Amazon's Custom AI Silicon Gains Ground as Nvidia's Dominance Faces Internal Competition

Amazon is rapidly scaling its custom AI chip ecosystem, challenging the near-monopoly held by Nvidia in the data center market. According to estimates from Epoch AI, Amazon's share of cumulative AI chip shipments rose from 1.4% in the first quarter of 2024 to 7.5% by the fourth quarter of 2025. This shift reflects a broader trend among major hyperscalers—including Alphabet, Microsoft, and Meta—to develop proprietary accelerators to reduce reliance on general-purpose GPUs and lower operational costs. For Amazon, this transition is central to the growth of Amazon Web Services (AWS), where custom silicon is used to optimize both training and inference workloads. CEO Andy Jassy reported that the chips business achieved an annual run rate exceeding $20 billion in the first quarter. Jassy noted that if the operation were a standalone entity selling to other cloud providers, it could potentially generate $50 billion in revenue. For context, Nvidia reported $216 billion in revenue last year, while AMD recorded $35 billion. The demand for specialized hardware remains aggressive; Trainium3 capacity for 2026 was nearly exhausted by the end of March. Additionally, Amazon's custom CPUs are now utilized by 98% of its top 1,000 customers, signaling a deep integration of proprietary silicon across its cloud infrastructure. While Nvidia remains the market leader due to its CUDA software and networking ecosystem, the rise of custom accelerators represents a structural shift in how AI compute is deployed.