As the push by global Big Tech giants to develop in-house artificial intelligence (AI) chips heats up, Amazon has joined the competition.

Amazon Web Services (AWS), the largest cloud provider in the world, unveiled its in-house chip ‘Trainium 3’, featuring improved power efficiency and performance, at its annual conference ‘AWS re:Invent 2025’ in Las Vegas, United States, on the 2nd (local time).

Trainium 3 quadruples computing performance compared with the previous generation, Trainium 2, while reducing energy consumption by 40%. AWS explained that using this product can cut AI model training and operations costs by up to 50% versus comparable Nvidia graphics processing units (GPUs), and can also significantly shorten training time. It also emphasized that many customers, including Anthropic, have used the chip and system to sharply reduce inference costs.

AWS added that development of the successor AI chip, Trainium 4, has already begun and that it offers more than three times the performance of the third generation. The chip is expected to support the Nvidia interconnect technology ‘NVLink’, a move seen as aiming to improve customer convenience by reducing dependence on Nvidia while recognizing the strong preference for Nvidia GPUs in the cloud market.

Attempts by Big Tech companies to surpass Nvidia, the dominant force in the AI chip market, have intensified recently. In particular, the TPU (Tensor Processing Unit) released by Google last month is being evaluated as a potential threat to the position Nvidia has built with GPUs. The TPU was used in the development of the latest Google AI model, Gemini 3.0, which has been praised for record-high performance.

The rapid pursuit by Google is spurring several Big Tech companies beyond Nvidia. According to the Wall Street Journal (WSJ), OpenAI CEO Sam Altman recently issued an internal memo declaring a ‘code red’ and ordered the company to focus all-out on improving the quality of ChatGPT. This is interpreted as concentrating companywide resources on its flagship chatbot instead of new businesses such as healthcare and advertising, in order to respond to the rapid user gains by Gemini.