If AMD buys Xilinx for upwards of $30 billion -- and if the financials work out -- the combined entity would conceivably reach initial annual revenues of at least $10 billion based on 2019 numbers. That would make it the eighth largest semi company globally just behind Texas Instruments.
More important, AMD would grab Xilinx’s market-leading share of Field Programmable Gate Array (FPGA) chips to go head-to -head against Intel, the world’s other big FPGA provider. FPGAs, invented by Xilinx in the early 80s, can be programmed by customers to meet their unique computing needs, an essential ingredient for AI applications used in self-driving cars, robots and many other applications. Such capability is considered important for defense and communications applications as well.
Intel is the world’s largest chip designer and manufacturer, with $71 billion in 2019 revenues. Intel bought Altera in 2015 for its FPGAs, making it the only other significant player in such chips.
An AMD purchase of Xilinx fits into AMD CEO Lisa Su’s strategic approach. She told CNBC in July that it is important to “plan for the long term.”
Neither company has commented on reports of a coming deal, first announced by The Wall Street Journal late Thursday. But nobody has denied the deal is coming either, “which says everything,” said Patrick Moorhead, an analyst at Moor Insights & Strategy.
Moorhead ticked off several benefits to AMD in buying Xilinx, including the fact that AMD has a market capitalization of more than $100 billion, reached in 2020 with an 89% increase in share price. “Given AMD’s valuation, the company needs to broaden its market opportunity,” Moorhead said in an email to FierceElectronics.
“AMD’s stock appreciation over the last few years offers them the perfect opportunity to do this deal,” added Kevin Anderson, an analyst at Omdia.
Xilinx has strength in carrier infrastructure, automotive and aerospace markets to help AMD diversify, and expands AMD’s datacenter presence, Moorhead added. The FPGAs would add to AMDs compute arsenal of CPU and GPU strength used in PCs and data centers that are in greater demand with COVID-19 forcing more people to work and study at home.
Moorhead predicted “near-zero regulatory hurdles” to overcome with the deal, contrary to what Nvidia faces in the purchase of Arm. He said there’s “zero overlap” in the product sets of AMD and Xilinx, something important to regulators who want to increase market opportunities for customers.
One ding on the deal: there’s some risk with AMD’s “lack of acquisition experience,” Moorhead added.
Drilling down on the value of FPGAs to AMD, they “fill a specific gap that CPUs and GPUs alone don’t cover,” added analyst Jack Gold of J. Gold Associates.
“New work loads like machine learning, 5G and networking require virtualized, easily modified and software defined system that run at high speed,” Gold explained “FPGAs p provide a bridge between high performance CPU/GPU and the custom functions that specialized accelerators can provide. So, FPGAs fill the gap between programmable specialized functions and customized function specific chips that can take a long time to develop and are only viable in large volumes. “
With an FPGA, a customer can easily adjust algorithms, which is important in new emerging compute areas, while custom chips require a customer to commit to a single architecture or algorithm, Gold added.
“That’s why Intel bought Alterra, to provide its customers with that ability, and why others have partnered with Xilinx,” Gold added. “If AMD buys Xilinx, it gives them one more vector of processors they can provide to cloud, data center and specialized functions like network cores. And it further accelerates their growth in high performance data centers and cloud systems.”
Chris Morris, another analyst at Omdia, said cutting edge tech like AI, 5G and Advanced Driver Assistance Systems are growing in importance, increasing the need for FPGA adaptability. “We’ve seen time and time again that previous standard processors aren’t always the best, especially when paired with the need for maximum performance with maximum power efficiency,” Morris said.
“Xilinx FPGAs, with the ability to change their design with a software update, are proving to be one of the leading solutions to this problem,” Morris added. “FPGAs are able to update their design without needing a change in hardware and have been able to outperform traditional locked semiconductor designs such as ASICs while still being two or three generations old.”
Morris added that pairing FPGAs from Xilinx used in AI processing with the parallel processing power of AMD GPUs “could place AMD at the forefront of AI development.”
Xilinx has posted a lengthy list of potential FPGA applications on its web site, including defense and auto applications. In May, Xilinx introduced its space-grade, radiation-tolerant Kintex UltraScale FPGA family.
Leonard Lee, managing director of the NeXt-Curve analysis firm, agreed that an acquisition of Xilinx will also push AMD into the 5G infrastructure market as well as move AMD beyond CPUs and GPUs. Doing so means AMD would be more of a system architecture player in its business strategy, much the same as Nvidia is attempting to be with the purchase of Arm. “It’s the system that is scaling, not the CPU,” Lee noted.
On Monday,some investors raised questions about the financial merits of an AMD purchase of Xilinx, even if the technology component of the partnership is promising. Stone Fox Capital analysts in an article posted on Seeking Alpha called the deal "unlikely" and noted that AMD's stock dropped 4% on the news of a possible deal.
"At a list price of $30 billion, Xilinx wouldn't be easy for AMD to swallow considering the likely need to utilize debt," Stone Fox Capital said. "A good business combination from a technology viewpoint might not work from a financial point of view."