Jensen Huang confirmed that Nvidia’s next-generation AI systems, built on the Rubin platform, are in full production and will ship to customers on schedule in the second half of 2026.

The Rubin architecture is Nvidia’s successor to Blackwell, the GPU family that has already dominated data center buildouts across the planet. Analysts project approximately 5.7 million Rubin GPUs could ship in 2026.

What Rubin actually brings to the table

Nvidia is claiming the Rubin platform delivers 10 times the token processing capability at lower cost compared to its predecessor. The platform features GPUs capable of 50 petaflops of NVFP4 inference performance. Nvidia pairs those GPUs with its Vera CPUs, creating what the company describes as a unified AI supercomputer solution.

Huang unveiled the Rubin roadmap at CES 2026 in January, describing it as an “extreme co-designed six-chip platform.” During a subsequent Computex 2026 keynote, he elaborated on what he called a “five-layer cake” design framework, essentially Nvidia’s philosophy of optimizing every layer of the AI stack from silicon to cloud infrastructure simultaneously.

Rack-scale systems like the NVL72 can incorporate up to 72 GPUs in a single deployment, targeting the exact kind of massive compute clusters that hyperscalers are racing to build. AWS, Google Cloud, and Microsoft are all expected to deploy Rubin-based products once deliveries begin.

Why crypto investors should care about GPU roadmaps

The explosion in AI compute demand is driving infrastructure buildouts that also benefit decentralized compute networks. Projects like Render, Akash, and io.net are essentially trying to create decentralized versions of the GPU cloud that Nvidia’s biggest customers are building. When Nvidia ships faster, cheaper chips, the economics of these decentralized networks shift too.

The confirmation that Rubin is on track, with no noted delays in the production schedule, suggests Nvidia’s supply chain has matured significantly since the shortage-plagued days of 2021 and 2022. Huang has repeatedly emphasized supply chain reliability and production capacity as strategic priorities.

The competitive landscape and what to watch

AMD continues pushing its Instinct MI series, while custom silicon efforts from Google (TPUs), Amazon (Trainium), and Microsoft are all aimed at reducing dependence on Nvidia hardware. Startups like Cerebras and Groq are attacking different parts of the inference and training market.

Projections of 5.7 million Rubin GPU shipments in a single year suggest Nvidia’s manufacturing scale remains in a different league. The company’s comprehensive approach, designing CPUs, GPUs, networking, and software as an integrated stack, creates switching costs that competitors struggle to replicate.

All three major cloud providers are lined up as launch partners for Rubin. The key metric to track is how quickly Rubin adoption drives down the cost per token of AI inference, which affects the unit economics of AI-crypto hybrid projects that depend on affordable compute.