Nvidia Corp (NASDAQ:NVDA, XETRA:NVD) used its CES 2026 stage to make the case that the AI boom isn’t cooling, but broadening. In his keynote speech today, CEO Jensen Huang framed the next leg as a shift from “AI models” to “AI systems” that can be deployed at scale, more cheaply, and in more real-world settings — from data centres to cars and robots.
At the centre of that pitch was Rubin, Nvidia’s next-generation data-centre platform and successor to Blackwell.
“Computing has been fundamentally reshaped as a result of accelerated computing, as a result of artificial intelligence,” Huang said. “What that means is some $10 trillion or so of the last decade of computing is now being modernised to this new way of doing computing.”
The headline launch: Rubin, built as a six-part stack
Nvidia describes Rubin as its “first extreme-codesigned AI platform” — a six-chip system designed as one unit, rather than a standalone GPU upgrade. The core pieces are Rubin GPUs, Vera CPUs, NVLink 6, Spectrum-X Ethernet Photonics, ConnectX-9 SuperNICs and BlueField-4 DPUs.
Rather than pitching a faster GPU in isolation, Nvidia is arguing that the next wave of AI gains will come from tightly integrated systems — where compute, networking, memory and software are designed together to reduce bottlenecks and cost. In practice, that means fewer chips doing more useful work, and far more efficient “token generation” — the basic unit that underpins everything from chatbots to image generation and AI agents.
It reflects a broader shift under way in the AI market. Training ever-larger frontier models still grabs headlines, but the real commercial challenge is increasingly inference — running those models reliably, cheaply and at massive scale. As AI tools move from demos into everyday products, cost per query matters just as much as peak performance.
Rubin is Nvidia’s answer to that problem: a platform designed to make AI workloads cheaper to serve, not just faster to train.
From chips to ‘AI factories’
Throughout the CES presentation, Huang repeatedly returned to the idea of “AI factories” — data centres purpose-built to manufacture intelligence at industrial scale. Nvidia is positioning itself less as a chip supplier and more as the infrastructure backbone of the AI economy.
That helps explain why Rubin is being launched alongside new networking, data-processing units and photonics, rather than as a standalone GPU cycle. The message to customers is that the competitive edge now lies in whole-system optimisation — something Nvidia believes only it can deliver end to end.
It’s also a subtle acknowledgement of where pressure is building. Hyperscalers are increasingly designing their own silicon, especially for inference, and competitors are targeting narrower, cheaper workloads. Nvidia’s response is to move further up the stack, making itself harder to replace piece by piece.
The push into ‘physical AI’
The other major theme of the keynote was what Huang called “physical AI” — systems that don’t just generate text or images, but perceive, reason and act in the real world.
That concept underpins Nvidia’s renewed emphasis on robotics, autonomous vehicles and simulation platforms. If AI is moving beyond screens and into machines, the demands on compute change again: latency, reliability and real-time decision-making become critical.
For Nvidia, this is a familiar playbook. The company has spent years building software platforms and developer ecosystems around automotive and robotics, even when commercial uptake lagged. CES 2026 suggests Nvidia believes the technology — and demand — is finally catching up.
Importantly, this is also where Nvidia sees open models and rapid iteration as an advantage. As models improve every few months, the value shifts to the hardware and systems capable of running them efficiently in complex environments.
What this says about the tech outlook
Stepping back, Nvidia’s CES message was notably confident — not just about its own roadmap, but about the durability of AI investment more broadly.
The emphasis on cost, efficiency and deployment suggests the industry is moving out of its experimental phase and into one focused on scale and sustainability. That tends to favour incumbents with deep engineering resources and existing customer relationships — but it also raises the bar for returns.
For investors, the takeaway isn’t simply that “AI demand is strong.” It’s that the next phase of AI growth may look less like a gold rush and more like industrialisation: slower, more competitive, and increasingly shaped by who can deliver intelligence at the lowest marginal cost.
Nvidia is betting that tightly integrated platforms like Rubin — and a broader push into physical AI — will keep it at the centre of that shift. Whether the market agrees will depend less on keynote claims, and more on how quickly those systems show up in real-world deployments over the next 12 to 18 months.