Latest / HotTea: The AI Morning Briefing / AI's bill came due in public.
Transcript
- Sections 00:00 What we're covering today 00:20 1. Microsoft showed the revenue. Meta showed the burn 01:58 2. OpenAI offered 100,000 researchers the tool, not the system 02:58 3. Central banks now have to price an AI boom they cannot yet measure 04:13 4. A defense partnership put swarm coordination into the test lane 05:13 5. Memory makers printed records. Their stocks still lost the argument Transcript What we're covering today for Thursday, July 30, 2026. Microsoft showed how cloud demand can carry the spend, Meta showed how fast costs can outrun earnings, central bankers put AI inside the policy problem, OpenAI widened researcher access without opening its systems, defense autonomy moved into a swarm test, and record memory profits still failed to calm the market. Microsoft showed the revenue. Meta showed the burn. Microsoft reported $90.0 billion in quarterly revenue, up 18% year over year, and $35.8 billion in net income, up 31%. Azure and other cloud-services revenue grew 43%, and Microsoft said annual Azure revenue exceeded $100 billion for the first time. Axios reported that Microsoft's quarterly capital expenditures rose 70% to $41 billion. Meta reported $60.8 billion in revenue, up 28%, while costs and expenses rose 55% to $42.0 billion. Meta's net income fell 14% to $15.8 billion, free cash flow was $784 million, and quarterly capital expenditures including finance-lease principal were $31.1 billion. Meta narrowed its full-year capital-spending range to $130 billion to $145 billion by raising the low end. The pressure: This is not a clean head-to-head comparison. Microsoft and Meta have different businesses, accounting mixes, customer bases, and investment cycles. Meta's quarter also included $2.4 billion in legal charges and $1.18 billion in severance expenses, while Microsoft's GAAP income included investment effects. One quarter cannot establish the lifetime return on either company's infrastructure. What to watch: Microsoft's forecast for more than $50 billion of current-quarter capital spending, the share of that spend tied to short-lived GPUs and CPUs, Meta's third-quarter revenue range of $61 billion to $64 billion, Meta's post-layoff cost base, and whether either company discloses product-level returns rather than broad AI attribution. OpenAI offered 100,000 researchers the tool, not the system. Axios reported that OpenAI is launching a program to give 100,000 academic researchers free access to its most advanced hosted models through 2027. The first 10,000 participants are expected to begin receiving access this summer, and OpenAI said the broader effort is part of more than $250 million it plans to invest in external scientific research and discovery through 2027. The pressure: The program can widen access to expensive frontier inference, but the reported terms do not provide model weights or training data. OpenAI also told Axios that participant data would not be used for training and that business-grade privacy protections would apply; those remain company commitments whose implementation and research value need independent evidence. What to watch: Selection criteria, geographic and institutional distribution, published work from the first cohort, whether negative or safety findings can be released without restriction, and whether access produces reproducible results that researchers outside the program can audit. Central banks now have to price an AI boom they cannot yet measure. A new Bank for International Settlements bulletin said the AI boom is driving a large, increasingly debt-financed investment surge while its productivity payoff remains uncertain and uneven. The authors said AI affects demand and supply at the same time, blurring cyclical signals and increasing the risk of monetary-policy miscalibration. On July 29, the Federal Reserve held its target range at 3.5% to 3.75% by a 9-3 vote; the three dissents favored a quarter-point increase as inflation remained above the 2% goal. The pressure: AI can create near-term demand through data-center investment and market wealth while producing longer-term supply gains through productivity. Policymakers cannot safely assume the timing or size of either effect. The BIS bulletin is an analytical assessment, not an observed productivity verdict, and the Fed statement did not attribute its decision specifically to AI. What to watch: The Fed's year-end task-force findings on AI, productivity, labor, inflation measurement, and data collection; revisions to business-investment data; debt financing for AI infrastructure; labor-market separations in exposed occupations; and evidence that productivity gains are spreading beyond a small group of firms. A defense partnership put swarm coordination into the test lane. Axios reported that AeroVironment and Applied Intuition are collaborating on the Mayhem 10 launched-effect system and Applied's Acuity ISR/Strike software, and that they recently tested the drone's ability to synchronize and swarm. AeroVironment describes Mayhem 10 as an autonomous, multi-role system derived from its Switchblade family, while the reported software layer supports data sharing and coordinated behavior. The pressure: This is a company-reported test and partnership, not independent evidence of reliable battlefield performance. Public material does not establish failure rates, communications resilience, target-identification accuracy, rules of engagement, or the exact human-authorization boundary. No live operational, casualty, or targeting claim is included here. What to watch: Independent test results, the operator-approval model for any strike function, performance under jamming and degraded communications, audit logs for machine recommendations, acquisition contracts, and whether the Pentagon publishes test and evaluation criteria before fielding coordinated systems. Memory makers printed records. Their stocks still lost the argument. SK hynix reported record second-quarter results on July 29, and AP reported the following day that Samsung Electronics posted a record 89.5 trillion won operating profit for the April-June period, with nearly all of it coming from semiconductors. Yet AP said SK hynix shares fell more than 9% on Wednesday after its result missed higher market expectations, while Axios documented a broader pullback across semiconductor and memory names. The pressure: The contradiction is not weak current demand. It is that investors are testing how long elevated memory pricing, AI capital spending, and incumbent market power can survive new capacity and Chinese competition. Record profit is backward-looking evidence; planned fabs, customer concentration, and the cost of the next supply cycle determine whether it persists. What to watch: HBM contract pricing, SK hynix and Samsung capital-spending guidance, CXMT capacity and yields, customer concentration among hyperscalers, the pace of new fab construction, and whether rising supply closes the gap before AI-server demand slows. That is the signal before the noise. This briefing was produced from HotTea's verified daily edition and narrated with an AI-generated voice.