OpenAI has expanded access to Codex, its GPT-5.5-powered coding tool, across Nvidia’s workforce, signaling that the product is moving beyond limited testing and into broader corporate use. The rollout has drawn public support from both Sam Altman and Jensen Huang, while Nvidia says usage has already spread across multiple business functions.
Nvidia reports that more than 10,000 of its employees are now using Codex powered by GPT-5.5. The users are not limited to engineers, but also include teams in product, legal, marketing, finance, sales, HR, operations, and developer programs.
That wide adoption suggests the tool is being treated as a practical work assistant inside the company rather than a narrow developer experiment. Nvidia said the internal response has been strongly positive, with some employees describing the impact as “mind-blowing” and “life-changing.”
According to Nvidia’s blog, engineers at the company have already been using GPT-5.5 through the Codex app for several weeks. The company said the effects are visible and measurable in day-to-day work.
Nvidia also linked the deployment to a larger shift in software development practices. The company said tasks that once took days, such as debugging, can now be completed in hours, while experiments on complex codebases that previously stretched across weeks may now move forward overnight.
The infrastructure behind the rollout is also central to Nvidia’s message. Codex is running on the company’s rack-scale GB200 NVL72 system, which Nvidia described as an important foundation for enterprise-scale inference of frontier models.
In its own blog post, Nvidia said the system can deliver a cost that is 35 times lower per million tokens. It also claimed output tokens per second per megawatt can be 50 times higher than with previous-generation systems.
That efficiency has become a key talking point in the wider AI coding market. Token usage matters because heavier tasks consume more tokens, and competition among model providers is increasingly shaped by how much work a system can complete before hitting limits.
The context matters even more after some users of Anthropic complained that Claude token limits felt tighter than before. Against that backdrop, OpenAI has positioned Codex as a more efficient option that can handle similar work with far fewer tokens.
OpenAI has also expanded Codex beyond a basic coding helper into a broader software engineering environment. The company has integrated GPT-5.5 into the product as part of that push, including testing on Nvidia’s own compute infrastructure.
Sam Altman described the collaboration positively on X, saying OpenAI was trying a new approach with Nvidia to roll out Codex across the organization. He said the trial had gone very well and pointed to it as a promising step for the product’s development.
Jensen Huang offered a similarly direct endorsement to Nvidia staff. He reportedly sent an internal email encouraging employees to use Codex, making clear that AI adoption was being treated as part of the company’s operating direction rather than a side project.
His message was explicit about the broader shift. With the line, “Let’s jump to lightspeed. Welcome to the age of AI,” Huang framed the rollout as part of a larger transformation in how Nvidia works.
The collaboration also highlights how closely model providers and infrastructure companies are now working together. As competition sharpens with Anthropic’s Claude Code gaining attention for its ability to generate large amounts of code independently, OpenAI appears determined to show that Codex is ready for enterprise-scale deployment with measurable results.