Amazon’s cloud computing division announced on Tuesday that it will provide free computing power to researchers interested in using its custom artificial intelligence chips, seeking to challenge Nvidia’s dominance in the market. AWS revealed it will offer $110 million in credits for access to its cloud data centers, allowing researchers to utilize Trainium, its chip designed for developing AI models that compete with Nvidia’s offerings. Additionally, AWS is setting up a research cluster of up to 40,000 Trainium chips, which will be available to teams and students through self-managed reservations.
AWS said the Build on Trainium program would help researchers develop AI applications and build new technologies in partnership with academic institutions, scientists, and startups. It’s also putting up grants of up to $11 million in Trainium credits for universities with strategic partnerships and several individual grants of up to $500,000 for the broader AI research community. The company says it’s trying to address the difficulty of developing generative AI, which involves using computer programs to “teach” a machine how to create images or text. The technology is used in fields such as medicine and automotive design.
The company aims to gain traction in the generative AI market by providing a cloud platform for third-party developers and partnering with companies that have created trained AI models. On Tuesday, it announced a service called Bedrock that lets users build generative AI-powered apps using a selection of models from startups, including AI21 Labs, Anthropic, and Stability AI. The service is available in a limited preview.
In a nod to the growing popularity of generative AI, AWS also unveiled Elastic Cloud Compute (EC2) Inf2 instances powered by its custom Trainium chips, designed to handle large-scale AI tasks. The instance pricing varies but is typically more than double the price of EC2’s other high-performance computing instances, which are powered by Nvidia’s GPU chips.
Nvidia’s chips are widely used for developing generative AI because of their superior performance in calculating the complex mathematical operations needed to create graphics and text in digital content. However, several smaller companies have started offering alternative AI chips that aim to better match the computing requirements of those tasks. Earlier this year, Groq Inc. unveiled an AI chip to process natural language in chatbots. The company’s chip uses a technique known as neural net inference, which creates artificial neurons that mimic human brain activity. This allows the chip to run at lower voltages, consume less energy, and deliver better performance for a given price. This makes the chip suitable for smartphones and digital assistants, requiring fast processing speed and low latency.
