OpenAI recently published work on the infamous Navier–Stokes Millennium Prize Problem, sparking a massive community debate over the sheer scale of modern AI experiments. To tackle the problem, their AI agents exchanged 4.9 million messages and consumed nearly 300 billion output tokens—a workload that commenters calculate would cost a staggering $15 million at standard public API prices.
For early-stage founders, this highlights a stark reality about the current AI landscape: you cannot out-spend Big Tech on raw compute. If your startup’s moat relies on burning millions of tokens to brute-force a solution, your unit economics will collapse long before you reach a Series A.
For small backers and angel investors, the takeaway is clear: focus on capital efficiency. The most promising early-stage teams aren't the ones burning small fortunes on massive compute runs, but the agile builders using lean, targeted models to solve clear, everyday problems.