The Dawn of Agentic Knowledge Work
AI agents have redefined how developers approach their tasks, but the next leap forward lies in knowledge work: processing vast amounts of information, solving intricate problems, generating novel ideas, and fueling innovation. OpenAI’s Codex, an agentic coding application, is at the forefront of this transformation. Now powered by GPT-5.5, OpenAI’s latest frontier model, Codex runs on NVIDIA GB200 NVL72 rack-scale systems, delivering unprecedented performance and efficiency.
Internally, over 10,000 NVIDIA employees—spanning engineering, product, legal, marketing, finance, sales, HR, operations, and developer programs—are already leveraging GPT-5.5-powered Codex. Their feedback? Terms like “mind-blowing” and “life-changing” are common. This adoption marks a significant shift in how enterprise teams approach complex coding and knowledge tasks.
Codex and GPT-5.5: A Powerful Combination
The integration of GPT-5.5 with Codex on the GB200 NVL72 platform yields remarkable economics. Compared to prior-generation systems, the new infrastructure delivers 35x lower cost per million tokens and 50x higher token output per second per megawatt. These gains make frontier-model inference viable at enterprise scale, enabling organizations to deploy advanced AI without prohibitive costs.
NVIDIA engineers have had access to this setup for a few weeks, and the results are measurable. The combination of GPT-5.5’s reasoning capabilities and the GB200’s computational power is reshaping development cycles.
Measurable Gains Across NVIDIA
The impact on productivity is tangible. Debugging cycles that once stretched across days are now completed in hours. Experimentation that required weeks can be turned into overnight progress, even in complex, multi-file codebases. Teams are shipping end-to-end features directly from natural-language prompts, with stronger reliability and fewer wasted cycles compared to earlier models.
Faster Debugging and Experimentation
One of the most striking changes is the acceleration of debugging. Previously, identifying and fixing issues in large codebases could take multiple days of manual effort. With GPT-5.5-powered Codex, these cycles collapse to a matter of hours. Similarly, experimentation—testing new approaches or validating hypotheses—now happens overnight rather than over weeks. This speed allows NVIDIA teams to iterate faster and focus on higher-value work.
As NVIDIA founder and CEO Jensen Huang noted in a company-wide email urging everyone to use Codex: “Let’s jump to lightspeed. Welcome to the age of AI.” This sentiment reflects the transformative potential of the technology.
Enterprise-Grade Security and Deployment
For enterprise environments, security is paramount. Just as every human needs a dedicated workspace, every AI agent requires its own secure computing environment. Codex supports remote Secure Shell (SSH) connections to approved cloud virtual machines (VMs). This arrangement allows agents to work with real company data without exposing it externally.
A Sandbox for Every Employee
NVIDIA IT rolled out cloud VMs for every employee to run their agents safely. Each VM provides a dedicated sandbox where the agent can operate at maximum capabilities while maintaining full auditability. Users control the Codex agent running in the cloud VM from a familiar user interface, ensuring ease of use without compromising security.
A zero-data retention policy governs NVIDIA’s deployment, meaning no data is stored beyond the session. Agents access production systems with read-only permissions through command-line interfaces and Skills—the same agentic toolkit NVIDIA uses to run automation workflows across the company. This approach balances power with strict governance.
A Decade of Collaboration
The GPT-5.5 launch and Codex rollout represent more than just technological progress—they are the result of over 10 years of collaboration between NVIDIA and OpenAI. The partnership began in 2016, when Jensen Huang hand-delivered the first NVIDIA DGX-1 supercomputer to OpenAI. Since then, the two companies have worked together on optimizing models, scaling infrastructure, and pushing the boundaries of AI.
This ongoing collaboration ensures that frontier models like GPT-5.5 are not only powerful but also cost-effective and energy-efficient. By working with every frontier model company, NVIDIA accelerates the use of AI agents internally while helping partners build the world’s best models for everyone.
For more on how NVIDIA is enabling enterprise AI, explore Enterprise-Grade Security and A Decade of Collaboration.