Exclusive: Amazon developing AI chips, a significant investment in creating its own technology

Amazon.com is developing its own AI chips to enhance its AI operations and achieve greater control over its technology in the face of soaring demand for generative AI, four sources familiar with the matter told Reuters.<br><br>The retail and cloud-computing giant is working on developing chips customized for training large language models, such as its Bedrock AI and chatbots, including its Alexa technology, according to three sources. It is also looking into developing chips for inference, which is the process of generating responses based on the model's training, one of the sources said.<br><br>The company's AI chips are being designed by its Annapurna Labs business, which focuses on creating custom chips, the sources added.<br><br>Amazon said it has been developing a range of chips to support its AI initiatives.<br><br>"We are working on a wide range of AI-powered products and services, including developing chips to support training and inference," spokeswoman Zico Kolter said, but declined to elaborate.<br><br>The exact timeline for the release of Amazon's AI chips is unclear, but the company is expected to make an initial push into inference before potentially moving into training, two of the sources said.<br><br>"This is something that they are doing in earnest. They are hiring a lot of people to work on this. They are working with a lot of partners," said one of the sources.<br><br>Several Big Tech companies, including Alphabet, Microsoft and Meta, are developing custom chips to power their AI operations, with some already in use.<br><br>Amazon is investing heavily in its AI initiatives, including spending $4 billion last month to buy a prominent AI startup, Anthropic. <br><br>Its AI technology Bedrock powers its chatbot called Claude.<br><br>"It is a necessary investment in order to push the industry forward," said Gartner analyst Chirag Dekate.<br><br>An AI chip custom-designed by Amazon would allow the company to have better control over its technology, reduce its reliance on external suppliers and achieve greater performance.<br><br>"This is not just about costs. It's also about performance," said one of the sources.<br><br>"It is a necessary investment in order to push the industry forward," said Gartner analyst Chirag Dekate.<br><br>An AI chip custom-designed by Amazon would allow the company to have better control over its technology, reduce its reliance on external suppliers and achieve greater performance.<br><br>"This is not just about costs. It's also about performance," said one of the sources.<br><br>Amazon has been developing its own technology, including computer chips, for more than a decade, including its Graviton server chip and its Trainium chip for AI.<br><br>But developing AI chips that can handle training and inference at scale is a much more complex task, said one of the sources.<br><br>"It is a much, much harder problem to solve," said the source.<br><br>"If you are training a large language model, you are dealing with a very, very large amount of data. You need to be able to process that data very quickly. That's a very challenging task from a hardware standpoint."<br><br>By Nivedita Balu and Stephen Nellis