OpenAI and AWS sign a seven-year deal in which OpenAI will pay $38B for AI compute, including training its models using Amazon's data centers and using its CPUs
Companies' first computing partnership is part of Amazon's efforts to grow its cloud business
Wall Street JournalBerber Jin
Context & Ripple Effects
This is OpenAI and AWS’s first computing partnership, converting Amazon’s cloud-growth push into a long-term customer commitment rather than a one-off infrastructure purchase. The arrangement was operational from the outset: OpenAI said it would begin running workloads on AWS immediately using large-scale Nvidia GPU capacity.
Related coverage shows the partnership later broadened beyond raw compute, with OpenAI models slated for availability through AWS. That makes the initial seven-year commitment a foundation for a deeper cloud and model-distribution relationship.
First-order effects
OpenAI secures seven years of AWS compute capacity for model training, including use of Amazon data centers and CPUs, while taking on a $38B spending commitment.
AWS gains a major committed AI-compute customer and immediate OpenAI workloads, supporting its effort to expand cloud demand.
Second-order effects
The deal gives AWS an incentive to provision and operate more AI-oriented capacity for a single large customer, while OpenAI’s workload placement makes AWS a more consequential part of its infrastructure mix.
The later move to offer OpenAI models through AWS suggests the compute relationship can create a distribution channel for AWS customers, not just infrastructure revenue.
Third-order effects
If such commitments become standard, frontier-model developers and cloud providers will be tied together by longer-duration capacity contracts, making access to compute more dependent on a small set of hyperscale operators.
The subsequent combination of investment, capacity commitments, and model availability points toward cloud partnerships that bundle financing, infrastructure, and distribution rather than treating them as separate markets.
The trend: AI compute is being commercialized through long-term cloud commitments that increasingly connect capacity supply, financing, and model distribution.
JUST ANNOUNCED: OpenAI will start using AWS GB200 and GB300 NVL72. Note that OpenAI will mostly be using NVIDIA ConnectX NICs on AWS and not be using AWS EFA NICs, which have subpar user experience. Also, OpenAI will not be using Sagemaker Hyperpod Managed K8 / SLURM which is [im…
Amazon deal represents only 6.5% of the aggregate value of these three OpenAI deals unveiled since September. That's tiny. Oracle: $300 billion over five years (WSJ) Microsoft: $250 billion in additional Azure orders Amazon: $38 billion over seven years
.@awscloud and @OpenAI just announced a multi-year, strategic partnership that provides AWS's world-class infrastructure for OpenAI's advanced AI workloads. The compute clusters are designed to support various workloads, from serving inference for ChatGPT to training next
New multi-year, strategic partnership with @OpenAI will provide our industry-leading infrastructure for them to run and scale ChatGPT inference, training, and agentic AI workloads. Allows OpenAI to leverage our unusual experience running large-scale AI infrastructure securely, [i…
Had this a few weeks ago but could not get a solid source beyond hearsay. The systemic risk is spreading! So many companies will be short changed — www.aboutamazon.com/news/aws/ aws...
The move to find other places to run that aren't Azure is downstream of OpenAI saying they wanted to see other clouds and Azure saying sure, I'll just stop building more special capacity for you, you know how to cook your own dinner, right babe? [embedded post]