Amazon CEO Andy Jassy used the company’s 2026 annual shareholder letter to address multiple competitors while defending Amazon Web Services’ (AWS) decision to spend $200 billion in capex. As reported by TechCrunch, the letter references competitors across several areas, but the primary technology focus is AWS’s push for its own AI accelerators, particularly Trainium chips, and a shift toward reducing exclusive reliance on NVIDIA hardware.
AI Silicon Becomes Central to AWS’s Competitive Strategy
In his letter, Jassy takes a measured approach rather than issuing direct challenges. For example, when discussing NVIDIA, he writes that Amazon “have a strong partnership with NVIDIA, will always have customers who choose to run NVIDIA” and that AWS “will always support these chips in its cloud.” This framing signals an ongoing support strategy: AWS can offer infrastructure that works with NVIDIA GPUs while simultaneously optimizing for workloads that run on Amazon-designed accelerators.
At the same time, Jassy argues that market conditions for AI infrastructure are changing. According to TechCrunch, he states: “Virtually all AI thus far has been done on NVIDIA chips, but a new shift has started.” This suggests that AWS sees increasing willingness among customers to evaluate alternative compute stacks, including Amazon’s own chips.
Jassy attributes this shift to economics. AWS customers, he says, “want better price-performance.” In infrastructure terms, price-performance typically refers to the cost per unit of useful work (training or inference), which depends on chip performance, power efficiency, and how easily models and software pipelines can be deployed.
Trainium Demand and Capacity Constraints
The most concrete signal in Jassy’s letter concerns demand for Amazon’s current and next-generation Trainium chips. According to TechCrunch, Jassy states that capacity for Trainium3 is “nearly sold out.” He also notes that capacity for Trainium4 is “nearly sold out” even though Trainium4 is “still 18 months away from being available.“
These statements indicate demand extending far ahead of delivery, suggesting that customers are planning AI workloads with Amazon’s roadmap in mind. The timing difference—18 months—points to forward-looking procurement rather than reactive capacity planning.
Jassy also quantifies the business scale of the Trainium line. According to TechCrunch, Trainium has reached a “$20 billion annual revenue run rate.” He adds a hypothetical comparison: “if Amazon were a chipmaker that sold its wares to others, it would be at $50 billion ARR.” This suggests AWS is positioning Trainium as a credible competitive product category based on demand signals, though the $50 billion figure represents a hypothetical scenario rather than actual revenue.
Positioning Against NVIDIA
Jassy’s letter acknowledges NVIDIA’s current scale while suggesting momentum is shifting. According to TechCrunch, Jassy presents Trainium as a competitive alternative, while noting that NVIDIA’s financial results remain substantially larger: “Nvidia did $215.9 billion in actual revenue last year.“
For infrastructure planning purposes, the competitive narrative appears to center on multi-vendor deployment rather than outright replacement of NVIDIA. Jassy’s emphasis that AWS will “always support” NVIDIA chips indicates that, in AWS’s model, Trainium competes on workload economics and system-level integration rather than requiring customers to discontinue NVIDIA use entirely. This approach can influence how model training and inference pipelines are planned, with customers potentially choosing accelerators based on cost targets, availability, and performance characteristics.
Capex Investment and AI Infrastructure Strategy
Jassy’s $200 billion capex commitment appears connected to the AI silicon strategy through the Trainium demand and capacity constraints. While TechCrunch does not detail how capex maps to data center buildouts, networking, or semiconductor supply, the letter links the investment narrative to the accelerator roadmap by highlighting Trainium capacity constraints.
From an infrastructure perspective, this connection suggests that AWS’s AI hardware roadmap is intertwined with its broader infrastructure expansion. If Trainium capacity is “nearly sold out” for both current and future generations, compute supply—both silicon and supporting systems—could become a bottleneck that capex addresses. AWS appears to be treating AI infrastructure supply as a strategic priority.
Jassy’s letter references a broader set of competitors beyond NVIDIA and Intel, as TechCrunch notes he addresses “Nvidia, Intel, Starlink, more.” However, the primary technology focus remains on the AI chip narrative and Trainium’s capacity story.
Source: TechCrunch