$87.2B goes to training compute vs $15.3B to data — is VC equity the wrong capital for data?

abhiadesai · x · 2026-11-19

The quoted post is bullish on data markets: soon companies will buy data and fine-tune their own models — and eventually it won't even be the companies, but agents autonomously buying data and fine-tuning models. Abhi Desai adds 2026 spending structure: $87.2B in training compute (hyper-financed) vs $15.3B in data and related services (not financed at all), asking whether venture equity is the right capital type for data businesses or whether founders deserve less dilutive, better-aligned capital.

Original post →