Four years ago, Amazon (AMZN -0.15%) started using its own Trainium AI chips in its cloud infrastructure platform, Amazon Web Services (AWS). Those first-party chips became even more powerful with the launches of the Trainium2 in 2024 and Trainium3 in 2025. That expansion indicated that Amazon wanted to reduce its dependence on Nvidia (NVDA +0.84%), which still provides the majority of its data center GPUs.
Several of Nvidia's other top customers -- including Microsoft (MSFT -2.04%), Alphabet's (GOOG -1.05%) (GOOGL -1.11%) Google, and Meta -- also produced their own AI chips for the same reason. Google and Microsoft even plan to sell their own chips to third-party customers that want to break free from Nvidia's sticky ecosystem.
Image source: Getty Images.
That's why it wasn't surprising when recent reports suggested that Amazon would hop aboard the bandwagon and start selling its Trainium chips to external customers. Could this seismic shift shake up Nvidia's booming data center business?
Nvidia faces long-term threats
Amazon's Trainium3 chips can't compete against Nvidia's top-tier Blackwell GPUs on their own. But by densely stacking 144 Trainium3 chips into its UltraServers, Amazon can actually match the rack-scale performance of Nvidia's Blackwell systems at a much lower cost. Microsoft and Google are utilizing that same "system-level stacking" strategy to challenge Nvidia's chips.
Many privacy-oriented markets, such as Europe, want to expand their cloud infrastructure without storing their data on servers operated by American hyperscalers. To solve that, they'll likely purchase more third-party chips from Amazon, Microsoft, and Google to build their own cloud platforms. Other large companies that don't want to rely on those tech giants or become too dependent on Nvidia's chips will likely follow the same playbook.
NASDAQ: NVDA
Key Data Points
But Nvidia still has a wide moat
Amazon's sales of third-party AI chips would certainly represent a long-term challenge for Nvidia, but it probably won't meaningfully impact its near-term sales.
Nvidia still locks in its customers with its proprietary software ecosystem, CUDA, and most AI models, libraries, and frameworks are natively optimized to run on its industry-standard GPUs. Many companies that have already invested in Nvidia's ecosystem won't eagerly sever those ties to buy new chips from Amazon, Microsoft, or Google.
For now, Nvidia's investors shouldn't worry too much because the demand for its data center GPUs is still easily outstripping its supply. However, they should still keep a close eye on how its biggest customers are gradually evolving into formidable competitors.
