Amazon AWS Officially Launches Trainium3 AI Chip: A Cheaper Alternative to NVIDIA GPUs

[{"selector":"#anim-925cc462-984c-4914-88f3-e45b2b7b9974","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-2bdbab34-322a-4e16-98a2-4d1dabe71b73","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] Amazon AWS Officially Launches Trainium3 AI Chip: A Cheaper Alternative to NVIDIA GPUs

[{"selector":"#anim-d042915c-dedf-4651-a5d8-ccccb5d99caa","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-22d139d9-257f-4f13-9748-02c25ccbb628","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] Key Highlights – Unveiled at AWS re:Invent 2025 – Built on 3nm architecture – Offers 4.4× higher performance and 4× more memory bandwidth – 4× better performance-per-watt vs Trainium2

[{"selector":"#anim-c857a092-25ff-445b-9ab3-13bb95ada1cb","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-d889be6e-e6bf-43f2-a608-5e4f48878d29","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] Performance & Scalability – Each UltraServer supports 144 chips (362 FP8 PFLOPS) – 1 million+ chips can link in UltraClusters (10× previous gen) – 50% lower training & inference costs than GPU clusters – 2× computing performance & 40% more energy efficiency

[{"selector":"#anim-b5c68b79-fe37-466e-80bf-dfcd6a471742","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-14b6f1a2-6a84-4c04-80c7-3833f41f4243","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] Adoption & Benefits – Used by Anthropic , LLM Karakuri , Decart , SplashMusic – Users report lower inference costs – Designed to deliver faster AI training , better efficiency , and lower operational cost – Positioned as a strong, cost-effective NVIDIA GPU alternative