OpenAI has launched its latest frontier model, GPT-5.5, to power its agentic coding application, Codex, running exclusively on NVIDIA's high-performance GB200 NVL72 rack-scale systems. This deployment marks a significant milestone for enterprise AI, with over 10,000 NVIDIA employees across all departments already utilizing the tool to report transformative productivity gains. The infrastructure upgrade delivers measurable economic and performance benefits. Served on the GB200 NVL72, the system achieves a 35 percent reduction in cost per million tokens and a 50 percent increase in token output per second per megawatt compared to previous generations. These efficiency improvements make running frontier models viable at an enterprise scale. In practice, debugging cycles that previously took days are now completed in hours, and complex experimentation requiring weeks of work is now finishing overnight. Teams are shipping end-to-end features directly from natural language prompts with greater reliability and fewer wasted resources. Jensen Huang, founder and CEO of NVIDIA, encouraged all employees to adopt the tool in a company-wide email, declaring it the dawn of an age where work jumps to lightspeed. To ensure these capabilities operate safely within corporate environments, NVIDIA implemented a security-first deployment strategy. Each agent is assigned a dedicated cloud virtual machine via secure remote SSH connections, preventing external data exposure. NVIDIA IT provisioned these virtual machines to provide a sandbox for maximum performance while maintaining full auditability. A zero-data retention policy governs the deployment, ensuring no information is stored long-term. Agents access production systems with read-only permissions through command-line interfaces and a specific toolkit called Skills, which NVIDIA uses for internal automation workflows. Employees control the agents through a familiar user interface, balancing powerful capabilities with strict security protocols. This collaboration represents more than a decade of partnership between the two companies, dating back to 2016 when Huang personally delivered the first NVIDIA DGX-1 supercomputer to OpenAI. The relationship has evolved into a full-stack integration, with NVIDIA serving as a day-zero partner for OpenAI's open-weight models and optimizing their weights for frameworks like TensorRT-LLM, vLLM, and Ollama. Looking ahead, OpenAI has committed to deploying more than 10 gigawatts of NVIDIA systems for its next-generation infrastructure, placing millions of GPUs at the foundation of future model training and inference. Furthermore, the companies function as early silicon and co-design partners. OpenAI provides critical feedback that shapes NVIDIA's hardware roadmap, while NVIDIA grants OpenAI early access to new architectures. This symbiotic relationship culminated in the joint bring-up of the first GB200 NVL72 cluster containing 100,000 GPUs. This massive system successfully completed multiple large-scale training runs, setting a new benchmark for system-level reliability at a frontier scale. GPT-5.5 represents the culmination of this infrastructure working at full strength, driving the next wave of innovation in AI agents and knowledge work.