Nvidia’s AI advantage is moving beyond the GPU
TechCrunch Russell Brandom ● Covered by 3 sources
Nvidia’s AI edge is no longer just about GPUs. It’s also winning the messy job of keeping giant AI systems fed, cooled, and moving.
Based on reporting by TechCrunch, Russell Brandom — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
For a while, the Nvidia story was simple: it had the best AI chips, everyone needed them, and the money followed. That picture is getting messier. Amazon and Google are building their own silicon now, investors have spent the past year worrying about GPU competition, and Nvidia’s stock stopped sprinting the way it did after the company’s market cap grew 10x between the start of 2023 and mid-2025.
But Nvidia’s latest earnings pushed a different idea to the front. The company’s real moat may be shifting from the GPU itself to everything around it. As AI systems scale up toward gigawatt-sized compute, the hard part is no longer just raw chip performance. It’s orchestration: getting data where it needs to go, when it needs to go there, without wasting power or time.
That’s where Nvidia is leaning in. Its Vera Rubin architecture pairs the Rubin GPU with other hardware, including the Vera CPU, the Groq 3 LPX inference accelerator, and related storage and networking racks. Those systems aren’t designed to do the same job as the GPU. They exist to make the rest of the machine behave better. Think less “more horsepower” and more “the engine finally has a working transmission.”
Jason Hardy, Nvidia’s VP of storage technology, says the Vera CPU is meant to solve the simple but brutal problem of memory bottlenecks. There’s only so much memory you can fit in one server, and getting the right data to the GPU fast enough is getting harder as deployments get bigger. Hardy said Nvidia has seen “upwards of 3x improvement” in those operations, and that the system helps flash storage reach its full potential without bottlenecking.
OpenAI is attacking the same problem from the other side. Its Jalapeño chip was built to minimize data movement and keep a whole workload inside one connected system. Different hardware, same obsession: move less data, waste less time, squeeze more work out of every watt. That’s the new fight.
And that fight may matter more than the old one. Nvidia still has to deal with hyperscalers and rival chipmakers, but the battleground has widened. It’s no longer just about building a better GPU. It’s about making the whole machine run like it was designed by people who actually had to pay the electric bill.
My take — AI-written commentary, not fact-checked reporting
The GPU was always only half the story, and anyone calling compute a commodity was doing it from a safe distance. The real money now sits in the unglamorous plumbing: traffic control, memory, storage, networking, the stuff that keeps giant AI systems from turning into very expensive heaters. Nvidia knows that, which is why the next chip war looks a lot more like infrastructure warfare with better branding.
Read more about this at: TechCrunch