'It was awesome to see it work': OpenAI deploys GPT-5.5 Codex across Nvidia Blackwell systems – 50x efficiency boost and 35x cost reduction makes AI ‘viable at enterprise scale’

News
By published

OpenAI has deployed Codex on Nvidia systems

An Nvidia GB200 server rack in a data center with one rack open and exposed.
(Image credit: Nvidia)
  • Nvidia says OpenAI Codex deployment offers 50x increase in token output
  • Offers 35x reduction in cost compared to GPT-4o
  • 10,000 NVIDIA employees now operate codex across HR, legal, engineering and more

OpenAI has deployed its Codex model powered by GPT5.5 across Nvidia in an AI-native enterprise deployment.

The model is now available to over 10,000 Nvidia employees, and has reportedly resulted in a 35x reduction in cost and 50x increase in token output per megawatt compared to GPT-4o.

Codex has already delivered “mind-blowing” and “life-changing” results, according to Nvidia employees.

Article continues below

‘Frontier-model inference viable at enterprise scale’

In a blog post, Nvidia says, “Debugging cycles that once stretched across days are closing in hours. Experimentation that previously required weeks is turning into overnight progress 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.”

The model has been deployed on Nvidia's GB200 NVL72 rack-scale systems, and this is what has made the huge savings in cost and token output possible, Nvidia says.

The Nvidia-Codex adoption moves beyond the typical AI deployments most businesses have seen. Nvidia CEO Jensen Huang said, “Chatbots answer questions. Agents do work."

Codex’s original function was to aid GitHub users as a Copilot-style coding assistant, but can now operate as an agentic assistant that can perform tasks across non-technical departments such as product, legal, marketing, finance, sales, HR, and operations.

An email from Jensen Huang, shared by Sam Altman on X, says, “Incredibly excited to all [that] will use Codex to accelerate and do work never possible before. Please convey my congratulations to your team on showing the world the frontier yet again. And thank them again for inventing GPT that gave us the springboard to reason, plan, use tools, and beyond.”

The deployment of Codex on Nvidia systems comes as OpenAI’s rivalry with firms such as Anthropic is reaching new heights, especially as Anthropic’s Claude Mythos draws widespread attention.

While both models have wide-ranging and potentially revolutionary applications, Anthropic’s Mythos focuses on patching zero-day vulnerabilities at scale (as demonstrated by Firefox), whereas OpenAI’s GPT-5.5 Codex can handle real company data in a secure cloud sandbox, allowing employees to move beyond chatbots and interact with an AI agent capable of securely automating entire workflows.

Google logo on a black background next to text reading 'Click to follow TechRadar'

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.

Benedict Collins
Benedict Collins
Senior Writer, Security

Benedict is a Senior Security Writer at TechRadar Pro, where he has specialized in covering the intersection of geopolitics, cyber-warfare, and business security.

Benedict provides detailed analysis on state-sponsored threat actors, APT groups, and the protection of critical national infrastructure, with his reporting bridging the gap between technical threat intelligence and B2B security strategy.

Benedict holds an MA (Distinction) in Security, Intelligence, and Diplomacy from the University of Buckingham Centre for Security and Intelligence Studies (BUCSIS), with his specialization providing him with a robust academic framework for deconstructing complex international conflicts and intelligence operations, and the ability to translate intricate security data into actionable insights.

You must confirm your public display name before commenting

Please logout and then login again, you will then be prompted to enter your display name.