OpenAI scales Habitat storage platform to 1 billion ChatGPT users

The headline number, 22 million requests per second, is striking, but the more consequential detail is the organizational one: a library originally written for research use was pressed into production duty and then had to be rebuilt around it, rather than replaced. That path dependency shapes every architectural decision that follows.

Modelwire has no prior coverage in its archive that connects directly to this story. It belongs to a cluster of infrastructure-at-scale reporting that has been building across the industry, covering how frontier AI services discover that model capability ships faster than the operational plumbing beneath it. The Habitat story is a concrete, named example of that pattern, which until now has mostly been discussed in the abstract when companies talk about reliability and latency at consumer scale.

Watch whether OpenAI publishes latency percentile data or open-sources any Habitat components in the next six months. If they do, it signals confidence that the architecture is stable enough to withstand external scrutiny. If the post remains a one-time blog entry with no follow-up, the more likely read is that this was a recruiting and credibility signal rather than a technical contribution.