Nvidia Corp (NASDAQ:NVDA, XETRA:NVD) used its CES 2026 keynote to underline what it described as “very high” demand for AI computing, unveiling its next-generation Vera Rubin platform and reinforcing its leadership across AI chips, networking and systems, according to Bank of America.
At the keynote and analyst Q&A, CEO Jensen Huang laid out Nvidia’s roadmap for the Vera Rubin platform, which is now in full production and on track for launch in the second half of 2026.
Bank of America said the timeline is consistent with management’s prior commentary following chip tape-out in December.
A key highlight was the announcement of six new chips that will form the backbone of Vera Rubin. These include the Vera CPU, featuring 88 custom Nvidia cores and triple the system memory of the prior Grace platform, and the Rubin GPU, which the bank said delivers up to 5x inference and 3.5x training performance versus Blackwell using NVFP4.
Nvidia also introduced NVLink 6 for scale-up connectivity, the ConnectX-9 800G SuperNIC for east-west traffic, the BlueField-4 DPU for north-south traffic, and the Spectrum-6 Ethernet switch, offering 102.4 terabits of scale-out infrastructure using co-packaged silicon photonics.
Bank of America said the announcements reinforce Nvidia’s deepening competitive “moats” across compute, networking and software, noting that the company continues to run every major large language model in production today.
The analysts also highlighted Nvidia’s view on AI scaling, with model sizes increasing 10-fold per year, token generation rising five-fold annually, and token generation costs falling 10-fold each year. Despite a more modest increase in transistor count versus Blackwell, Rubin is expected to deliver a similar 10x cost reduction per generation, supported by tighter hardware-software co-design. Nvidia also expects a four-fold reduction in the number of GPUs required for training, alongside faster rack assembly and servicing.
Beyond data centres, Nvidia showcased progress in “physical AI,” announcing Alpamayo, an open AI model and toolset for autonomous vehicles, and highlighting its GR00T models for robotics. The first passenger vehicle using the new reasoning model is expected to launch in early 2026 on Nvidia’s DRIVE platform.