Amazon is moving to sell its Trainium AI chips externally, following Microsoft and Google into direct competition with Nvidia. While Trainium3 chips individually cannot outperform Nvidia's top-tier Blackwell GPUs, Amazon's system-level stacking strategy allows it to match Blackwell's performance at lower cost. This reflects a broader shift among major cloud providers and privacy-conscious markets seeking to reduce reliance on Nvidia's proprietary ecosystem, though Nvidia's software moat and supply-demand imbalance currently protect its near-term business.
Summaries like this, in your inbox every morning.
Sign up free →What happened
Amazon is reportedly preparing to sell its Trainium AI chips to external customers, following similar moves by Microsoft and Google. Amazon's Trainium3 chips can match the rack-scale performance of Nvidia's Blackwell systems at lower cost when stacked densely into UltraServers.
Why it matters
Major cloud providers and privacy-focused markets in Europe are seeking alternatives to Nvidia's ecosystem to reduce dependence on a single supplier and avoid storing data on American hyperscalers' servers. Companies already invested in Nvidia's software and infrastructure may face pressure to evaluate competing options, though switching costs remain high.
What to watch
Nvidia still maintains a strong hold through its proprietary CUDA software ecosystem and the fact that most AI models and frameworks are optimized to run on its GPUs. Current demand for Nvidia's data center GPUs continues to outstrip supply, but investors should monitor how the company's largest customers evolve into competitors.
Amazon began using its own Trainium AI chips in AWS four years ago and has strengthened its position with Trainium2 (2024) and Trainium3 (2025) launches. This reflects a deliberate strategy to reduce dependence on Nvidia, a pattern replicated by Microsoft, Alphabet's Google, and Meta—all of which developed proprietary AI chips for their own infrastructure. Google and Microsoft have already signaled plans to sell their chips to third-party customers, and Amazon's reported move to do the same represents an acceleration of this competitive shift.
The appeal of alternative chips extends beyond the hyperscalers' internal needs. Privacy-conscious regions, particularly Europe, face regulatory and practical pressure to build cloud infrastructure that does not rely on American technology giants' servers. Smaller enterprises that fear over-dependence on a single vendor similarly represent a growing market for alternatives. By offering system-level solutions through UltraServers—matching Nvidia's performance at lower cost—Amazon, Microsoft, and Google are attacking both the technical and economic barriers to switching.
Nvidia's position remains formidable despite these long-term headwinds. Its CUDA software ecosystem creates significant switching costs: most AI models, libraries, and frameworks are natively optimized for Nvidia GPUs, meaning customers who have already invested in that platform face friction in migrating to rivals. Additionally, supply constraints mean Nvidia's demand continues to exceed its available output. However, the article suggests that investors should monitor how Nvidia's largest customers are gradually transforming into competitors—a structural shift that, while unlikely to threaten near-term revenue, may reshape the market over time.
No discussion yet for this article
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
