Amazon Web Services (AWS) announced on Tuesday that it will integrate key Nvidia technology into future generations of its artificial intelligence chips. The move is part of Amazon’s wider plan to attract major AI customers and strengthen its cloud computing services.
AWS said it will use Nvidia’s “NVLink Fusion” in an upcoming chip called Trainium4. While no release date was provided, the company highlighted that NVLink enables extremely fast communication between different types of chips. The technology is considered one of Nvidia’s most important innovations.
The announcement came during AWS’s annual cloud conference in Las Vegas, which brings together roughly 60,000 attendees.
Nvidia has been working to get more chipmakers to adopt NVLink. Intel, Qualcomm, and now AWS have all agreed to use the technology.
For AWS, NVLink will make it possible to build larger and faster AI servers. These systems will be able to coordinate more efficiently, which is essential for training large-scale AI models that rely on thousands of connected machines. As part of the partnership, AWS customers will also gain access to “AI Factories,” specialized AI infrastructure deployed inside their own data centers for faster performance.
Nvidia CEO Jensen Huang said the collaboration is designed to build the “compute fabric for the AI industrial revolution,” bringing advanced AI capabilities to companies around the world.
Separately, Amazon introduced new servers powered by a chip called Trainium3. These servers, available starting Tuesday, contain 144 chips each and offer more than four times the performance of AWS’s previous AI generation while using 40% less power. Dave Brown, AWS vice president of compute and machine learning, said AWS plans to compete with rivals like Nvidia by offering better price-to-performance benefits.
Brown said customers need chips that deliver high performance at the right price point, adding that this combination will help convince companies to adopt AWS’s hardware.
Amazon also rolled out updated versions of its AI models, known as Nova. The new Nova 2 model is faster and more responsive. It includes a multimodal version that can generate responses from images, text, speech, or video prompts. Another model, Sonic, can produce speech outputs in a natural, “human-like” way, according to AWS CEO Matt Garman.
Amazon has faced challenges gaining broad adoption for Nova compared to competitors like OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini. However, AWS still posted a 20% increase in sales last quarter, driven largely by its AI and cloud infrastructure services.
At the conference, Amazon also unveiled Nova Forge, a service that lets businesses build their own AI models using their proprietary data. Garman said the tool allows companies to create models that understand their information deeply without losing the original training data.
Amazon shares rose 0.9% to $235.98 by midday Tuesday.
