Oracle's Nvidia Chips Hit 97.9% Utilization as Old GPUs Reprice 20% Higher

Oracle's disclosure lands at a moment when Nvidia investors have had to take demand on faith. The company spent Thursday evening's earnings call putting numbers on something usually inferred: 850 megawatts delivered, more than 300,000 GPUs, and a fleet running 97.9% full. The 850 megawatts was almost triple the prior quarter's total and 73% of everything Oracle delivered in fiscal 2026, and revenue followed the hardware, with cloud infrastructure revenue up 121% year over year to $7.4 billion.

The more striking detail is what happened to aging hardware. Computing gear is supposed to get cheaper as it ages, and these GPUs predate Nvidia's Blackwell generation entirely. Instead, Oracle's co-CEO Clay Magouyrk said renewals and resales carried a 20% premium, with the majority of those chips four years or older. Prices tend to move that way when new hardware is hard to get.

The caveats are real. Oracle's utilization figure covers one operator's fleet, not the industry's, and some of that fullness reflects the $664 billion backlog of contracts customers signed quarters ago. Yet Nvidia's own disclosures point the same way: revenue rose 106% year over year to $96.2 billion, and CFO Colette Kress said its compute was fully utilized across every cloud it serves. The argument that Nvidia is limited by manufacturing capacity rather than appetite hinges on whether old GPUs keep repricing higher; a future quarter showing discounts on 4-year-old chips would be an early sign the shortage is easing.