With AI gaining center stage, tech giants want to cut back on their dependence on chipmakers while increasing their return on investment. In a recent report, Microsoft-backed OpenAI, the company that created ChatGPT, was said to be attempting to enter the AI chip market. Now, another report claims that the Windows maker intends to announce the company’s first chip made to process AI models at its annual developers’ conference.
According to The Information, which cited a source with firsthand knowledge of the situation, the launch, which represents “the culmination of years of work,” could assist Microsoft in reducing its reliance on Nvidia-designed AI chips, which are now in limited supply due to a surge in demand.
The “Athena”-codenamed microprocessor may make its public debut on November 14 in Seattle at Microsoft’s Ignite annual conference.
According to reports, the Microsoft chip is comparable to Nvidia GPUs and was created for data center servers that train and operate extensive language models, the program that powers conversational AI capabilities like OpenAI’s ChatGPT.
Microsoft now powers LLMs for cloud customers such as OpenAI and Intuit as well as for AI capabilities in Microsoft’s productivity programs using Nvidia GPUs. The technology would make it possible for Microsoft reduce its dependency on Nvidia’s H100 GPU, which is rumored to be experiencing supply issues as a result of rising demand.
OpenAI might produce its own AI chips
The change occurs at a time when a report said that OpenAI was also thinking about developing its own AI processor. Since at least last year, the corporation has engaged in conversations around AI chip strategy, according to news agency Reuters. Sam Altman, the CEO of OpenAI, has also prioritized the company’s purchase of additional AI chips, according to the statement.
The Tensor Processing Unit, or TPU, is used by Google to train its sizable generative AI models like PaLM-2 and Imagen. Both training and inferencing are supported by proprietary chips from Amazon (Trainium and Inferentia).
