Key Takeaways:
- AWS launched OpenAI's GPT-5.6 Sol, Terra and Luna on Bedrock on July 24
- The OpenAI-AWS partnership commits $38 billion in cloud spending through 2027
- Amazon is winding down most in-house Nova models to focus on infrastructure
뒤로
AWS brings OpenAI's GPT-5.6 Sol to Bedrock in 3-model lineup
AWS is opening its enterprise cloud to OpenAI's frontier models, marking a strategic shift from building its own AI to hosting the competition.
Amazon Web Services made OpenAI's GPT-5.6 Sol, Terra and Luna available on Bedrock on July 24, deepening a partnership that has already committed $38 billion in cloud spending through 2027.
"This gives enterprise customers the frontier models they want inside the AWS security perimeter," an AWS spokesperson said, noting that usage runs through Bedrock's IAM, PrivateLink and guardrails.
The three-model lineup includes Sol, the flagship frontier model that set a new state of the art on Terminal-Bench 2.1 for command-line workflows; Terra, designed for daily work tasks; and Luna, a more affordable variant. They join GPT-5.5 and GPT-5.4, which became generally available on Bedrock in June, alongside the Codex coding agent used by more than four million people weekly.
The launch comes as Amazon winds down most of its in-house Nova models, concentrating resources on a single frontier effort expected at re:Invent this fall. AWS is betting its infrastructure layer — Bedrock, custom Trainium silicon, and a $100 billion commitment from Anthropic — will matter more than owning the best model.
A $38 Billion Cloud Deal Takes Shape
The OpenAI-AWS partnership, announced in November 2025, runs OpenAI workloads on Amazon EC2 UltraServers with hundreds of thousands of Nvidia GB200 and GB300 GPUs, scalable to tens of millions of CPUs. Full capacity is slated by the end of 2026, with an option to expand into 2027. The product layer followed in April 2026 with three OpenAI offerings in limited preview on Bedrock.
For AWS, the calculus is straightforward: enterprise customers want to build with frontier models, and if Amazon cannot supply the best one itself, it will host whoever can. The company's own Nova Premier, Omni, Reel and Canvas models have been deprecated, with employees describing their status internally as "KTLO" — keep the lights on.
Anthropic's $100 Billion Bet Runs in Parallel
Amazon's entanglement with Anthropic runs even deeper. On April 20, 2026, the companies announced an expanded partnership: Amazon invested another $5 billion, bringing its cumulative Anthropic investment to roughly $13 billion, with up to $20 billion in milestone-based capital available. In return, Anthropic committed to spending more than $100 billion on AWS technologies over a decade and securing up to 5 gigawatts of Trainium capacity.
Project Rainier, the jointly operated Trainium cluster, has been fully operational since October 2025. Anthropic uses more than one million Trainium2 chips there to train and serve Claude. More than 100,000 customers run Claude models on Bedrock, making it one of the most-used model families on the platform.
Amazon shares trade at roughly 22x forward earnings. The company plans roughly $200 billion in capital expenditure for 2026, with AI demand as the primary driver. By hosting both OpenAI and Anthropic on Bedrock, AWS captures AI workloads regardless of which lab wins the model race — a hedge that could protect its 31% cloud market share against Microsoft Azure's OpenAI advantage and Google Cloud's Gemini push. Q2 results are due July 30.
This article is for informational purposes only and does not constitute investment advice.
