Oracle delivered 850 MW of AI capacity — the fleet still ran 97.9% full

WIRE · HYPERSCALERS · Data as at 10 Sep 2026

What happened. On 10 September 2026 Oracle reported Q1 FY27 results: it delivered 850 MW of additional AI datacenter capacity and more than 300,000 GPUs to AI Cloud customers since end-Q4 — almost triple Q4 FY26 deliveries — while Cloud Infrastructure revenue rose 121% to $7.4 billion. The company said customer demand for AI Cloud training and inferencing still grows faster than supply; it booked more than $30 billion of new AI cloud contracts in the quarter and lifted remaining performance obligations to $664 billion. On the earnings call, co-CEO Clay Magouyrk said GPU utilization stayed at 97.9%, and renewals of mostly four-year-or-older GPUs repriced about 20% above prior contracts. (Oracle press release, 10 Sep 2026; Q1 FY27 earnings call.)

Why cash cares. On the hyperscalers layer, Oracle sits with the buyers who set capex and pull GPU, networking, power, and halls. Who gets paid while contracted backlog outruns installed megawatts is whoever can deliver capacity that customers accept — not only who booked the contract. A fleet that stays nearly full after a record quarter of adds means the short clock is still capacity delivery and attach, not a pause in AI cloud demand. Upstream chips and power stay pulled; designs waiting for halls wait.

What to watch. Whether Q2 FY27 capacity deliveries keep pace with the $664 billion RPO clock, and whether GPU renewals still clear at a premium once Vera Rubin systems start shipping (Oracle said first deliveries in Q2).

Read: Hyperscalers: The Capex-Setting Layer — and Hyperscaler Explained for how this layer sits on the map.

Get the Wire. Free daily brief on-site. Email digest at most twice a week. Subscribe free →

Source: Oracle Corporation press release, 10 Sep 2026 — company press; utilization and renewal premium from Q1 FY27 earnings call.