AI agents are reshaping how developers work, and the latest breakthrough comes from OpenAI’s Codex application, now driven by the cutting-edge GPT-5.5 model. Deployed on NVIDIA’s powerful GB200 NVL72 rack-scale systems, this combination is delivering unprecedented efficiency gains. Over 10,000 NVIDIA employees across diverse teams—from engineering to HR—are already experiencing transformative results. This article explores five essential insights into this collaboration, highlighting the technology, measurable benefits, security measures, and the decade-long partnership making it all possible. Use the links below to navigate directly to any section.

1. GPT-5.5 and Codex: A New Era in AI-Driven Coding

OpenAI’s latest frontier model, GPT-5.5, powers the agentic coding application Codex, which runs exclusively on NVIDIA GB200 NVL72 rack-scale systems. This infrastructure enables a 35x reduction in cost per million tokens and a 50x increase in token output per second per megawatt compared to previous generations. These economics make frontier-model inference viable at enterprise scale, allowing NVIDIA to deploy AI agents across the company. The result is a dramatic acceleration of complex tasks, from debugging to feature development, all from natural-language prompts. Codex essentially gives each employee a dedicated AI assistant that operates securely within the company’s infrastructure.

2. Measurable Gains: Debugging and Development at Lightspeed

NVIDIA engineers have accessed GPT-5.5 through Codex for just a few weeks, yet the gains are already measurable. Debugging cycles that once took days now close in hours. Experimentation that required weeks is now overnight work, even in complex, multi-file codebases. Teams are shipping end-to-end features from natural-language prompts with stronger reliability and fewer wasted cycles than earlier models. Employees describe the results as “mind-blowing” and “life-changing.” This speed-up is not just incremental; it fundamentally changes how developers approach problem-solving, allowing them to iterate faster and innovate more freely.

3. Enterprise Security: Zero-Data Retention and Secure Sandboxes

To ensure seamless operation within secure enterprise environments, Codex supports remote Secure Shell (SSH) connections to approved cloud virtual machines. NVIDIA’s IT team rolled out individual cloud VMs for every employee, providing a dedicated sandbox for each agent. This setup maintains full auditability while letting the agent operate at maximum capabilities. A zero-data retention policy governs the deployment, and agents access production systems only with read-only permissions through command-line interfaces and the Skills toolkit—the same agentic system NVIDIA uses for automation across the company. Users control their Codex agent from a familiar UI, ensuring ease of use without compromising security.

4. A Decade of Full-Stack Collaboration Between NVIDIA and OpenAI

The GPT-5.5 launch and Codex rollout are the result of over 10 years of collaboration between NVIDIA and OpenAI, starting in 2016 when NVIDIA founder Jensen Huang personally engaged with the OpenAI team. This partnership extends beyond just using NVIDIA hardware; it involves co-optimizing the entire stack—from the silicon to the software—to achieve the lowest cost and highest performance for frontier models. NVIDIA works with every leading model company, but its deep integration with OpenAI allows for breakthroughs like the GB200 NVL72 system, which is purpose-built for these demanding workloads. The collaboration is a testament to how hardware and AI research can advance together.

5. Adoption Across NVIDIA and Jensen Huang’s Call to Action

Over 10,000 NVIDIANs across engineering, product, legal, marketing, finance, sales, HR, operations, and developer programs are now using GPT-5.5-powered Codex. The adoption is company-wide, driven by a clear mandate from CEO Jensen Huang. In a company-wide email urging everyone to use Codex, he said: “Let’s jump to lightspeed. Welcome to the age of AI.” This internal deployment demonstrates the technology’s readiness for enterprise use and highlights NVIDIA’s commitment to dogfooding its own innovations. The feedback loop from these thousands of users also helps refine the system, making it even more powerful for future external releases.

Conclusion: OpenAI’s GPT-5.5 running on NVIDIA GB200 NVL72 represents a significant leap in practical AI for enterprise development. With dramatic cost and speed improvements, robust security, and a foundation of deep collaboration, Codex is already transforming how NVIDIA works. As more organizations adopt similar agentic systems, the age of AI-driven knowledge work is truly here.