News
NVIDIA Reveals Details Of Powerful DGX GH200 Supercomputer
The DGX GH200 architecture will enable hundreds of chips to act as a single GPU, and has been built with generative AI in mind.
NVIDIA has revealed details about the company’s next supercomputer, the DGX GH200, which has been built to help companies develop generative AI models.
The machine uses a new NVLink Switch System to enable 256 chips to perform as if they were a single GPU. NVIDIA claims the system offers nearly 500 times as much memory as you’d find in previous DGX A100 computers, quite probably making it the most powerful supercomputer on the planet.
Several key players are interested in the DGX GH200. Google Cloud, Meta, and Microsoft will be some of the first to put the system through its paces as they test its capacity for generative AI workloads.
Also Read: Dubai Launches AI System To Reduce Traffic Congestion
NVIDIA says the new machine will be available by the end of 2023, by which time they should have already developed Helios, an even more powerful version combining four DGX GH200 systems.
Meanwhile, on the gaming front, NVIDIA’s Avatar Cloud Engine (ACE) has been revealed, allowing developers to create custom AI models for speech, conversation, and animation. NVIDIA claims that ACE for Games will “give non-playable characters conversational skills so they can respond to questions with lifelike personalities that evolve”.
News
NVIDIA Puts GPT-5.5 Codex In Hands Of 10,000 Staff
The chipmaker has significantly expanded OpenAI’s latest model across teams from engineering to HR under tight internal controls.
NVIDIA has started rolling out OpenAI’s GPT-5.5 model through the Codex coding agent to more than 10,000 employees, extending the tool well beyond software teams and into core business functions.
The deployment covers engineering, product, legal, marketing, finance, sales, HR, operations and developer programs. Staff are using Codex for coding, internal research and routine knowledge work as companies test whether AI agents can move from demos to daily use.
GPT-5.5 is running on NVIDIA’s GB200 NVL72 rack-scale systems, linking OpenAI’s newest model directly to the chipmaker’s latest infrastructure push. NVIDIA said the systems cut cost per million tokens by 35 times and raise token output per second per megawatt by 50 times versus earlier generations.
Inside the company, it says the effects are immediate. Debugging work that once took days is being finished in hours and experiments across large codebases that used to stretch over weeks are now handled overnight. Teams are also building features from natural-language prompts with fewer failed runs.
In a company-wide note urging staff to adopt the tool, CEO Jensen Huang wrote: “Let’s jump to lightspeed. Welcome to the age of AI.”
Security remains central to the rollout. Codex can connect through Secure Shell to approved cloud virtual machines, allowing agents to work with company data without moving it outside approved environments. NVIDIA said it assigned cloud VMs to employees so agents run in isolated sandboxes with full audit trails.
Also Read: Deezer Says AI Tracks Now Make Up 44% Of Uploads
The company added that the setup uses a zero-data-retention policy. Access to production systems is read-only through command-line tools and internal automation layers.
The move also highlights NVIDIA’s long relationship with OpenAI. NVIDIA said the partnership began in 2016, when Huang personally delivered the first DGX-1 AI supercomputer to OpenAI’s San Francisco office.
The two companies have since worked across hardware and model deployment. NVIDIA also said OpenAI plans to deploy more than 10 gigawatts of NVIDIA systems for future AI infrastructure.
For Gulf markets pouring money into sovereign AI and enterprise automation, the signal is clear: internal AI agents are moving from pilot phase to standard tooling.
NVIDIA Puts GPT-5.5 Codex In Hands Of 10,000 Staff
Deezer Says AI Tracks Now Make Up 44% Of Uploads
Yango Ride Introduces In-Chat Trip Planning In ChatGPT
Nano Banana 2 Arrives In MENA For Google Gemini Users
