Meta cooling systems and Nvidia outlook frame AI infrastructure constraints

Key Developments

Meta makes cooling architecture part of the AI-capacity story

Meta published an August 27 explainer on closed-loop liquid cooling, saying newer AI hardware has made air cooling less efficient for GPU servers and that the majority of its newest AI-optimized data centers use closed-loop liquid cooling (Meta Newsroom). The company described the system as a sealed water-and-glycol loop that moves heat away from server racks, passes through heat exchangers, and returns the cooled liquid to the hardware instead of expelling it from the facility (Meta Newsroom). Meta said the coolant can be used for up to a decade without replacement, and that a typical AI-optimized data center using closed-loop liquid cooling with dry coolers uses less water annually than a couple of full-service restaurants (Meta Newsroom). The same post said direct-to-chip liquid cooling can avoid nearly doubling the server tray size required for equivalent air cooling, allowing more GPUs in the same rack footprint (Meta Newsroom). Meta also said a reinforcement-learning pilot at one data center reduced air-cooling supply-fan energy consumption by an average of 20% and water usage by 4% across weather conditions (Meta Newsroom).

The read-through is that AI infrastructure competition is increasingly moving into facility engineering rather than only accelerator selection. Closed-loop liquid cooling does not eliminate power constraints, but it changes the usable density of a data-center shell and gives Meta a water-intensity argument at a moment when AI campuses face more local scrutiny. The reinforcement-learning detail matters because it links Meta’s AI software stack back into its physical operations: if simulator-trained controls reduce cooling overhead without direct live-site experimentation, infrastructure efficiency becomes an operating feedback loop rather than a one-time design choice.

What to watch: Track whether Meta ties closed-loop cooling and reinforcement-learning controls to future AI-campus approvals, Open Compute Project designs, or data-center water disclosures; the next useful signal is whether the 20% fan-energy and 4% water-use pilot metrics appear across a broader fleet (Meta Newsroom).

Nvidia’s post-results setup shifts attention to 2028 demand visibility

CNBC’s August 27 Morning Squawk said Nvidia reported a more than 100% year-over-year quarterly revenue increase and a better-than-expected revenue forecast after its latest results (CNBC). CNBC also reported that CFO Colette Kress told analysts Nvidia expects 70% revenue growth for fiscal 2028, compared with analysts’ 44% projection, even with supply constraints (CNBC). The same CNBC item said Nvidia disclosed that Amazon Web Services would purchase 2 million graphics processors (CNBC). CNBC quoted CEO Jensen Huang responding to criticism of Nvidia’s financing role in the AI ecosystem by saying critics were “missing a very big point” and that startups’ large funding needs were “really the nature of AI” (CNBC).

Figure 1 — Nvidia CFO Colette Kress told analysts the company expects 70% revenue growth for fiscal 2028, well above analysts’ 44% projection, even with supply constraints. Source: (CNBC).

The analytical point is that Nvidia’s debate is no longer simply whether the current data-center cycle is strong; it is whether customers can keep converting capacity plans into funded deployments through fiscal 2028. The 2 million-GPU AWS disclosure gives the demand side a named hyperscaler anchor, while the financing quote shows why Nvidia’s ecosystem role is receiving more scrutiny as compute buyers, model developers and infrastructure financiers become intertwined. The constraint to watch is supply and capital formation: if the 70% fiscal-2028 growth framework depends on accelerated GPU availability and funded AI start-ups, then order quality matters as much as reported demand.

What to watch: Nvidia’s September 10 Goldman Sachs Communacopia + Technology Conference presentation is the next scheduled investor-facing venue, and Nvidia said the webcast replay will remain available for 90 days (NVIDIA Newsroom).

Meta extends fraud-ad verification in Poland after scam-report declines

Meta said on August 28 that it is rolling out an enhanced AI system in Poland to reduce impersonation scams and that Poland is in the first group of European countries receiving the new anti-scam system (Meta Newsroom). The company said existing AI systems reduced user reports involving commonly misused celebrity likenesses by more than 80% (Meta Newsroom). Meta also said it will require identity verification from 100% of financial-services advertisers targeting Poland, with implementation expected to finish in the coming weeks (Meta Newsroom). The policy is part of Meta’s stated target for 90% of global ad revenue to come from verified advertisers by the end of 2026, versus 70% in 2025 (Meta Newsroom). Meta said scam-ad reports in Poland fell 83% from July 2024 to June 2026, and that it removed 137,000 scam ads in Poland from July 2025 through June 2026, with more than 88% removed before any user report (Meta Newsroom).

The operating implication is that ad integrity is being pulled into country-level compliance and monetization architecture. Requiring all financial-services advertisers in Poland to verify identity makes fraud prevention more measurable, but it also adds friction to a high-risk ad category that platforms have historically scaled through self-service tools. The 90%-of-revenue target is the key business constraint: Meta is not only reducing bad ads after detection; it is trying to increase the verified share of its global ad base before fraud campaigns migrate across markets and creative formats.

What to watch: The important follow-up is whether Meta expands the 100% financial-advertiser verification rule beyond Poland and whether future transparency reports separate prevention-before-report metrics from post-report removals (Meta Newsroom).

Google Flow and Apple TV pricing show consumer AI and services moving on different clocks

Google said on August 27 that Gemini Omni 1.1 Flash updates are rolling out in Google Flow, adding start-and-end-frame controls, 1080p or 4K exports, faster 360p concept drafts, and 720p downloads after upscaling (Google Keyword). Google said the capabilities are available starting that day in Google Flow (Google Keyword). Separately, TechCrunch reported on August 28 that Apple TV subscriptions are rising to $14.99 per month from $12.99 per month, annual pricing is moving to $119 from $99, and Apple One is increasing to $21.95 per month from $19.95 (TechCrunch). TechCrunch described the Apple TV change as the fourth price increase in four years and noted streaming-price increases by Netflix in March and Peacock earlier in August (TechCrunch).

The common thread is that consumer technology platforms are monetizing engagement through two very different levers. Google is adding granular creative controls to make generative video more usable for production workflows, where the question is whether AI tools can move beyond novelty into repeat creative work. Apple is testing services pricing power in streaming and bundles, where the question is whether content, device integration and cross-service packaging can offset subscription fatigue. The second-order issue is margin mix: AI creative products may consume infrastructure before monetization is clear, while mature media subscriptions can raise price but risk churn if perceived value does not rise with it.

What to watch: For Google, the signal is whether Flow usage shifts toward higher-resolution export workflows rather than low-resolution experimentation; for Apple, the signal is whether the September 9 iPhone event and broader services bundle messaging absorb the $14.99 monthly Apple TV price point without forcing heavier promotional discounting (Google Keyword) (TechCrunch).

This is an AI Briefing — AI-generated analysis published under TLCapital.AI. It is not personal research or positions, and it is not investment advice. Figures are sourced to primary filings with dates noted throughout. Do your own diligence.