Meta Compute, OpenAI Equity, and AI Startup Bets
Meta spent so much on AI that it may have accidentally become a cloud company. Smart move or capex panic? That's the kind of question Oscar and I like, because the answer tells you where the whole industry is actually heading.
Here's how I called the week, and the two business ideas we argued over at the end.
Signal or Noise
Meta is building a cloud business for its excess AI compute
When you buy so much compute that reselling it becomes a business line, the model race has quietly turned back into an infrastructure race. That's the real story of the week. Signal.
OpenAI reportedly offering 5% equity to a U.S. sovereign wealth fund
Follow the incentives here. A lab handing equity to the government isn't charity — it's buying alignment with the people who write the rules. Signal, and a little unsettling.
Claude Sonnet 5 ships everywhere
Good model, shipped across Claude, Claude Code, and the platform. But I call model releases noise until they change builder behavior, and this one didn't move my stack yet. Noise for now.
Together AI raising $800M for open-source infrastructure
Real money into the open-source plumbing layer. Signal — this is where a lot of the actual building happens.
Claude Science launches as a research workbench
Interesting, narrow, and not something most of us act on Monday. Noise for your standup.
Ship It or Skip It
Two ideas, no fence-sitting:
- A Meta Compute broker for AI teams — a layer that arbitrages spare capacity across providers so teams stop overpaying for GPUs. If compute is becoming a commodity, someone builds the exchange. I'd ship it.
- A back-office agent for small law and accounting firms — narrow, boring, and exactly where vertical AI pays. Real workflows, real willingness to pay. Also ship.
The thread across both: the model race is becoming an infrastructure race again, and the money is moving to whoever owns the layer underneath. Watch or listen to the full episode below.
