The three major US cloud providers — Amazon Web Services, Microsoft Azure, and Google Cloud — are collectively spending over $300 billion on AI infrastructure through 2026. Custom chips, purpose-built data centers, liquid cooling systems, undersea cables — the scale of investment is staggering. For US businesses trying to pick a cloud AI platform, understanding who's winning matters.
The GPU War Is Over (NVIDIA Won)
All three hyperscalers continue to rely heavily on NVIDIA GPUs — H100s and the newer H200s — as their primary AI compute substrate. NVIDIA's CUDA ecosystem, now 18 years old and deeply embedded in AI research workflows, proved impossible to dislodge. Google's TPUs and Amazon's Trainium chips are real and used in production at scale, but they serve specific niches rather than replacing NVIDIA's general-purpose dominance.
What this means for you: GPU availability is tighter on AWS and Azure than Google Cloud right now, because Microsoft's OpenAI partnership consumed a disproportionate share of their H100 allocation through 2025. If you need H100s on-demand, Google Cloud has had consistently better availability.
AWS: The Infrastructure Incumbent
AWS remains the largest cloud provider by revenue and the most widely used for general cloud workloads. Its AI story centers on Bedrock — a managed API service that lets customers access models from Anthropic, Meta, Cohere, and others without managing infrastructure. Bedrock has gained strong traction with enterprise customers who want AI capability with the security and compliance controls they already trust from AWS.
Where AWS lags: its proprietary AI model story is weak. Amazon's Titan models are competitive but not leaders. AWS is essentially a neutral aggregator, which is a reasonable position but doesn't give it a differentiating narrative the way Microsoft's OpenAI partnership does.
Azure: The Enterprise AI Leader
Microsoft's $13 billion investment in OpenAI has paid off in one key way: enterprise sales. When a Fortune 500 company's IT department hears "GPT-4 on Azure with our existing security and compliance controls," it's a compelling pitch. Azure OpenAI Service is the fastest-growing product in Microsoft's history, and it's pulling companies deeper into the Azure ecosystem.
Azure's Copilot integrations across Microsoft 365, GitHub, and Dynamics 365 create a sticky ecosystem play. Companies already in the Microsoft stack often choose Azure for AI simply because integration is lower-friction. The risk: over-reliance on OpenAI. If that relationship changes, Azure's AI differentiation weakens.
Google Cloud: The AI Research Leader
Google invented the transformer architecture that powers all modern LLMs. DeepMind, Google Brain, and Google Research remain the world's most prolific AI research organizations. Vertex AI, Google's managed ML platform, is technically the most sophisticated of the three hyperscaler offerings. And Gemini Ultra, Google's frontier model, is genuinely competitive with GPT-5 on most benchmarks.
Where Google Cloud struggles: enterprise sales and trust. Google has a history of killing products (RIP Google+, Stadia, and dozens more), which makes enterprise IT departments nervous about building on Google infrastructure. Google Cloud is also smaller than AWS and Azure, meaning its support organization and regional footprint are thinner in some markets.
Who Should You Choose?
Already on AWS? Use Bedrock. The path of least resistance for access to frontier models with your existing security infrastructure. Already on Azure? Azure OpenAI Service and Copilot integrations make the choice easy. Already on Google Cloud? Vertex AI and Gemini are genuinely excellent, and you'll likely get better GPU availability. Starting fresh? Evaluate on specific use case, regional data center availability, and support quality — the AI capability difference between the three is now marginal for most workloads.