AMD (AMD) - Get Report and Xilinx (XLNX) - Get Report are eager to work together not just on new server platforms, but also on joint offerings for 5G base stations, cars and much else, its CEOs stress.
Three weeks after The Wall Street Journal first reported that a potential $30 billion-plus deal was being discussed, AMD announced on Tuesday morning that it’s buying top FPGA developer Xilinx in an all-stock deal that values Xilinx at more than $33 billion based on AMD’s current trading price.
The companies aim to close the deal by the end of 2021, and to generate $300 million worth of annual cost savings within 18 months of closing. AMD shareholders stand to own 74% of the post-merger company, and Xilinx shareholders the other 26%.
In tandem with the announcement, AMD posted its Q3 report, which was originally scheduled to arrive on Tuesday afternoon. The company comfortably beat Q3 estimates, reporting 45% annual revenue growth on the back of strong server CPU and notebook and game console processor sales, and also issued above-consensus Q4 sales guidance that implies 41% annual growth at its midpoint.
AMD’s stock, which was up 79% on the year as of Monday’s close, was down 4.1% in Tuesday trading to $78.88. Xilinx’s stock, which previously soared on the WSJ’s report, is up 8.6% to $124.35. That still leaves it about $10 below the AMD deal’s implied price.
Here are some initial thoughts on the AMD-Xilinx deal:
1. AMD’s 2020 Stock Gains Make the Deal a Lot Easier to Digest
When the WSJ’s report arrived, AMD’s stock was up 89% on the year, while Xilinx’s stock was up just 6%. This massive delta between the companies’ 2020 performances opened the door for AMD to negotiate a deal that would dilute its shareholders far less than a deal negotiated at the beginning of the year would have.
2. Data Center Efforts Will Be a Priority...
Though Xilinx historically hasn’t received a large percentage of its revenue from data center end-markets, the percentage has been rising in recent quarters, thanks to organic investments, acquisitions and engagements with internet/cloud giants (the proverbial hyperscalers) such as Microsoft (MSFT) - Get Report. During its September quarter, Xilinx’s data center revenue rose 30% annually and accounted for 14% of total revenue.
Among other things, Xilinx’s FPGAs are used within servers to perform AI inference (the running of trained AI models against real-world content), power SmartNICs (networking cards that can offload various network, security, virtualization and/or storage processing functions from CPUs) and storage and video acceleration.
During a conference call held on Tuesday morning, AMD CEO Lisa Su and Xilinx CEO Victor Peng both talked up the potential to create server platforms that encompass Xilinx FPGAs and AMD’s server CPUs and GPUs. Su particularly highlighted Xilinx’s efforts in the SmartNIC market. This is a field where Nvidia is also investing aggressively, following its recent acquisition of Mellanox Technologies, and where firms such as Intel (INTC) - Get Report and Marvell (MRVL) - Get Report are also present.
Should the Nvidia-Arm and AMD-Xilinx deals both close, Nvidia and AMD will both have server CPU, GPU and SmartNIC offerings to market, as will Intel.
3. ...But AMD Also Sounds Pretty Interested in Other Xilinx End-Markets, Too
Thanks to their hardware-programmability, which helps OEMs quickly bring products to market and reprogram the silicon that powers them later on, FPGAs have long been a common sight within mobile base stations, cars, wireline telecom equipment, airplanes, medical equipment and various other embedded systems.
Many of these markets are ones in which AMD currently has little or no presence in. But Su suggested that AMD is hungry to leverage Xilinx to expand the addressable market for AMD’s CPUs and GPUs.
“I think the broader markets in communications, in automotive, in industrial, and aerospace and defense, those are all very attractive markets that can use CPUs and GPUs,” she said. “And so I think the market presence that Xilinx has and the leadership in those markets also help.”
5G infrastructure -- a market where Xilinx has engagements with the likes of Samsung and Ericsson (ERIC) - Get Report -- is especially an area worth paying attention to, given the growing adoption of virtual RAN (vRAN) architectures in which off-the-shelf servers handle a 5G network’s baseband processing needs, while also potentially running other server workloads.
Conceivably, AMD and Xilinx could work on end-to-end 5G infrastructure solutions that pair AMD’s server CPUs with Xilinx’s radio processing, fronthaul processing, SmartNIC and (though ASIC-based solutions are quite popular here) baseband processing offerings, with OEMs able to pick and choose which offerings they want to use.
It’s worth noting that making a larger push into embedded processing would likely spell additional competition for AMD not only with Intel, but also with developers of Arm-based embedded processors such as NXP Semiconductors (NXPI) - Get Report and STMicroelectronics (STM) - Get Report.
4. Tech Synergies Go Beyond Pairing CPUs and GPUs With FPGAs
In areas such as chip manufacturing and packaging technology, AMD and Xilinx’s engineers could learn a lot from each other, Su suggested on the call.
“In manufacturing, we will be able to apply learnings from AMD's early adoption and high-volume production ramp at advanced process nodes across Xilinx's products. In packaging, Xilinx has demonstrated leadership and significant IP in the 2.5D and 3D die-stacking technologies that are becoming increasingly important as Moore’s Law slows” she said.
Su added that (as Nvidia and Intel also outline plans to create end-to-end software stacks for their data center silicon) AMD and Xilinx also “see opportunities to combine our open-source software offerings into a unified stack, optimized to accelerate computing at the platform and the system level. “
5. Chinese Regulatory Approval Is a Potential Wild Card
In theory, it shouldn’t be hard for an AMD-Xilinx deal to get cleared by regulators, given the enormous differences in each company’s product lines and end-market/customer concentration.
Nonetheless, as Qualcomm and NXP can attest, such differences don’t necessarily guarantee that Chinese antitrust regulators will approve a chip M&A transaction in the current geopolitical environment. Chinese regulators did approve the Nvidia-Mellanox deal, but only after a lengthy review -- and that was before fresh sanctions were imposed on both Huawei’s HiSilicon chip unit and top Chinese foundry SMIC.
Worries about Chinese approval might have something to do with why Xilinx is still trading at a healthy discount to the AMD deal’s implied price.
6. AMD’s Server CPU Momentum Should Aid its Planned Data Center Efforts With Xilinx
It’s getting overshadowed by the Xilinx news, but AMD just reported another very strong quarter for its Epyc server CPU business. As was the case in Q2, Epyc sales more than doubled annually in Q3, aided by continued strength with hyperscalers and growing enterprise traction.
(All of this contrasts a lot with what Intel just reported about its server CPU unit's Q3 performance, as the business deals with both tougher competition from AMD and weak enterprise server demand.)
Moreover, Su indicated AMD expects a pretty strong initial reception for its third-gen Epyc server CPU line (codenamed Milan), which will begin shipping to cloud and HPC clients in Q4 and to major enterprise server OEMs in Q1. She asserted that pre-launch customer engagements for Milan are stronger than pre-launch engagements for AMD’s second-gen Epyc line (codenamed Rome), that “more than 100 new or refreshed Milan systems are in development” and that Milan’s big single-thread performance gains relative to Rome will help it win new customers.
In a nutshell, AMD’s server CPU business is -- thanks to both strong execution and Intel’s manufacturing technology setbacks -- much bigger than it was not too long ago. And all signs suggest it’s poised to see another year of strong growth and share gains in 2021, as Intel finally rolls out a server CPU platform (codenamed Ice Lake) that was once expected to compete against Rome rather than Milan.
And this in turn should give AMD far more opportunities to sell companies on platforms featuring both its CPUs and Xilinx’s FPGAs (and in some cases, AMD’s GPUs) than it would have if the companies merged in, say, 2018.