AMD has announced it will acquire FPGA developer Xilinx in a $35 billion-dollar, all-stock deal. The deal, which we discussed earlier this month, will give AMD access to new markets that it hasn’t previously played in, including FPGAs and artificial intelligence.
During its Q3 2020 earnings call, AMD CEO Lisa Su claimed that Xilinx was an attractive acquisition target because it offers “industry-leading capabilities in SoC design, serdes and high-speed IO, mixed-signal RF, advanced 2.5 and 3-D silicon integration and packaging, as well as targeted software stacks for key verticals.”
AI was not a major focus of the announcement, though it’s been a hot topic in the larger semiconductor world. While both Lisa and Xilinx CEO Victor Peng mentioned AI as one area where the combined AMD Xilinx would be able to leverage technology more effectively, the focus was more on broad enterprise opportunities across hyperscaling and telecoms worldwide. AI was generally discussed as adjacent to those efforts, though AMD did talk up the importance of software ecosystems, and announced it wants to combine its own open-source offerings into a unified product stack.
This is a bit curious because Xilinx has been doing some work in AI throughout 2020, including an effort just a few weeks ago to detect COVID-19 at the edge using AI. There’ve been several AI-related product announcements in the past six months. During the call, however, AMD played up more generic opportunities in edge, industrial, 5G, and data centers, with AI mentioned as one product in the mix.
One possible explanation for this is that AMD is still working out what it wants its AI response to be. When we’ve asked the company questions about its long-term goals in the segment, the response has typically emphasized AMD’s goal of improving general computing performance as a whole rather than focusing on the AI market. This approach has paid huge dividends for AMD the past few years and shows no sign of running out of steam, so it may just be that AMD doesn’t want to distract from the near-term places where it will be deploying hardware to talk about longer-term plans for future opportunities.
On the other hand, it’s possible that AMD doesn’t see the Xilinx acquisition as doing all that much to help their AI position. In CPUs, AMD is challenged by Intel and AVX-512. It has shown no particular interest in leaping into that fray, and given the difference in their respective core counts, AMD can still argue that an awful lot of workloads benefit more from higher core counts faster clocks as opposed to AVX-512 support. On the GPU side of things, AMD is hemmed in by Nvidia and CUDA. Acquiring Xilinx doesn’t do much to change that situation.
According to its most recent financial reporting, Xilinx’s top markets are:
A&D, Industrial, TME (Aerospace and Defense, Telecommunications, Media, and Entertainment): 44 percent.
Automotive, Broadcast, and Consumer: 16 percent.
Wired and wireless: 26 percent.
Data center: 14 percent
Channel: 0 percent.
Hilariously, “Channel” is listed as 0 percent but is also listed as having declined -1 percent from its FY 2020 level of 0 percent.
AMD refers to using Xilinx’s extensive market in telecom as a way for Epyc to get a foot in the door of those markets, with the implication that the two companies may find new synergies through similar arrangements in other segments.
There is one other reason AMD might be playing things a little closer to its chest. The company got absolutely roasted for its overall handling of the ATI merger and had great difficulty bringing its first APUs to market. AMD may be planning to take a somewhat more measured pace to developing long-term plans that involve merging AMD and Xilinx products on the same piece of silicon. At the same time, given AMD’s leading position in chiplet design, it wouldn’t be surprising to see an x86 / FPGA hybrid chip design of some sort in the medium-term future.