Morgan Stanley’s Joseph Moore and his team attended the Hot Chips annual conference, which focuses on developments in the semiconductor industry, and they write that specialty artificial intelligence (AI) solutions stood out.
Moore writes that the most interesting themes were focused on deep learning and AI and the enabling semiconductor building blocks, across graphics, field-programmable gate array products (FPGAs), and custom chips. He notes t hat several memory intensive ecosystems are emerging to serve the growing needs of rapidly growing deep learning markets.
More detail from the note:
While all three product categories are significant, we continue to be struck by the growing importance of FPGAs, with 3 of the world’s top 7 cloud vendors presenting elements of their FPGA strategy – in stark contrast to investor skepticism on the use of FPGAs in the cloud. On the training side, several companies presented alternatives to NVIDIA (NVDA) graphics chips, including Google (GOOGL) with 2 presentations on its Tensor Processing Unit – though we still see NVIDIA’s Volta in a fairly dominant position for most training applications over the next couple of years.
Moore writes that FPGAs will probably play a bigger role than investors believe in machine learning “inference”, or “real time usage of deep learning trained neural nets in a production environment” (that’s a mouthful) and that Xilinx (XLNX) is as key beneficiary. Baidu (BIDU), a Xilinx customer, is building a new architecture that could “considerably broaden the use of FPGAs, and Amazon’s (AMZN) Amazon Web Services, which uses FPGAs for machine learning, provided an update of their Xilinx FPGA F1 service at the conference, a business that is a big part of Moore’s bullish thesis on Xilinx. Microsoft (MSFT) also presented a view of its Altera FPGA based implementation of machine learning inference, Brainwave, and Moore believes Xilinx may be able to break into that company as well in the future.
The Technology Select Sector SPDR ETF (XLK) is lower this afternoon.