As the race to develop customized artificial intelligence (AI) hardware heats up, traditional semiconductor manufacturers could be at risk from new custom chips from some of their largest customers, such as Apple Inc. (AAPL) and Alphabet Inc. (GOOGL). As Mountain View, California-based search giant Alphabet continues to release next-gen versions of its home-grown Tensor Processing Unit (TPU), which first launched in 2016, one team of analysts on the Street suggests that the firm will use the chips to build computers.
Susquehanna analyst Mehdi Hosseini was out with a note following a trip to Taiwan in which he spoke with various sources in the tech industry, as reported by Barron's. The analyst wrote that a major trend he gathered from his research is that more and more companies are doubling down on the manufacturing of server computers, or "custom boxes," indicating that a great deal of the activity is coming from cloud computing providers. He expects server computer shipments to grow 10% in 2018, an increase from his initial forecast of 8%.
Cloud Giants Building 'Custom Boxes' In-House
Hosseini suggested that leading U.S. cloud players such as Google have started to ramp up their application-specific integrated circuit (ASIC) solutions to be used for AI applications. Google's TPU has often been included in the ASIC category, and has been boasted by the company as offering more speed than chips from traditional industry leaders like NVIDIA Corp. (NVDA) and Intel Corp. (INTC). (See also: Intel’s Chip Lead Is ‘Disappearing’.)
"In fact, this is the first time in Taiwan where we have consistently heard of ASICs in servers and for AI applications. Our view here is that after so many years of new concepts like 'AI' and 'IoT' dominating investor imaginations, it is finally beginning to materialize with leading Cloud players like Google stepping up and actually building out capacity," wrote Susquehanna.
Hosseini expects Taiwan Semiconductor (TSM) to be the primary foundry partner for Google as it ramps up its ASIC solution. (See also: Qualcomm at Risk From Apple, Samsung, Huawei Chips.)