Meta's rental of computing power is similar to using existing resources to fund new ones, rather than ceasing the pursuit of high-end computing power.
Related
CITIC Securities: Significant fluctuations will not alter the AI supercycle; emphasize the importance of domestic computing power as a "Plan B".
According to Mars Finance, CITIC Securities points out that news of Meta's plan to lease out some of its computing power has once again triggered market concerns about computing power oversupply. This, coupled with concerns about cloud vendors' cash flow pressures, the continued rise in upstream prices, the slowdown in Capex growth, and overcrowding in AI transactions in recent months, is the main reason for the significant volatility in tech stocks. In the short term, Meta's move is primarily aimed at revitalizing its existing, outdated computing power assets. Considering its continued development of advanced models and investment in next-generation computing hardware, leasing out computing power is not contradictory to further increasing its investment in computing power. Furthermore, since computing power rental fees have continued to rise recently, concerns about computing power oversupply are unfounded. This round of tech stock adjustments is more of a deleveraging and rebalancing process in the recent global liquidity tightening environment, rather than a reversal of the AI industry trend. For the medium to long term, it is crucial to pay close attention to whether the next few months will see a similar breakthrough in AI capabilities as seen with OpenClaw and Coding Agent at the beginning of the year. In addition, it has been observed that after overseas AI assets entered a phase of high crowding, high correlation, and high volatility, international funds are beginning to seek differentiated sources of return. Domestic computing power with differentiated value, dubbed "Plan B," remains resilient and is expected to attract foreign investment. With the earnings season approaching, we recommend focusing on sub-sectors with high earnings certainty and reasonable valuations: In terms of growth prospects, we recommend domestic FAB (Featured Adhesives and Materials) and equipment sectors with positive narratives, as well as the optical communication sector with relatively low valuations; in the price increase chain, segments with high AI exposure and those that have already experienced price increases have a higher probability of realizing their earnings gains, such as memory and upstream PCB industries. (Cailian Press)
US computing power rental providers Nebius and CoreWeave turned positive in pre-market trading, shaking off the shadow of Meta's plan to build its own cloud business.
According to Mars Finance, on July 2nd, based on BIT (bit.com) market data, the pre-market share prices of US computing power leasing companies Nebius and CoreWeave turned positive, rising 3% and 2% respectively, after closing down 17% and 14% respectively yesterday. Yesterday's news that "Meta plans to build its own cloud business and sell surplus AI computing power" triggered a market-wide decline in computing power leasing companies.
Zuckerberg signals the commercialization of computing power, suggesting that Meta infrastructure may enter a dual-track model of "self-use + external leasing".
According to BlockBeats, on July 1st, Bloomberg reported that Mark Zuckerberg stated in a May shareholder call, "Almost every week, different companies come to us, either wanting us to build an API service or asking if we have computing power to sell to them, at a higher price than we paid. We haven't done that yet because we have our own need for that computing power," Zuckerberg said at the time. "But obviously, if we realize we've overbuilt, then that's an option, which partly gives us the confidence to invest in scaling up." In the rapidly evolving AI race, Zuckerberg has repeatedly stated that he believes the entire industry is facing a bottleneck in computing power, and that Meta should accumulate as much computing power as possible before deciding how to use it. Despite facing numerous complex challenges, Meta CEO Mark Zuckerberg has signaled to investors that he is willing to sell excess computing infrastructure and even launch so-called API services, allowing customers to pay based on AI usage—a business typically measured in "tokens" (i.e., the amount of data generated and consumed when a customer queries). Discussions surrounding Meta Platforms' potential commercialization of some of its AI computing power have raised concerns about a potential "oversupply" of computing power. However, analysts argue that this assessment is oversimplified. Meta currently does not have any significant surplus computing power available for sale. On one hand, it continues to expand its investment in AI infrastructure and has signed large-scale computing power agreements with companies like Crusoe. On the other hand, its existing H100/H200 resources are primarily used for internal recommendation systems and AI model training, and demand remains tight. Therefore, the so-called "selling computing power" is more of a future option for asset utilization than a current reality. Overall, it reflects a shift in AI computing power from a single scarcity to a multi-generational, tiered use and long-term capacity planning phase.
SemiAnalysis: Meta will accelerate, rather than slow down, its computing power procurement; it is in talks with Anthropic to build its own AI model service platform.
According to BlockBeats, on July 3rd, SemiAnalysis stated in its latest report that after news broke that Meta might become a new Neocloud, the market's initial reaction was to sell off computing power cloud companies like CoreWeave and Nebius, and to renewed concerns about "AI computing power oversupply." However, the firm's assessment is the opposite: this concern may be wrong. Meta's data center and computing power procurement will not slow down, but rather continue to accelerate. The article mentions that in the first half of this year alone, Meta has already signed contracts for over 5GW of capacity in the cloud services and managed data center sectors, and this does not include its accelerating self-built projects. SemiAnalysis states that Meta is in final negotiations with Anthropic, hoping to obtain access to Claude private instances. If this comes to fruition, the significance goes beyond "Meta buying more computing power"; it suggests that Meta may be building its own AI model service platform. This model is somewhat similar to AWS's Bedrock, Microsoft's Foundry, and Google's Vertex. Meta can initially use Claude internally, or it can package the model capabilities as a token-as-a-service to provide services externally in the future. In the short term, it might use its own models externally and Anthropic models internally; in the long term, Meta might even incorporate Anthropic and OpenAI models into its external service system. SemiAnalysis states that the underlying logic is that Meta has computing power, advertisers, social network distribution capabilities, and consumer-end entry points. If it can combine cutting-edge models, intelligent agents, and sales and marketing SaaS, it will not just be a company that buys GPUs, but will be moving towards the upper layers of AI application and model distribution.
HTX DeepThink: Meta's sale of AI computing power raises concerns, marking a new phase in the generational differentiation of computing power assets.
Chloe (@ChloeTalk1), a columnist for HTX DeepThink and a researcher at HTX Research, points out that Meta's recent sale of some AI computing power has sparked concerns about an oversupply of computing power, putting pressure on GPU cloud service providers and the semiconductor sector. However, from the perspective of the actual situation in the industry chain, this change cannot be simply interpreted as the peak of AI computing power demand. It is more likely that computing power assets are beginning to show generational differentiation: the previous generation of GPUs and non-core computing resources are gradually shifting to commercial leasing, while new-generation high-performance clusters such as Blackwell and Rubin remain relatively scarce.
Industry insiders: Computing power rental prices continued to rise this month; demand for training cards and inference cards may diverge.
According to Mars Finance, regarding Meta's plan to launch a cloud infrastructure service called "Meta Compute," which aims to sell AI computing power to external customers, an industry insider in the computing power leasing sector told the Science and Technology Innovation Board Daily reporter that Meta has previously invested heavily in AI but the results have been slow to materialize. Leasing computing power can not only recoup cash flow but also signifies that it is distancing itself from its previous generation of outdated hardware and its blind expansion strategy. He also revealed that its leasing server supplier mentioned that prices would continue to rise at the beginning of this month. He predicts that the demand for inference cards and training cards may diverge. As the scale and performance of open-source large-scale models continue to climb, some applications are shifting from self-developed fine-tuning to directly calling existing models, creating a significant "temperature difference" in the market between high-end new training cards and inference/older training cards. (Science and Technology Innovation Board Daily reporter Huang Xinyi)