Will Meta's plan to sell computing power lead to its withdrawal from the development of cutting-edge models? What are the key events and signals that can reverse market expectations and sentiment in the future? Xiao Zha admitted at yesterday's Meta internal meeting that the progress of AI agent development in the past four months has been slower than expected, and the restructuring effect has not met expectations. However, the company still maintains an optimistic attitude towards AI returns in the next 3-6 months and plans to continue to promote related investments and adjustments. Simply put, it means acknowledging that the progress is below expectations, while being optimistic about the future without reducing investment, and still proceeding according to plan. On Tuesday, Bloomberg revealed that Meta plans to sell computing power. There are several concerns in the market: 1) Is there an excess of computing power? 2) Is Meta selling computing power for more efficient monetization or is its AI progress falling short of expectations? More concerned is whether Meta will withdraw from the development of cutting-edge models? 3) Will Meta reduce capital expenditures?. So last night, Xiao Zha's response was equivalent to answering the last two questions. Although the progress did not meet expectations, he still invested as planned (without reducing expenses in the near future). Of course, Xiao Zha's speech was actually based on the premise of expecting AI returns to improve in the next 3-6 months. The implicit meaning is that AI returns have not improved in the next six months and may need to be adjusted again. Let's talk about a few questions: 1. Will Meta abandon cutting-edge AI models? From a personal perspective, Meta's AI is not an independent ToB software, but a system capability deeply embedded in advertising, recommendations, Reels, Instagram, WhatsApp, Messenger, AI glasses, and future agents. It is not just a question and answer robot company, but a large-scale consumer Internet distribution network. For Meta, model quality, inference cost, latency, personalization, content understanding, and advertising conversion rate are part of the same value chain and cannot be left to third-party APIs for the long term, meaning they cannot be held back by others. The real value and difference of Meta in the future lies in its "proprietary model+proprietary user scenarios+proprietary advertising/social data loop+proprietary inference infrastructure" Of course, the problem is that the progress of AI is lower than expected, but as long as there is still hope, I personally think it's difficult for Xiao Zha to give up so quickly. 2. Will Mete do cloud computing? It's also a natural progression, after all, Meta has invested so much in computing power, and the significance of doing cloud computing is: 1) Monetization of idle capacity: Jefferies research data shows that Meta's current internal infrastructure utilization rate is about 65%, with 35% of idle capacity remaining. Bank of America pointed out that the demand for enterprise AI can help absorb any excess computing power if the speed of Meta construction exceeds the speed of internal use. 2) Improving Free Cash Flow: Selling idle computing power "helps monetize its massive AI infrastructure investment and improve free cash flow expectations. 3) Reducing tail risks of overinvestment: If consumer facing AI products take longer to scale, enterprise AI demand can provide a more stable source of income while reducing pressure on profit margins. Enterprise level AI may become a significant new source of revenue for Meta, helping it monetize its growing computing power. Even if Meta only gains a small share of the enterprise AI market, given its deployed infrastructure scale, it can still generate meaningful revenue. This provides Meta with revenue diversification beyond advertising - although advertising is profitable, it is highly sensitive to economic cycles. Being able to optimize the investment return rate of meta, of course, from a more practical perspective, is also a form of financial discipline and utilization management. 3. When will the market concerns brought by Meta dissipate? This may be what everyone is most concerned about. Meta's sale of computing power is just a catalyst and triggering factor, essentially a clearing of ultra-high crowding and high leverage trading. In mid June, it was discussed that this is the real risk in the market. https://(x.com)/qinbufark/status/2072513357077524671? s=46&t=k6rimWsEbo2D2tXolYcM-A。 Of course, looking deeper, the real concerns of the market are threefold: First, does the model company burn money to buy numeracy, but its income cannot keep up? Secondly, is CapEx excessively pre positioned by cloud providers, leading to a collapse in computing power prices in the future? Thirdly, is the AI Infra chain building capacity for future non-existent demands? Of course, everyone is still concerned about whether there is enough revenue to undertake the AI infrastructure investment with so much money? If there is not enough income to undertake, it will eventually lead to excess computing power and Capex reduction. To reverse this situation and restore market confidence. The hardest evidence is not the founder speaking out, but the income. if OpenAI、Anthropic、Google Gemini、Microsoft Copilot/Azure AI、AWS Bedrock、 The revenue of enterprise agents continues to exceed expectations, and the market will re understand the entire chain: It's not an excess of computing power, but rather AI application revenue is catching up with computing power supply. The first weight that can truly reverse market expectations in the medium term is the sustained high demand for AI, with the core being the sustained high performance of large model ARR and super large cloud vendors. You can break this down into four linked indicators instead of one indicator: 1) Continuous improvement of the large model ARR; 2) Hyperscale cloud revenue continues to accelerate; 3) Backlog/RPO continues to expand; 4) The price and gross profit of AI cloud will not collapse. If all four indicators are met simultaneously, Meta's sale of cloud computing power will be reinterpreted as "monetization of scarce assets"; If two or more of these four begin to weaken, Meta's sale of cloud computing power will be interpreted by the market as "the first card of AI infrastructure collapse". Figure 1 shows the weight ranking of key events and signals that can reverse market sentiment and expectations from a personal perspective: the highest weight is not clarified by Meta itself, but continuously verified by AI demand side revenue, orders, prices, and gross profit; Meta's disclosure of the proportion of external sales capacity and the progress of self-developed models are necessary but secondary signals. At the beginning of June, here is https://(x.com)/qinbafrank/status/2063103873867452886? S=46&t=k6rimWSEbo2D2TXolYcM-A has discussed that only when the fundamentals (AI commercialization growth rate) really have problems, adjustments need to be made before waiting for more reversal signals. Because fundamental problems can reset all logic, rather than completely starting over, but the market impact is significant. If there is no progress in the industry for a long time, the adjustment range can easily evolve from a small level to a specific level. Then the market waits for the industry to provide new evidence that it has returned to rapid growth, with scale and growth exceeding expectations, before confidence can return. This content is sponsored by @ BITstocks_CN. Buy BIT-16000+US stocks and ETFs on the US stock market, hold real positions, and enjoy dividends.