Larry Ellison says Oracle will spend “billions” on Nvidia GPUs and “three times” that on Ampere and AMD CPUs in 2023, as the company expands its cloud services
Stephen Nellis / Reuters :
Context & Ripple Effects
Oracle’s cloud push began as a bid to challenge AWS, and its compute catalog had already added Nvidia A100 accelerators and Arm-based instances. This spending plan turns that product direction into a large procurement commitment spanning both AI accelerators and server CPUs.
The commitment is an early step in an expansion that later showed up in rapidly growing cloud-infrastructure revenue, tying Oracle’s cloud strategy more directly to its ability to secure and deploy compute.
First-order effects
- Oracle commits substantial 2023 infrastructure spending to expand its cloud capacity, with CPU outlays described as roughly three times its Nvidia GPU spending.
- Nvidia gains a major cloud-provider GPU buyer, while AMD and Ampere stand to benefit from Oracle’s much larger planned CPU purchases.
Second-order effects
- A mixed GPU-and-CPU buying plan makes Oracle’s cloud buildout dependent on several chip suppliers rather than a single accelerator vendor, broadening the supplier base behind its service expansion.
- As Oracle adds capacity, its cloud offerings can compete more directly for workloads that require both accelerated compute and conventional server resources; the later cloud-infrastructure growth report indicates that this segment became increasingly material.
Third-order effects
- If such commitments persist, cloud competition increasingly turns on long-duration access to power, data-center capacity, and semiconductor supply—not just software features.
- Oracle’s later reported commitments to exceptionally large data centers suggest this procurement model can evolve into a capital-intensive, financing-sensitive infrastructure strategy, though this 2023 announcement alone does not establish its eventual scale.
The trend: This is one data point in the AI infrastructure supercycle, in which cloud providers pair accelerator purchases with broad CPU and data-center expansion to build deployable capacity.