Microsoft has been working on a new artificial intelligence (AI) chip named Athena since 2019, according to The Information. The tech giant is expected to release the chip during 2021 and it could be implemented in the company’s own and OpenAI’s core technology. However, the chip is not seen as a threat to the major chip-provider Nvidia, it is instead a signal of the need for hyperscalers to build custom silicon to support large language model (LLM) training.
The intense development and growth of advanced generative AI models requires more compute capabilities and this has been a concern for many organisations. According to Gartner analyst Chirag Dekate, the increasing number of parameters needed to optimise generative AI models already exceed our current capabilities and require Exascale calculations. Thus, Microsoft, in the same way as Google (TPU) and Amazon (Trainium and Inferentia processor architectures), decided to take matters into their own hands and develop their own AI chip for faster training, with cost saving benefits.
Nvidia’s work around AI chips, in terms of market share, is dominating, reaching an 88% share according to John Peddie Research. Nonetheless, these custom chips are also looking to make their presence felt in the inference process, an area in which Nvidia is not as strong. According to Jack Gold, analyst at J. Gold Associates, this custom silicon would provide a cheaper option to customers not needing the high grade Nvidia chips and it could also cut costs at the hyperscale’s end when using the A100 and H100 GPUs.
Nvidia has been on top of the game when it comes to ruling the AI/ML space; they were behind the so-called AI revolution and have crafted powerful platforms and software-centred approaches that have seen their stocks rise. Nonetheless, Microsoft’s Athena is an important milestone for the industry, with the possibility for disruptive economies of scale through non-traditional technology approaches.
Chirag Dekate also highlighted Nvidia is still the major force for generative AI development, their expertise and experience are hard to match. Nonetheless, custom AI ASPs will be an important asset for both creating economies of scale and providing an optimised version of the AI chips to different users. As Microsoft develops their Athena strategy, the semiconductor industry will be looking to find means of engaging in the AI market. In order to do so, they will have to innovate according to the “last phase of Moore’s law”, which means heterogenous acceleration comprising GPU and chips specific to the applications.
Microsoft is a renowned technology giant and one of the most valuable companies in the world with a market cap of 1.59 trillion dollars. Founded in 1975, the tech firm offers services and develops products ranging from personal computers to enterprise software, hardware and related services.
John Peddie is a well-known analyst that has been following technological industry trends for over 35 years. He founded John Peddie Research in 1988 and since them, his firm has established a reputation as leading industry analyst covering the PC graphics and semiconductor markets.
