Latest / HotTea: The AI Morning Briefing / The cash behind Meta's AI buildout
Transcript
- Sections 00:00 What we're covering today 00:27 1. Meta spent $34.9 billion on capital projects, more than it brought in from operations 01:44 2. Nvidia says its AI-factory units can deploy in as little as 14 days once sites are prepared 02:32 3. Thousands of AI agents turn mouse-brain data into research leads 03:24 4. China's new policy gives companies more say over research priorities 04:16 5. Cloudflare adds browser-agent tools to Commerce.com stores 05:10 Visit Hot Tea Disclosure Narration uses an AI-generated voice. Transcript Welcome to Hot Tea for Thursday, October twenty-second, twenty twenty-six. This briefing uses an A I-generated voice. the chipmaker says its factory-built systems can shorten deployment once a site is ready. We'll also cover A I agents examining mouse-brain data, China's research policy, and Cloudflare's browser tools for shopping carts. Shoppers still make the final checkout click themselves. Meta spent $34.9 billion on capital projects, more than it brought in from operations. Meta reported $64.60 billion in third-quarter revenue on October 21, up 26% from a year earlier. It spent $34.91 billion on capital projects, including finance-lease principal. That was more than its $32.79 billion in operating cash flow, leaving non-GAAP free cash flow at negative $2.12 billion. Meta narrowed its full-year capital-spending forecast to $135 billion to $145 billion. The previous range started at $130 billion. The forecast covers company capital spending, but the release does not identify every dollar as an A I expense. Operating income fell 37%, but the comparison includes a $9.89 billion charge for Meta's August agreement with 52 attorneys general. Meta says operating income would have been $22.88 billion without that charge. The release does not establish how much of the change came from A I spending. In Meta's next report, check whether operating cash flow covers capital spending. Watch for revenue or savings that Meta ties to specific A I services. Keep the one-time legal charge apart from recurring costs. the chipmaker says its A I-factory units can deploy in as little as 14 days once sites are prepared. It introduced DSX POD on October 21, factory-integrated units built around its Vera Rubin systems. Nvidia says offsite integration can cut onsite deployment from six months to as little as 14 days. Partners worldwide plan deliveries for early 2027. That timeline begins only after power, cooling, and networking are ready. Factory assembly could move work offsite. The announcement supplies no independent field results, and it does not show that permits or grid connections can be completed in two weeks. The first delivered units will matter. So will the measured time from delivery to useful service, alongside how long site preparation took. Thousands of A I agents turn mouse-brain data into research leads. OpenAI, PerturbAI, and Scripps Research said on October 21 that distributed agents analyzed existing data from millions of mouse brain cells in 32 minutes. The agents examined individual gene disruptions and linked the results to scientific literature. In blinded comparisons, experts preferred their reports over single-agent reports 75% of the time. The output is still a set of biological explanations and hypotheses that need experiments. The team's comparison with human analysis is an estimate, and the work appears in a preprint. Neither speed nor expert preference shows a treatment benefit. The accompanying repository provides code and research reports for inspection. Independent reproduction and laboratory tests will show which proposed mechanisms hold up beyond a plausible written explanation. China's new policy gives companies more say over research priorities. An October 21 opinion from the general offices of the Chinese Communist Party and State Council calls for closer links between research and industry. It supports wider A I adoption and a stronger role for enterprises in setting research priorities. It also supports measures that let smaller firms use some research results before paying. The policy aims to link research funding with industrial use. Whether it changes anything depends on program rules, budgets, and implementation. Publishing the opinion does not show higher productivity or completed commercial adoption. It also provides no single new A I spending total. Ministry and local plans may show which firms qualify, how funding works, and what milestones they set. The official explanatory document distinguishes proposed mechanisms from results. Cloudflare adds browser-agent tools to Commerce.com stores. Cloudflare announced a WebMCP rollout across Commerce.com storefronts on its network on October 21. That network covers almost 100,000 domains. The tools let shoppers search for products, build carts, and open a checkout link in the browser session. Merchants can opt out. That domain count shows potential reach, not completed purchases or demand for buying agents. Cloudflare's flow leaves the final checkout click to the shopper. WebMCP remains an experimental standard with limited browser support. Cloudflare sells the infrastructure behind this rollout, and its announcement offers no independent purchase or revenue results. Watch successful checkout rates, merchant participation, and independently measured failures. A product-search demonstration still does not answer the questions about purchases. That is the signal before the noise. This briefing was produced from Hot Tea's verified daily edition. For the complete briefing and every source link, visit Hot Tea dot A I.