Oracle’s GPU Fleet Ran Nearly Flat Out Last Quarter, Offering a Window Into AI Chip Demand

Oracle used its fiscal first-quarter 2027 earnings call, covering the period ended Aug. 31, to detail the scale of its artificial intelligence buildout — and the numbers underscore how heavily its data center hardware is being worked.

The company said it delivered 850 megawatts of AI capacity to customers during the quarter, hardware that houses more than 300,000 graphics processing units. Notably, Oracle reported that its GPU fleet operated at 97.9% utilization, meaning almost every chip in its fleet was actively serving customer workloads.

The vast majority of those processors come from NVIDIA. Oracle’s cluster designs for its largest AI deployments link together hundreds of thousands of Nvidia GPUs, and its most ambitious planned installations are designed to scale to as many as 800,000 units.

The disclosure offers a rare, quantified look at demand for Nvidia’s accelerators. While Nvidia itself did not report earnings this week, its largest cloud customers’ capacity commitments and utilization rates serve as a proxy for how strained supply of high-end AI silicon remains across the industry.

Market snapshot

Oracle shares closed at $158.78, up 2.9% from the prior close of $154.30, giving the software and infrastructure company a market capitalization of roughly $415.5 billion. Nvidia, whose shares slipped 0.73% to $223.67 from a prior close of $225.31, carries a market cap of approximately $5.56 trillion — a reminder of how much value investors have assigned to the demand environment Oracle’s figures illustrate.

Oracle’s cloud business has increasingly positioned itself as a builder of massive, GPU-dense data centers for AI training and inference, competing for hyperscale workloads alongside larger rivals. High utilization rates suggest capacity is being absorbed almost immediately upon delivery, though they also leave little headroom for maintenance, failures, or demand spikes without additional buildout.

The reported 850 megawatts of delivered capacity represents power consumed by AI infrastructure — a metric that has become a common yardstick for data center scale as power availability, rather than floor space, increasingly constrains expansion.

What to watch

• Nvidia’s upcoming quarterly earnings report and any commentary on supply constraints and data center revenue.
• Oracle’s next earnings release for updates on GPU deployment totals and capacity delivery targets, including progress toward its 800,000-GPU planned builds.
• Further disclosures on power procurement and data center construction timelines, which will shape how quickly Oracle can convert its order book into delivered capacity.

Source: original release