NVIDIA’s US$96 Billion Quarter Says the AI Race Has Barely Begun

NVIDIA has spent much of 2026 serving as both the great beneficiary of the artificial-intelligence boom and its most closely watched stress test. Every bout of technology volatility, every question over hyperscaler spending and every warning about power, memory or financing has revived the same argument: has the industry begun building ahead of the economics?

The latest numbers suggest otherwise. NVIDIA reported fiscal second-quarter revenue of US$96.2 billion, more than double a year earlier, while data-centre revenue climbed 117 per cent to US$89 billion. It expects US$108 billion this quarter, despite assuming no data-centre compute revenue from China.

More significant was the view beyond the quarter. NVIDIA expects revenue to grow approximately 70 per cent in fiscal 2028, dramatically ahead of what Wall Street had been modelling. Crucially, management described that forecast as “supply-constrained”. Demand, in other words, is not yet the principal limitation. The constraint is how quickly the industry can manufacture, finance, power and connect the infrastructure required to satisfy it.

“AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.” — Jensen Huang, founder and chief executive officer, NVIDIA

This also explains why the competitive landscape is changing. The contest is no longer simply NVIDIA against AMD. Google, Amazon and other hyperscalers are developing custom accelerators, while semiconductor groups are investing heavily in chips designed specifically for inference, where trained models are deployed millions or billions of times at the lowest possible cost. AMD’s recent expansion into specialised inference technology and rising demand for custom silicon underline where the next battle is moving.

That transition matters because inference potentially turns AI from an episodic training expense into a continuous computing economy. Every coding agent, enterprise assistant, autonomous system, robot and AI search request consumes compute after the model has been trained. Jensen Huang’s argument that “compute is revenue” increasingly describes the commercial architecture being constructed around AI.

The physical consequences are enormous. New AI campuses require GPUs, memory, networking and cooling, but also transformers, generators, cables, cement, land and increasingly dedicated power generation. The American data-centre boom is already radiating through industrial supply chains, while projects measured in hundreds of megawatts demonstrate how quickly AI is becoming an infrastructure market rather than merely a technology market.

Why Does It Matter?

Because NVIDIA’s quarter suggests the AI race is entering a different phase. The first chapter was about proving that increasingly capable models could be built. The next is about deploying them economically, everywhere.

That creates formidable risks: memory shortages, power constraints, permitting, enormous financing requirements, export controls and the possibility of overbuilding. Competition will also intensify as customers seek cheaper alternatives to NVIDIA’s architecture.

But none of those pressures points to an AI economy standing still. They reveal something more consequential: the race is moving out of the laboratory and into the physical economy. NVIDIA’s forecast suggests that transition has only just begun.

Get the stories that matter to you.
Subscribe to Cyber News Centre and update your preferences to follow our Daily 4min Cyber Update, Innovative AI Startups, The AI Diplomat series, or the main Cyber News Centre newsletter — featuring in-depth analysis on major cyber incidents, tech breakthroughs, global policy, and AI developments.