At the Consumer Electronics Show in Las Vegas, Jensen Huang, chief executive of Nvidia, said the company’s latest AI computing platform, built around its new Vera Rubin chips, has entered full production and is already being tested by major cloud providers and AI labs. The platform combines a new generation of GPUs and CPUs, delivering up to five times the AI computing power of Nvidia’s current chips for chatbots and other generative models, while significantly reducing per-token costs.
The move comes at a time when Nvidia, now the world’s most valuable listed company, is facing intensifying competition from AMD as well as more capable in-house AI accelerators developed by Google, Amazon, and other large cloud clients.
Rubin Platform Promises Big AI Performance Boosts
Jensen Huang said Nvidia has finished manufacturing the chips behind its new Vera Rubin platform and is now assembling full systems inside its own labs. The platform combines six Nvidia chips, including the Rubin GPU and a newly designed CPU, marking the company’s next major step in AI hardware. A single Rubin server contains 72 GPUs and 36 CPUs, while multiple servers can be linked into pods of more than 1,000 chips for training and running large AI models in massive data centres.
According to Nvidia, the Rubin platform delivers up to five times faster AI inference and 3.5 times quicker training than systems based on its current Blackwell chips. In some cases, it can also cut the number of GPUs needed to train large models by four times. Huang said the improvements come not from a dramatic jump in transistor count, but from changes in how the chips process numbers and generate tokens, allowing the system to produce far more output from only a modest increase in silicon.
Cloud Giants Among First Rubin Customers
Nvidia said CoreWeave will be one of the first cloud providers to deploy complete Vera Rubin systems. Larger players, including Microsoft, Oracle, Amazon and Alphabet, are expected to follow once the platform becomes generally available in the second half of 2026. Some early Rubin hardware is already with a small group of AI firms, which are using it to test new models and measure performance against Blackwell systems and competing accelerators.
Rather than selling Rubin as just another chip, Nvidia is rolling it out as a complete platform that tightly integrates hardware with its CUDA software, networking tools, and AI frameworks. For customers already running Nvidia systems, that makes upgrading far simpler than switching to a different architecture. Analysts say this tight integration continues to give Nvidia a powerful edge, locking customers into its ecosystem and raising the bar for even well-funded rivals trying to break in.
Competition And Regulatory Pressures Intensify
Nvidia may have one of the most ambitious roadmaps in the AI chip industry, but its lead is increasingly being tested. Rivals and some of its biggest customers are moving quickly to build their own accelerators, aiming to chip away at Nvidia’s dominance. At the Consumer Electronics Show, AMD promoted its latest MI455 GPUs, while Intel rolled out new processors for servers and PCs, underlining just how crowded the high-performance market has become.
The pressure is also coming from customers. Google, Amazon and Meta are all investing heavily in their own chips to cut reliance on Nvidia, with Google’s TPU designs often cited as one of the biggest long-term threats to the Rubin platform. Nvidia chief Jensen Huang said the company’s recent deal to bring in talent and technology from startup Groq would not disrupt its main business, but could open the door to new products later on.
Beyond competition, Nvidia is also facing growing regulatory scrutiny in the US and Europe over its market power and pricing. At the same time, US export controls are narrowing what it can sell to China, even as demand there remains strong for older H-series GPUs.
Stay updated with the latest global technology and business news on Inspirepreneur Magazine.
