AI's unusual bet: buying compute for years, selling intelligence by the token
vaibhavbetter · x · 2026-10-28
Ramaswmy Sridhar published a 16-part thread analyzing the AI industry's unusual bet: committing to multi-year compute purchases while selling intelligence priced per token, importing risk structures from banking, commodity trading, airlines, utilities and cloud into AI. Each sector explains a different facet of the risk.
Vaibhav amplified it with the core question: "whether demand keeps outgrowing committed cost" — whether revenue growth can truly cover years of locked-in compute commitments, or whether the industry is repeating the classic overcapacity cycle.
More from Venture
- Personal AI assistants cost $3,000-$7,000 per user per year, five startup theses emerge — FinanceYF5 · 2026-10-28
- SaaS analytics tool sells for $4.5K at 0.6x multiple in 97 days on TrustMRR — marclou · 2026-10-28
- Norn Group raises $9.1M of $15M to build the datasets AI for aging will need — juanbenet · 2026-10-28
- Hypothetical $2T Anthropic IPO framed as channeling $100B+ a year into the real economy — sanjaykalra · 2026-10-28
- Founders Fund II turned $220M into $37.4B, a 183.5x return led by SpaceX — FinanceYF5 · 2026-10-28
- Chinese AI SEO community Web.Cafe draws 499 signups for a single event — gefei55 · 2026-10-28