saved
Gavin Baker on why everyone can win in AI
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
In a three-post X thread, investor Gavin Baker argues for a hybrid, multi-model AI future where open-weight models carry most tokens and fragmentation at the model layer helps infrastructure, applications, and ordinary people. He says lower model margins raise compute demand, that Anthropic's commitments are manageable, and that external financing may be near its peak.
ideas
- Fragmentation beats oligopoly. A two-lab model layer would squeeze suppliers and applications; open-weight growth since May makes competition stronger.
- Cheap tokens raise compute demand. Every model token costs similar compute, so lower markups on open-weight tokens send relatively more revenue to clouds.
- Enterprises will own weights. Post-trained open models behind a router with frontier models can deliver frontier performance at lower cost.
- Commitments are smaller than they sound. Baker puts Anthropic's leaked contracts at about $31 billion a year of cloud spend plus Broadcom spending financed externally.
- Scarcity prevents a bubble. Shortages of watts, wafers, and capital slow the buildout, which he sees as protection against an actual overbuild.
quotes
“TLDR: Everyone can win. Open-source, frontier labs, infrastructure, applications and most importantly the American people.”
“Lower margins at the model layer mechanically, even if counter-intuitively, drive more demand for compute.”
“We are structurally short watts, wafers and increasingly capital.”