An in-depth look at AWS' underperformance in the GenAI era, as Microsoft Azure and Google Cloud gain ground, and how Anthropic could spark an AWS AI resurgence
Two-and-a-half years ago, we flagged a looming “cloud crisis” at AWS. Today, the evidence has mounted. X: @rohanpaul_ai and @dylan522p X: Rohan Paul / @rohanpaul_ai : AWS is betting heavily on its custom Trainium chips, with Anthropic as the anchor customer, to regain momentum in the AI cloud race. ~ A solid Semi Analysis report. AWS is building multi-gigawatt data centers packed with Trainium2 hardware, designed to give a better cost per [image] Dylan Patel / @dylan522p : AWS cloud revenue projections can actually be mapped if you follow they datacenter builds, accelerator purchases, and flow through to rental income of specific clients. Yippit can never do this.
Context & Ripple Effects
AWS had already positioned itself as a multi-model platform through early access to models from Anthropic and other AI providers. Its tighter alignment with Anthropic deepened when Amazon committed another $4 billion and Anthropic made AWS its primary training partner in the expanded AWS-Anthropic partnership.
The new analysis frames Trainium2 capacity as the test of whether that partnership can translate into a stronger AI-cloud position while Azure and Google Cloud gain momentum. AWS's recent revenue growth did not remove that strategic question, particularly after second-quarter operating income came in below estimates.
First-order effects
- AWS is concentrating its AI-infrastructure bet on Trainium2 data centers, with Anthropic as an anchor workload that can validate the chips' cost-performance proposition at scale.
- Anthropic becomes more tightly tied to AWS's custom-silicon roadmap, while AWS customers gain a clearer alternative to GPU-led AI infrastructure if Trainium delivers as intended.
Second-order effects
- Azure and Google Cloud face added pressure to defend AI-training and inference workloads on both model availability and infrastructure economics, rather than relying only on general cloud breadth.
- Large customers evaluating generative-AI deployments may gain more leverage to compare accelerator cost and performance across clouds; AWS's planned AI-agent marketplace initiative could provide another route to distribute workloads if the infrastructure gains traction.
Third-order effects
- If anchor-model providers increasingly shape cloud hardware roadmaps, AI-cloud competition could shift from generic compute capacity toward vertically integrated pairings of models, chips, and data centers.
- The outcome remains contingent on Trainium adoption and performance, but the pattern points to a more capital-intensive cloud market in which proprietary accelerators are a central differentiator rather than a side offering.
The trend: This is one data point in the AI infrastructure capital cycle, where cloud providers are using custom silicon and anchor AI partners to turn massive capacity buildouts into durable platform advantage.