Instinct founder Noah Shinn said the personal AI agent company has seen user growth of about 10% to 11% a day in recent weeks, pushing its compute demand close to doubling every week.
Shinn said roughly 40% of his time is now spent figuring out how much compute capacity the company should buy ahead of demand. The problem, he said, is that compute cannot be purchased in real time at the same pace users are arriving.
He broke down the math this way: if Instinct buys only 2x its current capacity, that supply may be exhausted in about a week. If it buys 5x, it could still be full in less than three weeks. Going straight to 10x capacity would mean committing a large amount of cost upfront.
Shinn also said many resources need to be reserved months in advance, while last-minute compute purchases may cost 3x to 4x more.
Even if Instinct's daily growth rate falls from 10% to a range of 5% to 8%, Shinn estimates that after three to four months of continued growth, the company may still need to prepare compute for a user base of 100 million. The issue is that this is also roughly the amount of time needed for new capacity to come online.
That leaves the company balancing two risks: buy too little and growth cannot be supported; buy too much and costs expand sharply if growth slows.
Shinn said this scaling challenge is harder than the expansion problem faced by typical consumer internet products. When Instagram or Facebook doubles its users, the main task is handling more requests. With personal agents, each additional user continues to consume significant model inference.
He added that the pace of scaling Instinct is dealing with is even faster than the similar issues previously encountered by Claude Code and Codex.