OpenAI is making one of the boldest infrastructure bets in tech, spreading more than $600 billion across Amazon Web Services (AWS), Oracle, and Microsoft to secure the compute needed to power the next generation of artificial intelligence.
OpenAI’s multi-cloud turn
Following the end of its exclusive cloud-computing partnership with Microsoft, OpenAI has reallocated spending across multiple hyperscalers: an estimated $250 billion back to Microsoft, $300 billion to Oracle Cloud, and a new $38 billion multi-year pact with Amazon Web Services. While the AWS deal is the smallest, it’s a deliberate diversification strategy to reduce concentration risk as demand for high-performance compute surges.
Securing scarce compute
The AWS agreement includes access to hundreds of thousands of NVIDIA GPUs—among them the latest GB200 and GB300 series—and the ability to tap tens of millions of CPUs. This isn’t just about training tomorrow’s frontier models; it’s about keeping today’s massive inference workloads for ChatGPT running at global scale.
“Scaling frontier AI requires massive, reliable compute.” — Sam Altman, OpenAI co-founder and CEO
Hyperscalers feel the heat
Although AWS remains the largest cloud provider by market share, rivals have posted faster cloud-revenue growth by capturing new AI workloads. OpenAI’s deal is both a validation of AWS’s large-scale AI capabilities and a bid to anchor a marquee AI customer. AWS says it can run clusters exceeding 500,000 chips and is building purpose-built architectures using EC2 UltraServers for ultra-low-latency GPU networking—critical for large-scale training.
“The breadth and immediate availability of optimised compute demonstrates why AWS is uniquely positioned to support OpenAI’s vast AI workloads,” said Matt Garman, CEO of AWS.
Reality check on timelines
“Immediate” is relative in the world of silicon. The full capacity from OpenAI’s latest cloud AI deals won’t be fully deployed until the end of 2026, with options to expand into 2027. The takeaway: the hardware supply chain is complex, capital-intensive, and locked into multi-year schedules that outlast typical IT planning cycles.
What enterprise leaders should do now
- Build vs. buy is settled: OpenAI is spending hundreds of billions on rented hardware. Most organizations should leverage managed AI platforms like Amazon Bedrock, Google Vertex AI, or IBM watsonx, where hyperscalers absorb infrastructure risk.
- Avoid single-cloud lock-in: OpenAI’s pivot underscores the need to mitigate concentration risk. For CIOs, relying on one provider for compute that underpins core processes is increasingly a gamble.
- Treat AI as capital planning: AI budgets are no longer variable IT expenses. Securing compute is a long-term commitment—closer to building a factory or data center than provisioning servers.
Bottom line
OpenAI’s $600 billion cloud bet is reshaping the AI infrastructure map. The future of AI isn’t just algorithms—it’s access to massive, reliable compute, and that demands thinking in decades, not quarters.
