TLDRocket
4 August 2026
The infrastructure backbone of AI is buckling under its own weight. Texas, which has aggressively courted data centers with tax breaks and deregulation, just hit pause. Governor Greg Abbott declared a moratorium on new data center grid connections, forcing ERCOT to audit existing projects as the interconnection queue swelled to 474 gigawatts—more than five times the state's peak demand and nearly all data centers. The move signals that even hospitable jurisdictions have limits. Meanwhile, the hardware vendors powering this expansion are consolidating control. SpaceX has now spent $624 million on Tesla Megapacks for xAI data centers, while Anthropic locked in $10 billion of compute from Volta, a Norway facility built with Bitdeer and Nvidia's Vera Rubin chips. AMD's data center revenue doubled year-over-year to $6.7 billion, now 58 percent of the company's total revenue. These aren't marginal businesses anymore—they're existential. OpenAI, meanwhile, is shifting from training to inference economics. Its Astra model proved ten mathematical theorems at an estimated $2,000 in API tokens, quantifying frontier reasoning work for the first time. ChatGPT Work reached 10 million users in three weeks by becoming an agent that lives in the cloud, completing tasks across Slack and email. Microsoft, watching its employees rack up monthly AI bills in the thousands, introduced token budgets and made GPT-5.6 the default. The message is clear: infinite scaling has given way to managed consumption. What happens when demand outpaces both grid capacity and willingness to pay?
Read the full briefing →