A few days ago we learned that OpenAI has closed the largest fundraising round in its history, totaling $110 billion.
If NVIDIA and SoftBank each invested $30 billion, the biggest contribution comes from AWS. The latter unlocked $50 billion. It will inject $15 billion to start and plans to award the remaining amount “in the coming months, subject to certain conditions.”
In parallel with this financial infusion, OpenAI promises to spend an additional $100 billion over the next eight years on compute resources at AWS*. This commitment complements a $38 billion agreement signed in November 2025.
The prospect of an runtime tailored for AWS
The alliance doesn’t stop at compute. AWS will also be the exclusive third-party cloud distributor for Frontier, the agentic orchestration layer OpenAI unveiled in early February.
The two companies also intend to launch, “in the coming months,” an agentic runtime built on OpenAI models and optimized for the AWS infra (native to Amazon Bedrock). They aren’t saying much more, other than this environment will preserve context (memory, identities, tools…), thereby avoiding manual orchestration.
The partnership also involves designing custom models for Amazon developers. In the backdrop, in-house LLMs that aren’t universally welcomed internally.
* AWS talks about consuming 2 GW of Trainium capacity. In November 2025, it referenced access to “hundreds of thousands” of GPUs. And a possibility of expansion to “tens of millions” of CPUs. The announcement mentioned UltraServer configurations equipped with GB200 and GB300.
Further reading:
The AI Factory, a grammar now legitimate at AWS?
OpenAI learns to think inference without NVIDIA
Perplexity bows to Microsoft Foundry… without letting go of AWS
How OpenAI shapes its development as compute evolves
From intuition to analysis, a taxonomy of LLM reasoning errors
