NVIDIA’s Earnings Show the AI Boom Still Has Room to Run

For much of this year, the semiconductor trade has been caught between two competing stories. One says artificial intelligence is still in the opening phase of a once-in-a-generation infrastructure buildout; the other says valuations, capital spending and expectations have run too far. NVIDIA’s latest quarter did not settle that argument, but it shifted the burden of proof back toward the optimists.

Revenue reached US$96.2 billion in the fiscal second quarter, up 106% from a year earlier, while Data Centre sales rose 117% to US$89 billion. NVIDIA guided to roughly US$108 billion in the current quarter, even while assuming no Data Centre compute revenue from China. Then came the surprise on the call: management indicated fiscal 2028 revenue could grow around 70%, well above the roughly 45% Wall Street had been modelling, with Jensen Huang arguing that the constraint is increasingly supply rather than demand.

That distinction matters. NVIDIA is no longer simply selling more GPUs into a hot market; it is trying to become the operating architecture of the AI factory, spanning CPUs, accelerators, networking, rack-scale systems, software and physical AI. Vera Rubin is already in full production, following Blackwell at remarkable speed, and the company expects the transition to accelerate rather than interrupt demand.

Huang put the moment more plainly in NVIDIA’s official announcement:

“AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.” He also described a “golden age”

of new AI labs, open models and physical AI coming online around the world. The language is expansive, but the numbers beneath it are becoming harder to dismiss.

The broader chip complex tells a similar story, although not with equal strength. TSMC is expanding advanced-node and CoWoS capacity, Broadcom is benefiting from custom AI accelerators and networking, AMD is pushing harder into rack-scale systems, while Intel is still trying to turn manufacturing weakness into a foundry opportunity. Memory suppliers including SK Hynix, Samsung and Micron are becoming more strategic as high-bandwidth memory moves from supporting component to critical constraint.

This is where the optimism needs discipline. The next phase of the AI race will be limited by things semiconductor investors once treated as background: packaging, transformers, electricity, cooling, fabrication capacity, memory and financing. NVIDIA can see demand accelerating, but it cannot build an electrical grid, speed up every fab or remove geopolitical exposure from Taiwan and China. Hyperscaler capex therefore remains both the strongest evidence for the boom and its largest financial risk.

Still, the alignment looks more constructive than it did only weeks ago. Supply chains remain tight, but they are expanding; memory investment is rising, packaging capacity is being added, and the next generation of chips is arriving into an ecosystem better prepared than during the first Blackwell ramp. Wall Street is not abandoning the AI race. It is becoming more selective about which parts deserve capital.

That may be the healthiest development of all. The market is beginning to separate AI enthusiasm from AI economics, rewarding companies that can translate compute into utilisation, revenue and cash flow.

NVIDIA’s results suggest the infrastructure cycle still has considerable distance to run, but the next winners will be those capable of turning intelligence into a physical, financed and functioning industrial system.

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.