cloud giants.
According to a July 2 report from Zhidx, which cited people familiar with the matter via Bloomberg yesterday, Meta is developing a plan for a cloud infrastructure business to sell AI (Artificial Intelligence) computing power and AI model access to external customers, competing directly with the three major cloud giants: Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. Yesterday, Meta's stock price surged by over 10% at one point and closed up 8.81%. Meanwhile, AI computing cloud service providers CoreWeave and Nebius fell by 10.8% and 12.4%, respectively.
On July 1 local time, the Philadelphia Semiconductor Index closed down 6.27%. At the individual stock level, Micron Technology and Western Digital fell by over 10%; Applied Materials, Lam Research, and Intel dropped by more than 9%; ASML, TSMC (Taiwan Semiconductor Manufacturing Co., Ltd.), and AMD declined by over 6%. NVIDIA and Broadcom showed relative resilience, falling by approximately 1.25% and 2.23%, respectively. At the close on July 2, the A-share semiconductor and computing hardware sectors declined broadly. In the semiconductor equipment sector, NAURA Technology hit the daily limit down, while Piotech and SkyVerve fell by over 15%. In the memory chip sector, GigaDevice hit the daily limit down, and Shannon Microelectronics dropped by over 13%. In the computing hardware sector, Eoptolink fell by over 11%.
According to the Securities Times, the Wind Semiconductor Index fell by over 5% on the day. At the individual stock level, Puya Semiconductor dropped by over 12%; Sicote and Yoke Tech hit the daily limit down; Leon Micro and Galaxy Micro fell by over 9%; JCET, Montage Technology, and GigaDevice declined by more than 7%. In addition, Cambricon and AMEC both fell by over 5%. According to TechCrunch, the cloud infrastructure business falls under Meta's internal initiative named "Meta Compute," which is co-led by Santosh Janardhan, Head of Infrastructure at Meta; Daniel Gross, Co-Head of Meta's Superintelligence Lab; and Dina Powell McCormick, President of Meta.
Relevant business planning is still advancing, and the strategic plan remains subject to adjustment. Behind this move lies Meta's massive investment in AI infrastructure and the delayed direct returns. Meta expects to spend up to $145 billion (approximately RMB 984.6 billion) on AI infrastructure this year, but its self-developed model Muse Spark, released in April, has not yet been opened to developers. Major tech giants, concerned about excessive AI investments, are seeking to recoup their massive AI investment costs.
Meta is no exception. To support its AI development vision, it continues to make massive investments in high-cost data centers and supporting infrastructure. Now, the company plans to establish a new business segment to sell idle computing power to external customers, revitalizing data center resources and opening up short-term profit channels. Meta is also not the first company to sell computing power to recoup costs. In May this year, SpaceX leased the computing power of its Memphis data center to companies including Anthropic and Google. For the Anthropic deal alone, the monthly charge is as high as $1.25 billion (approximately RMB 8.5 billion).
01.Two Business Routes Advancing in Parallel - Hosting AI Models and Selling Raw Computing Power The first route Meta is considering is to allow developers to access AI models (including Muse Spark) hosted on Meta's infrastructure and pay for the computing power required to run these models. This is similar to AWS's Bedrock service, which allows developers to invoke AI models from various companies. The other route is to emulate neoclouds like CoreWeave and directly sell raw computing power. This strategic move is not a spur-of-the-moment decision.
At Meta's shareholder meeting held in May, Mark Zuckerberg, Founder and CEO of Meta, stated that venturing into the cloud computing business is "under clear consideration," and revealed that companies reach out almost every week hoping to purchase access to Meta's AI models or idle computing power. What is driving Meta to shift towards selling computing power and models externally is the reality of slow commercialization progress of its self-developed AI. Meta launched Muse Spark in April this year, its first AI model built by a heavily funded team, but it has not yet been opened for developer use.
The Wall Street Journal previously reported that there is no clear timeline for the model's external launch. Unlike Google and OpenAI, external market demand for Meta's self-developed AI models and supporting services is relatively weak. Meta's financial reports do not separately disclose the revenue of Meta AI and its open-source weight model series, Llama, and management's public statements mostly focus on internal AI applications. This means that Meta's AI business has not yet formed an independent revenue segment of significant scale. Selling idle computing power and model access to external customers can precisely bypass this commercialization challenge, turning massive infrastructure investments into tangible cash flow first.
02.Meta Stock Surges Over 10% in a Single Day - Two Computing Power Leasing Companies Hit Hard The capital market's reaction was immediate. As of Tuesday, Meta’s stock was down nearly 15% year-to-date, underperforming the S&P 500 and trailing AI peers. Following the news of its cloud business expansion, Meta's stock price surged by over 10% in a single day, significantly alleviating the pressure.
Neoclouds, however, suffered heavy blows, with CoreWeave and Nebius falling by 10.8% and 12.4% on the day, respectively. Meta is a key customer for both companies. On April 9 this year, CoreWeave announced an expanded partnership with Meta, signing a new agreement worth approximately $21 billion (approximately RMB 142.6 billion) to provide AI cloud computing power for Meta. This new agreement is built upon a $14.2 billion (approximately RMB 96.4 billion) contract signed by both parties in September 2025. Adding the two contracts together, the total cooperation amount between CoreWeave and Meta has reached approximately $35 billion (approximately RMB 237.7 billion).
Nebius and Meta signed a five-year agreement with a total value of up to $27 billion (approximately RMB 183.3 billion) on March 16 this year. Under this agreement, Nebius will provide Meta with approximately $12 billion (approximately RMB 81.5 billion) worth of dedicated computing infrastructure. In addition, Meta has committed to purchasing up to an additional $15 billion (approximately RMB 101.9 billion) of Nebius computing capacity over five years. Gil Luria, Managing Director at investment bank D.A. Davidson, pointed out that Meta's entry into the computing power market is more likely to impact neoclouds rather than major cloud giants. Companies like CoreWeave and Nebius rely on Meta for growth, and Meta may no longer need them. Reuters, citing analysts, noted that competing directly with Amazon, Microsoft, and Alphabet can help Meta capture the strong demand for AI services from enterprise customers while reducing its reliance on the advertising market.
03.Conclusion: A Hundred-Billion-Dollar Gamble for Returns - Data Centers Become Key Assets in the AI Race Meta's follow-up in cloud business expansion sends a signal: the ultimate winner in the AI track may not be the companies with the best models and services, but the manufacturers holding data center resources. The premise for the above judgment to hold true is that computing power demand remains strong and the value of data center assets remains stable. Some skeptics warn that the global rush among enterprises to expand AI infrastructure is breeding a bubble, and the industry is highly dependent on chips that depreciate extremely fast. Others question whether AI companies can generate sufficient end-user revenue to support industry investments at the trillion-dollar level. Such concerns have not stopped Meta from continuously increasing its investment in AI computing infrastructure.
Meta expects to spend up to $145 billion on AI infrastructure this year, accounting for a significant proportion of the total AI investment of over $700 billion (approximately RMB 47.5 trillion) by major tech companies this year. As of the end of the first quarter this year, Meta has committed to investing $182.9 billion (approximately RMB 12.4 trillion) in building AI infrastructure over the next few years, including large-scale projects underway in Louisiana and Ohio. Zuckerberg stated that the Ohio project is as large as Manhattan and is expected to be put into operation within this year.
Sources: Bloomberg, TechCrunch, Reuters