18/07/26 Investor’s Daily Brief - AI Demand Is Booming, Yet Moats Are Shrinking Usage is growing, but pricing power is slipping and capital is getting more selective. Market snapshot: S&P 500: 7,475.69, -1.0% Friday close Nasdaq: 25,520.24, -1.4% Friday close WTI: $82.49, +4.48% Friday settlement Bitcoin: $63,934, +1.27% over 24 hours Saturday Inside today: • Cheap models weaken the chip premium • Treasury plumbing raises duration risk • $IBM exposes legacy AI vulnerability • $SPCX validates secure compute demand • Policy speed becomes a product Friday’s selloff was not a rejection of AI. It was a repricing of where AI profits will accrue—and how much investors should pay for distant earnings when capital is becoming less forgiving. CHEAPER MODELS, THINNER MARGINS Moonshot AI’s Kimi K3 intensified the semiconductor unwind, with the PHLX Semiconductor Index down 10% for the week. The problem is not falling AI usage. It is improving cost-performance: open models can expand adoption while compressing pricing power for chips and proprietary-model vendors. TREASURY CALM LOOKS CONDITIONAL The U.S. must absorb roughly $2 trillion of additional annual Treasury issuance as ownership shifts toward more price-sensitive, leveraged buyers. That makes long-duration assets more vulnerable to weak auctions, tighter repo funding or another jump in long-end yields. IBM LOSES THE BENEFIT OF THE DOUBT IBM’s 25% plunge after a weak outlook showed that “AI exposure” is no protection when the reported numbers fail. Legacy software and consulting moats are being tested by cheaper modernization tools and faster-moving competitors. COMPUTE DEMAND IS STILL REAL SpaceX’s talks to provide the Pentagon with billions of dollars of data-center capacity show that infrastructure demand has not disappeared. But more capacity and lower-priced entrants could shift profits from model scarcity toward secure compute, power and utilization. INFORMATION SPEED GETS MONETIZED Trump Media’s proposed Truth API would sell faster access to market-moving posts for as much as $100,000 a month. Policy latency is becoming proprietary market data. Bottom line: AI demand remains strong, but the easy-margin phase is fading. The cleaner exposure is no longer “own AI.” It is owning scarce infrastructure, resilient cash flows and distribution advantages that survive cheaper models and pricier capital.