sf

There are only 4 ways Frontier AI Labs will be able to earn defensible profits. 1. Monopoly/Duopoly via getting to RSI / AGI / ASI first with a large capability-gap away from Lab(s) #2/#3 which lasts for a sufficiently long time to farm significant superprofits 2. Monopoly/Duopoly via Regulation / Policy / Special Mandates / Partial Nationalization 3. Use frontier AI to design/utilize/sell superior new Hardware / Chips / Infra / Networking. 4. Compete with their own customers... in Domains where even a tiny advantage from SOTA AI Models compounds and results in winner-take-all effects and where SOTA is more easily monetizable (Applied Science, Bio, Pharma, Patentable Materials, Cyber, Trading, etc.) Frontier Labs are trying to do all 4 simultaneously. Some of these are obviously harder than others. Without successfully execution on some combination of these, margins will gradually shrink. Financiers of the coming OpenAI private round and the coming Anthropic IPO understand this calculus. That is why almost all of the marketing, messaging, blog posts and PR coming out of the labs is always trying to signal an improvement in one of the 4 items above.