NVIDIA, the chipmaker and AI infrastructure provider, is using GPT-5.5-powered Codex across its workforce, reporting faster software development cycles and lower computing costs.
The agentic coding tool Codex, built by OpenAI, now runs on NVIDIA’s GB200 NVL72 rack-scale systems, pairing the latest GPT-5.5 model with high-performance AI infrastructure designed for enterprise use.
More than 10,000 NVIDIA employees across engineering, product, legal, marketing, finance, sales, HR, and operations are already using the system. Internal feedback describes results as “mind-blowing” and “life-changing,” according to the media advisory.
Engineers who have used GPT-5.5 through Codex for several weeks report measurable gains. Debugging tasks that used to take days are now completed in hours. Complex experiments that once required weeks are being finished overnight, even across large, multi-file codebases. Teams are also building full features directly from natural-language prompts with improved reliability and fewer wasted cycles.
The efficiency gains are linked to NVIDIA’s GB200 NVL72 systems, which promises to deliver up to 35 times lower cost per million tokens and 50 times higher token output per second per megawatt compared with earlier-generation infrastructure. These improvements make large-scale AI model use more practical for enterprise deployments.
“Let’s jump to lightspeed. Welcome to the age of AI,” said Jensen Huang in a company-wide email.
NVIDIA said the deployment is designed for secure enterprise environments. Codex supports remote Secure Shell (SSH) connections to approved cloud virtual machines, allowing AI agents to work with internal company data without exposing it externally.
To strengthen security and auditability, NVIDIA IT assigned cloud virtual machines to each employee. These act as isolated environments where agents can run safely while maintaining full tracking of activity. Employees can control the Codex agent through a familiar user interface.
The system also follows a zero-data retention policy. AI agents access production systems with read-only permissions through command-line tools and internal “Skills,” the same toolkit NVIDIA uses to automate workflows across the company.