What just happened? Nvidia has rolled out OpenAI's latest frontier model internally, giving more than 10,000 employees access to GPT-5.5 through OpenAI's Codex agentic coding application. The deployment spans engineering, product, legal, marketing, finance, sales, HR, operations, and developer programs, with employees using Codex for both knowledge work and software development.
Codex now runs on GPT-5.5 hosted on Nvidia's GB200 NVL72 rack-scale systems. Nvidia says the systems deliver 35x lower cost per million tokens and 50x higher token output per second per megawatt than prior-generation systems, economics it argues make frontier-model inference viable at enterprise scale.
Inside Nvidia, engineers have been using the GPT-5.5-powered Codex app for several weeks, and the company says the gains are already measurable in how they build and maintain software. Debugging work that used to take days is now being completed within hours, and experiments that previously required weeks are progressing overnight in complex, multi-file codebases, according to the Nvidia post. Teams are also using natural-language prompts to deliver end-to-end features with greater reliability and fewer wasted cycles than earlier models, the company says.
Nvidia founder and CEO Jensen Huang urged employees to make use of Codex in an internal email. "Let's jump to lightspeed. Welcome to the age of AI," he wrote.
The rollout uses an enterprise architecture in which Codex agents run in sandboxed cloud virtual machines, enabling them to work with real company data without exposing it externally. Each agent runs on a dedicated cloud virtual machine, with the Codex desktop app connecting via Secure Shell (SSH) to approved VMs.
From a policy perspective, Nvidia says the deployment is governed by a zero-data-retention model and read-only integrations into production systems. Agents interact with those systems over command-line tools and "Skills" – the same internal toolkit used to run automation workflows across the company, which provides an additional control layer on what Codex can execute.
Nvidia frames the GPT-5.5 deployment as the latest step in a more-than-decade-long collaboration with OpenAI across hardware, software and model deployment. Their collaboration goes back to 2016, when Huang delivered an Nvidia DGX-1 AI supercomputer to OpenAI's San Francisco headquarters.
More recently, that collaboration produced what Nvidia calls a major milestone: the joint bring-up of the first GB200 NVL72 100,000-GPU cluster, which completed multiple large-scale training runs and "set a new benchmark for system-level reliability at frontier scale." GPT-5.5, the company concludes, "is the product of that infrastructure running at full strength."