According to ChainCatcher, Instinct founder Noah Shinn said the personal AI agent company has posted user growth of about 10% to 11% per day in recent weeks, pushing compute demand close to doubling every week.
Fast user growth is turning compute planning into a core constraint
Shinn said he now spends roughly 40% of his time thinking about how much compute the company should buy ahead of demand. The issue, he said, is that compute cannot be procured in lockstep with user growth.
If Instinct buys only 2x its current compute capacity, that capacity would be exhausted in about a week. Buying 5x would still be used up in less than three weeks. Buying 10x would mean committing a large amount of capital well in advance.
Lead times are measured in months, and last-minute capacity costs more
Shinn said many resources need to be reserved months in advance. If the company has to add compute on short notice, the price could be 3x to 4x higher. In his description, underbuying makes it hard to support growth, while overbuying magnifies costs if growth slows.
Even slower growth could still require planning for 100 million users
He estimated that even if daily growth falls from 10% to 5% to 8%, three to four months of continued growth could still leave the company needing to prepare compute for a user base of 100 million. He added that the wait time for new capacity to come online is roughly comparable.
Shinn said this is harder than scaling consumer internet products such as Instagram and Facebook because each additional user of a personal agent continuously consumes a large amount of model inference.
He also said Instinct is currently scaling even faster than the similar situations previously faced by Claude Code and Codex.