$200 billion. That's the capital expenditure figure Andy Jassy is defending in Amazon's 2026 annual shareholder letter - and he's not doing it quietly.
The letter reads less like a traditional investor update and more like a competitive manifesto. Jassy calls out Nvidia, Intel, Starlink, and others by name, positioning Amazon's custom chip development (the Trainium and Inferentia AI accelerators), AWS data center infrastructure, and Project Kuiper satellite internet as alternatives worth taking seriously.
On AI chips: Jassy argues Amazon's custom silicon can deliver better performance per dollar than Nvidia's GPUs for many workloads. That's a pointed claim. Nvidia's H100 and B200 chips have been the default infrastructure for AI model training, commanding premium prices partly because nothing else has been close. Amazon is arguing that's changing.
On connectivity: Project Kuiper, Amazon's satellite internet effort, gets positioned as a direct competitor to Starlink for enterprise customers and markets without reliable broadband. Jassy made clear this is not a hobby project.
The $200 billion in planned spending covers AI infrastructure, data centers, and logistics. Critics have questioned whether the scale is justified, particularly as more efficient AI architectures mean companies need less compute per task. Jassy's counter: efficiency doesn't shrink overall demand, it expands the number of use cases that become economically viable to run. More applications get built when the cost per query drops.
For businesses evaluating cloud providers, the letter signals Amazon is in a full infrastructure arms race with Microsoft Azure and Google Cloud and has no intention of pulling back. For the AI chip market, Jassy's directness about Nvidia is a sign Amazon sees its own silicon as genuinely competitive - not just a cost-cutting backup for price-sensitive workloads.