Microsoft, a leading software company, is striving to join the ranks of other AI-minded tech firms, such as Apple, Amazon, Google, Facebook, Tencent, and Alibaba, in launching its own custom-built AI chip for use in scale-out artificial intelligence, dubbed Athena. Microsoft’s major investment in OpenAI in 2019 gave birth to this endeavor, leading to the engineering of samples being made available to Microsoft and OpenAI employees.
Apple’s vertical integration of hardware and software within its product lineup, spanning billions of devices worldwide, has achieved success which Microsoft is striving to emulate. In the past, Windows RT on Arm ventures have been less than successful, but Microsoft’s optimistic outlook regarding their approach to AI hardware may find favor. Already, Microsoft’s Brainwave project has been using Intel Stratix 10 FPGA (Field Programmable Gate Array) cards since 2017, a first step into Microsoft’s AI strategy.
This ambitious endeavor is driven by a team of 300 dedicated engineers and has a roadmap that extends to multiple generations. The first generation of the chip is planned to be built with the 5nm process developed by Taiwan Semiconductor Manufacturing Company (TSMC). To further bolster this process, Microsoft has employed Mike Filippo, formerly an Apple CPU architect, to lend his expertise, although his primary focus has been with CPUs.
Microsoft is utilizing its ChatGPT platform to integrate its AI tech, with Microsoft 365, Bing, GitHub, and its Azure stack the key areas of deployment. Its estimated to cost billions of dollars annually, although Microsoft is also teaming up with Nvidia in building a powerful supercomputer, which may ease the burden as well as create further demand for their AI tech.
All in all, Microsoft appears to be well on its way in replicating Apple’s positive results in its efforts to launch a vertical stack for AI. Its investment in the chip project thus far is hopeful, and its long-term approach ensures that this ambitious endeavor may have a successful future.
