Every quarter, Wall Street dissects Amazon’s earnings, focusing on the colossal AWS cloud business. But a multi-billion-dollar giant is hiding in plain sight, and its performance may be the single best indicator of Amazon's future in AI.
The Most Important Business You've Never Heard Of
When people think
of AI chips, they think of Nvidia. But Amazon has been quietly building a powerful alternative for over a decade through its Annapurna Labs acquisition. The result is two families of custom silicon: Graviton for general computing and, more critically, Trainium for artificial intelligence. Trainium chips are purpose-built to do one thing: power the intensive calculations needed for training and running massive AI models, like those from Anthropic and OpenAI. The goal is simple: deliver performance that rivals the best in the market at a significantly lower cost. This isn't a side project; it's a strategic weapon. On its latest earnings call, Amazon revealed its custom chip business is growing by triple-digit percentages, with revenue commitments for Trainium soaring more than 20-fold year-over-year. This is not a rounding error; it's a behemoth in the making.
It's the AWS Playbook All Over Again
Remember when AWS was just an internal tool to support Amazon's retail operations? It was a cost center before it became the company's profit engine. The strategy with Trainium feels remarkably similar. Initially developed to lower the staggering costs of running AI on third-party hardware, Trainium gives AWS a massive structural advantage. It insulates the company from supply chain shocks and allows it to capture margins that would otherwise go to chip vendors like Nvidia. Now, customer demand is so strong that Amazon plans to begin selling the chips externally in late 2027, a move echoing how AWS was opened to the public. This transition from an internal cost-saver to a public-facing product is a classic Amazon move, and investors need a clear lens to track its progress.
The Real Battleground in the AI Cloud War
The fight for AI supremacy between Amazon, Microsoft, and Google isn't just about software—it's about the fundamental hardware. Google has its Tensor Processing Units (TPUs), and Microsoft is developing its own custom silicon. By burying Trainium's performance inside the nearly $50 billion quarterly revenue of AWS, Amazon is obscuring its true strength in this arms race. We know customers are flocking to Trainium for its price-performance, with some reporting cost savings of up to 50% over alternatives for training and inference. Major players like Anthropic have built huge AI supercomputers with nearly a million Trainium chips. Without separate financial disclosure, however, analysts and investors are left guessing. Is this a $25 billion business, as some reports suggest, or something even larger? We have no way to properly model its growth or its impact on AWS margins.
What Gets Measured Gets Managed
Right now, Trainium's results are blended into the broader AWS figures. While CEO Andy Jassy provides tantalizing clues on earnings calls—like customers saving over $2 billion in a single quarter by using custom silicon—these anecdotes aren't a substitute for hard numbers. Analysts are left to piece together the puzzle from context, a process that invites speculation when precision is needed. Separating the custom silicon business would force a new level of discipline and clarity. It would allow investors to value this high-growth segment independently, much like they do with AWS itself. It would reveal the true ROI on the massive capital expenditures—pegged at $220 billion for 2026—that Amazon is pouring into its AI infrastructure. More importantly, it would prove whether Amazon's most critical strategic bet of the last decade is paying off.
