The $38B Plot Twist: OpenAI Just Bet Big on AWS

This Week in Cloud — November 6, 2025

Welcome back to The Cloud Cover, your essential guide to navigating the shifting landscape of cloud and AI infrastructure. This week, OpenAI surprises everyone with a $38B pivot towards AWS, Microsoft doubles down on sovereignty and GPU supply, and Google turns “agents” into a native part of the data stack. The cloud wars are getting physical — let’s dive in.

The defining story of the week is OpenAI’s stunning 7-year, $38 billion strategic partnership with Amazon Web Services. This is a monumental development, primarily because OpenAI has long been the flagship AI partner for Microsoft, relying on Azure as its "sole source of computing power" for years.

This move signals that the primary bottleneck for AI dominance is no longer cloud allegiance but access to raw compute. OpenAI's demand for next-generation NVIDIA GPUs (specifically the GB200 and GB300 models) is so voracious that no single provider—not even Microsoft—can satisfy it. This $38 billion deal is an aggressive diversification to de-risk its most critical supply chain.

For AWS, this is a brilliant counter-move. After being perceived as trailing in the initial GenAI narrative, AWS has secured a massive contract to host the world's most prominent AI workloads. As AWS CEO Matt Garman noted, it's a "powerful reminder" of why organizations trust AWS for massive scale—a direct shot at Microsoft, implying it couldn't handle OpenAI's needs alone.

One crucial clarification: this is an infrastructure deal. It's about AWS providing EC2 UltraServers to run OpenAI's core workloads. This does not mean OpenAI's models are coming to Amazon Bedrock. AWS is continuing to build Bedrock around its own key partners, like Anthropic, which is using AWS's custom-built "Project Rainier" supercomputer.

🔍 The Rundown

📈 Trending Now: The Great Cloud Buildout Rolls On

This week, the "cloud" is looking decidedly physical. The AI arms race isn't just about who has the best model or agents; it's a brutal competition for a scarce physical supply chain: next-gen GPUs, power, and low-latency networking.

We're seeing every provider make massive, tangible infrastructure plays. AWS is vertically integrating, building its own custom-silicon AI supercomputer ("Project Rainier") for Anthropic while simultaneously laying its own private transatlantic subsea cable ("Fastnet") to own the network from end to end.

Microsoft, needing capacity now, is simply buying it, inking a $9.7 billion deal with IREN and another with Lambda just to secure data center space and GPUs. At the same time, it's "weaponizing sovereignty" by building specialized on-prem Azure Local hardware and new European regions (Austria and Belgium) to create regulatory moats that competitors can't easily cross. This is all about data gravity and data residency.

The game is no longer just about elastic, on-demand software. The winners will be those who can physically build—or buy—the most compute, power, and private networking, and place it in the right sovereign locations, fastest. The only question remaining is whether demand for this infrastructure can continue to keep pace.

📅 Event Radar

👋 Until Next Week

The buzzword of the week (or maybe the year), echoed by every single provider, was "agent." We saw GCP's "Data Engineering Agent," Oracle's "Agentic Finance," the Microsoft-SAP "agent-to-agent" integration, and the $38B AWS deal to power "agentic workloads."

We are clearly moving from Phase 2 of AI (calling a model via API) to Phase 3 (deploying and orchestrating autonomous agents). The big question for 2026 is no longer "who has the best model" but "who has the best platform for building, hosting, and orchestrating these agents."

It's a busy run-up to the end of the year. Stay tuned.

Do you enjoy these emails? Your friends and colleagues might, too! Help us grow the cloud community by sharing the newsletter with others.