Three years into the AI boom, Nvidia Corp (NVDA) isn't talking about when things might slow down. It's talking about how big things are getting. That was the message from a recent JPMorgan fireside chat with Nvidia's CFO, Colette Kress, who summed up the compute-demand environment in one word: "tremendous." Not cautiously optimistic. Not holding steady. Tremendous.

The Order Book Keeps Growing

Nvidia previously disclosed an over $500 billion backlog stretching through the end of 2026. That number hasn't been sitting still. While management didn't announce an updated total, Kress confirmed that additional customer agreements were signed after that disclosure, including deals with OpenAI and Anthropic. JPMorgan estimates the existing backlog alone could generate roughly $330 billion in data center revenue, which means Nvidia's visibility into future sales is getting clearer, not murkier.

What's particularly notable is how far ahead customers are planning. AI data center buildouts require long lead times, so Nvidia is already working with customers well before equipment gets deployed. Even for 2026 deployments, buyers are trying to expand their existing orders, suggesting demand is still outpacing what Nvidia can deliver.

Networking Becomes Part of the Package

While everyone focuses on GPUs, there's a quieter shift happening in Nvidia's business. The company now sees a 90% attach rate for its networking products with AI deployments, up from roughly 80% last year. Translation: nearly every new Nvidia-powered data center is buying the complete technology stack, not just the chips. That's the kind of lock-in that reinforces Nvidia's position far beyond just selling processors.

Physical AI Gets More Attention

Beyond cloud data centers, Nvidia is pushing harder into physical AI, meaning systems that actually interact with the physical world rather than just processing information. Automotive applications are the most visible example, but management pointed to expanding collaborations across multiple industrial sectors. JPMorgan sees this as a mostly untapped opportunity where Nvidia's integrated platform could once again give it an edge.

Not Without Speed Bumps

Licensing approval for H200 chip shipments to China remains in limbo, and any green light there would be upside to current expectations rather than something baked into forecasts. Meanwhile, Nvidia is preparing to manage multiple product platforms simultaneously while transitioning to its Rubin architecture in late 2026.

The bottom line? Nvidia isn't watching demand taper off. It's watching it pile up. When orders keep accumulating and customers keep buying more of your product suite, the AI story starts looking less like a temporary wave and more like a structural shift.