SANTA CLARA, Calif. — Nvidia this week delivered one of the clearest financial signals yet that the generative-artificial-intelligence boom is becoming a large-scale infrastructure cycle. The chipmaker reported $7.19 billion in fiscal first-quarter revenue and projected roughly $11 billion for the current quarter, far above the pace implied by its recent results, as demand for accelerated computing and large-model training drives customers toward its data-center processors.

Nvidia’s May 24 earnings release showed record Data Center revenue of $4.28 billion, up 18% from the prior quarter and 14% from a year earlier. Overall revenue was still 13% below the year-earlier period because gaming and other markets have not fully recovered, but the company’s second-quarter outlook of $11.0 billion, plus or minus 2%, effectively reset expectations for how quickly AI demand could change its growth trajectory.

AI moves from software story to capital-spending cycle

The company’s quarterly SEC filing makes clear that Nvidia is no longer depending primarily on consumer graphics. Data Center has become its largest business, with customers spanning cloud providers, internet companies, enterprises and research institutions. Chief Executive Jensen Huang said the computing industry is undergoing two simultaneous transitions — accelerated computing and generative AI — and that Nvidia is increasing supply to meet surging demand for its H100, Grace Hopper, networking and software platforms.

The demand is tied to a fundamental technical constraint. Training and serving large language models requires enormous parallel-processing capacity and high-speed memory bandwidth. Conventional CPUs can perform these workloads, but at far lower efficiency than specialized accelerators. Nvidia has spent years building a software ecosystem around CUDA and an expanding portfolio of GPUs, interconnects and networking products, giving it a strong position as developers rush to deploy models inspired by systems such as ChatGPT.

Enterprise partnerships show where the spending is headed

The week before the earnings report, Nvidia and ServiceNow announced a generative-AI partnership to develop custom large language models for enterprise workflows. That agreement is important because it moves generative AI beyond public chatbots into internal business processes such as customer service, IT support and employee productivity.

On May 23, Dell and Nvidia unveiled Project Helix, a blueprint for companies that want to build and run generative-AI systems on their own infrastructure. The offering combines Dell servers and storage with Nvidia H100 GPUs, networking and AI Enterprise software, addressing organizations that want access to large-model capabilities without placing proprietary information entirely in public clouds.

Nvidia also announced a Microsoft integration that will bring Nvidia AI Enterprise software into Azure Machine Learning. The relationship gives developers a supported path to train, tune and deploy models using Microsoft’s cloud while relying on Nvidia’s frameworks and accelerators underneath.

The platform strategy extends beyond a single chip

Nvidia’s advantage is not limited to the H100. At its March GTC conference the company introduced four inference platforms aimed at different generative-AI workloads, including the H100 NVL for large language models, L4 for AI video and Grace Hopper for recommendation systems. The strategy is to sell a full stack: processors, networking, systems and software that developers can use from training through inference.

That matters because inference — the process of running a trained model to produce answers, images or recommendations — can eventually consume as much or more computing capacity than training. As generative-AI applications move from experiments to products used by millions of people, demand can shift from a handful of massive training runs to persistent, round-the-clock serving infrastructure.

An extraordinary forecast still carries execution risk

The $11 billion revenue outlook implies a dramatic sequential increase, and achieving it will depend on supply. Advanced GPUs rely on leading-edge semiconductor fabrication and sophisticated packaging, both of which have finite capacity. Nvidia has said it is working with suppliers to increase production, but the pace of demand creates the possibility that customers will face long lead times.

Competition is also intensifying. Advanced Micro Devices is preparing new AI accelerators, major cloud providers are developing proprietary chips, and startups are attacking specialized portions of the machine-learning market. Customers that currently depend heavily on Nvidia have strong incentives to diversify if capacity remains constrained or prices rise.

For now, however, the company’s quarter suggests that generative AI is creating a spending cycle large enough to overwhelm weakness elsewhere in technology. Nvidia entered this year as the leading supplier of high-end AI accelerators. It is leaving May with a financial forecast that implies the market for those systems is expanding much faster than even many investors expected only weeks ago.