On November 3, 2025, U.S. AI leader OpenAI announced a seven‑year strategic agreement with Amazon Web Services (AWS) valued at $38 billion. According to an exclusive Reuters report, the agreement designates AWS as OpenAI's primary cloud platform for AI model training and inference, immediately providing NVIDIA GPU resources to support compute demands ranging from the Sora 2 video generator to GPT‑5–scale large language models. OpenAI CEO Sam Altman wrote on X: "AWS's global scale and innovation will accelerate our progress toward AGI." Amazon CEO Andy Jassy responded: "We welcome OpenAI into the AWS ecosystem to help bring AI applications to the masses." The collaboration marks OpenAI's move from sole reliance on Microsoft Azure to a diversified setup, potentially reducing Azure's share from 90% to 70%. TechCrunch analysts say the deal not only alleviates OpenAI's compute bottlenecks but also helps Amazon regain AI cloud market share and advance a global AI infrastructure "multi‑cloud era." In the AI arms race, the move highlights a "co‑opetition" logic among cloud giants while raising concerns within the Microsoft ecosystem and among regulators.
Background: The Inevitable Choice of OpenAI's "Multi‑Cloud" Strategy
OpenAI's close partnership with Microsoft began in 2019, with Microsoft investing a total of $13 billion and Azure becoming OpenAI's core cloud platform supporting GPT‑4o and Sora training. However, explosive AI demand in 2025 exposed OpenAI's compute constraints: the Stargate project requires 10 GW of compute and Azure supply was insufficient. CNBC reported that OpenAI had already turned to Oracle (a $30 billion compute agreement) and Google Cloud ($10 billion), with AWS becoming a fourth major partner. At the APEC summit (October 31), Altman hinted: "Single‑cloud risk is too great; multi‑cloud is the path to AGI." AWS is known for its global footprint (105 availability zones) and custom chips Trainium and Inferentia, with costs reportedly 20–30% lower than Azure.
Details of the agreement have not been fully disclosed, but The Guardian, citing informed sources, reported that OpenAI will train the Sora 3 video model and GPT‑5 on AWS, with initial investments undisclosed and full capacity expected in 2026. AWS Bedrock will be integrated with the OpenAI API to support enterprise‑level model fine‑tuning. Amazon is attracting OpenAI with EC2 UltraClusters (single‑cluster millions of GPUs), reducing dependence on Azure. Jassy said on an earnings call: "OpenAI is a strategic partner for AWS Bedrock, jointly developing multimodal AI." The deal includes immediate access to NVIDIA H100/B200 GPU clusters, with full capacity anticipated in 2026, supporting OpenAI's "equity plus supply" model.
Technical Details: Efficient Integration of AWS Trainium2 and OpenAI
The core of the collaboration is the integration of AWS's Trainium2 chips with OpenAI's models. Trainium2 is Amazon's in‑house AI training chip, delivering up to 2 petaflops per chip and costing roughly one‑third of an NVIDIA H100. OpenAI plans to use Trainium2 clusters to train GPT‑5, with expected speed increases of about 40% and energy use reductions of around 30%. WIRED reported that AWS Inferentia2 inference chips will optimize Sora 2 video generation, cutting latency by 50%. The Bedrock platform will add OpenAI model options and support hybrid deployments: training on Azure plus inference on AWS to lower costs by about 20%.
This integration stems from OpenAI's "multi‑cloud principle": Altman's July 2025 blog stated, "AGI requires global compute and avoidance of single‑point risk." AWS's global coverage (reported reach of 95%) and low pricing (Spot instances discounted by up to 50%) are key factors. Although Microsoft has invested $13 billion, OpenAI has already shifted 10% of its workload to Oracle, and the AWS agreement will cover an estimated 15%. The deal also includes AWS Nitro Enclaves isolated containers to support encrypted data training and mitigate data leak risks. AWS CEO Jassy emphasized that Trainium2's custom architecture is compatible with OpenAI's distributed training framework and supports parallelism for trillion‑parameter models.
Key technical highlights of the agreement:
- GPU clusters: Immediate access to hundreds of thousands of NVIDIA GB200/GB300 GPUs, expanding to the million‑GPU scale by 2026.
- Custom chips: Trainium2 for training plus Inferentia2 for inference, reducing both cost and power consumption.
- Security & privacy: Nitro Enclaves isolation and Bedrock data localization.
- Scalability: A seven‑year framework with support for extensions beyond 2027 and elastic Spot pricing.
- Integration: Bedrock seamless integration with the OpenAI API and enterprise‑ready fine‑tuning capabilities.
Market Impact: Amazon's Return to the AI Cloud Race and a Warning for Microsoft
For Amazon, the collaboration could reshape the AI cloud landscape. AWS held a 31% share of the cloud market in 2024 but only about 10% of the AI segment, trailing Azure's estimated 20% AI share. Jassy said the OpenAI partnership will drive 30% growth in Bedrock users and forecasted AI revenue of up to $100 billion by 2026. Amazon's stock rose 3.2% after the earnings call, increasing market value by $200 billion.
Microsoft will feel the impact: Azure accounted for an estimated 90% of OpenAI's cloud usage and could fall to 70% after this deal. CNBC analysis expects Microsoft's Azure cloud revenue to exceed $75 billion in 2025, but its AI share may decline by about 5%. Altman denied any betrayal of Microsoft, calling the move "multi‑cloud complementary." Nonetheless, Microsoft's shares fell about 1.5% after the earnings call.
Global effects: the multi‑cloud trend accelerates, benefiting Oracle, Google Cloud and AWS, and intensifying competition around AWS Bedrock. Forbes projects the AI cloud market could reach $500 billion by 2026, with the OpenAI partnership possibly accounting for around 10%. Enterprises stand to benefit from reduced risk and about 15% cost savings through hybrid cloud strategies.
Expert Views: Multi‑Cloud as a "Win‑Win" but Rising Regulatory Risks
Experts generally welcome the deal. Altman's post on X received widespread support: "With AWS on board, AGI is closer." Jassy framed the partnership as "AI democratization." Microsoft CTO Kevin Scott commented: "This is mutually beneficial; Azure remains core."
Risks include regulatory scrutiny: the EU's DMA may investigate AI cloud dominance and the U.S. FTC is watching for unfair competition. Privacy and data security remain concerns: AWS Bedrock's data localization features and OpenAI's compliance steps will be critical. Jassy reiterated that Trainium2's custom architecture is compatible with OpenAI's distributed training framework and supports parallelism for trillion‑parameter models.
Outlook: The Advent of a Multipolar AI Cloud Era
The OpenAI‑AWS agreement marks another milestone toward a multipolar AI cloud ecosystem. By 2026, OpenAI may emerge as a more "cloud‑neutral" player while AWS reasserts itself in the AI race. The tech ecosystem is shifting from a two‑player rivalry to a "multi‑cloud flourishing" model that could benefit developers worldwide, though regulators and cost pressures remain important considerations. The "cloud wars" in AI may well shape the future landscape.