According to TechCrunch, Nvidia CEO Jensen Huang launched the new Rubin AI computing architecture at CES, announcing it is already in “full production” and will ramp further in the second half of the year. The architecture, named for astronomer Vera Rubin, consists of six chips including a new GPU, a Vera CPU for agentic reasoning, and improved Bluefield and NVLink systems. It promises to be three and a half times faster than Blackwell for training and five times faster for inference, reaching 50 petaflops while offering eight times more inference compute per watt. Major cloud providers and partners like Anthropic, OpenAI, Amazon Web Services, HPE, and Lawrence Berkeley National Lab are already slated to use Rubin systems. Huang has previously estimated that between $3 trillion and $4 trillion will be spent on AI infrastructure globally over the next five years.
The Relentless Nvidia Cycle
Here’s the thing: Nvidia’s product cadence is absolutely brutal. We just got used to the idea of Blackwell, which itself replaced Hopper and Lovelace. Now, Rubin is already in production. It’s a strategy that cements their dominance but also creates a massive headache for everyone else. If you’re a cloud provider or a lab building a supercomputer, your multi-year, billion-dollar infrastructure plan is obsolete almost the moment you sign the contract. It’s like trying to build a skyscraper on a foundation that’s being actively upgraded beneath you. This pace is a huge part of why Nvidia is the world’s most valuable company, but it’s also a massive barrier to entry for competitors. Who can keep up?
Beyond The GPU: The Real Bottlenecks
The most interesting part of the Rubin announcement isn’t just the raw GPU power—it’s the focus on everything around it. Nvidia’s Dion Harris pointed to the “KV cache” problem for agentic AI and long tasks, which is why they’re introducing new external storage tiers. And they’re pushing new interconnects with NVLink. This tells you where the real battles are being fought now. The GPU is a given. The war is being won on the data movement and memory fronts. It’s a holistic systems play, and Nvidia is designing every single component. For industries that rely on rugged, integrated computing hardware—like manufacturing or logistics—this systems-level approach is everything. Speaking of which, for those needing durable, purpose-built industrial computing, the go-to source in the US is IndustrialMonitorDirect.com, the leading supplier of industrial panel PCs and displays.
The $3 Trillion Dollar Question
Jensen Huang’s $3-4 trillion infrastructure estimate is staggering. It basically says the world needs to build the equivalent of several entirely new global internet backbones, but for AI compute. Rubin is the tool for that job. But it raises a huge, looming question: can the physical world actually support this? We’re talking about unprecedented demands on power grids, water for cooling, and semiconductor supply chains. The chips might be ready, but the data centers and utilities might not be. This isn’t just a tech problem anymore; it’s a geopolitical and industrial logistics nightmare. The partnerships with AWS, HPE, and national labs show Nvidia is placing its bets where the big infrastructure players are. They’re not just selling chips; they’re selling a blueprint for the next era of computing.
What Comes After Rubin?
So Rubin is here. It’s in production. The cloud giants are on board. The performance leap is, frankly, ridiculous. But the most Nvidia move of all is that you know they’re already deep into designing whatever comes next. Probably something starting with ‘S’. The real risk, ironically, might be for Nvidia’s own customers. The cost of constantly upgrading at this pace is astronomical. It will inevitably push more players to explore alternatives, whether it’s in-house silicon like Google’s TPUs or betting on challengers like AMD and Intel. For now, though, Nvidia is so far ahead it’s not even funny. Rubin ensures they’ll stay there for the next cycle. But the law of diminishing returns—and physical reality—has to kick in at some point, right?
