Elon's Anthropic Deal, The Next AI Monopoly?, "FDA for AI" Panic, Trading the AI Boom

Elon Musk just leased all of Colossus 1 — SpaceX's H100-based data center with over 220,000 Nvidia GPUs and 300 megawatts of power — to Anthropic in a deal that addresses both companies' most urgent constraints. Anthropic was compute-starved; Claude users were hitting rate limits. XAI was cash-flow negative, burning capital on massive training clusters before revenue materialized. The deal solves both problems overnight.

Brad Gerstner estimates this will generate $4–5 billion in incremental SpaceX revenue this year, on top of mid-$20 billion analyst estimates. That's transformational for the IPO roadshow: SpaceX now has a five-layer cake — launch, Starlink connectivity, hyperscaler compute, space data centers, and applications (XAI, Grok). The biggest question hanging over the valuation was whether XAI could deliver revenue before capex spiraled. Now that pressure evaporates. Elon subsidizes Grok development while monetizing infrastructure at scale, competing directly with AWS, Azure, and GCP.

Chamath Palihapitiya called the deal a week ago, noting Anthropic's supply constraints and Elon's foresight in securing power before anyone else. The deal cements Elon's role as kingmaker in AI infrastructure. And it's only the beginning: XAI is already trading models at Colossus 2 (1.2 gigawatts of Blackwell GPUs), and cursor just signed on. Elon built AWS for the AI era — call it EWS.

The New York Times reported that the Trump White House is considering an executive order to create an AI working group that would examine «a review process for new AI models.» Kevin Hassett, director of the National Economic Council, invoked the FDA analogy on Fox Business, saying future AIs should be «proven safe» before release. Silicon Valley reacted viscerally: the phrase «FDA for AI» conjures pre-approval regimes that would kill innovation and entrench incumbents.

But according to Brad Gerstner and David Sacks, who both spoke to Hassett after the clip aired, the FDA analogy was poorly chosen. The administration does not support an approval regime. The real issue is cyber capability: Anthropic's Mythos model and OpenAI's equivalent can automate sophisticated cyberattacks. Within three to six months, all frontier labs — and Chinese models — will have these capabilities. The White House wants labs to coordinate, share models with government for hardening, and apply KYC (know your customer) during preview periods. That's not pre-approval; it's coordination.

Still, the incident revealed how close the industry came to regulatory capture. Bernie Sanders praised the idea of an FDA for AI. Think tanks and activist groups seized the moment to push for permanent oversight infrastructure. Sacks pointed out that both Anthropic and OpenAI acted responsibly by not releasing Mythos publicly — proving that pre-approval wasn't needed. The real work now is hardening systems, getting cyber-defense tools into the hands of CrowdStrike, Palo Alto Networks, and cutting-edge startups, and applying basic KYC before granting API access to frontier models.