This morning, AMD announced its completion of its Xilinx acquisition, the largest semiconductor deal ever and the second largest tech acquisition ever at its $49B value.
I had the chance to catch up with AMD CEO Lisa Su this weekend to talk about Day One of the combined AMD and Xilinx. It was a good conversation, one that indicates to me that a focused AMD has a big and bright future opportunity. I was also surprised to hear that AMD previously licensed Xilinx technology and has been integrating that IP into its processor roadmap with the first expected products in 2023. Surprise!
AMD and Xilinx in review
I think it would be helpful to first review what the companies do separately and then talk about what the combined entity can offer.
AMD is a $135B market cap valued company deriving most of its annual $16.4B revenue from high performance PC, server and game console CPUs, GPUs and SoCs with an annual $85B TAM. It primarily runs a horizontal development ($2.8B/year) and marketing strategy except for game consoles for Sony and Microsoft which is very much a specialized use case.
Xilinx is a $48B market cap valued company and drives most of its $3.1B annual revenues from embedded markets with FPGAs and very high-performance and programmable SOCs with an annual $50B TAM. Xilinx has a balanced horizontal and vertical development ($1B/year) and marketing approach and strength in automotive, carrier, industrial, aerospace and defense applications. Xilinx has mature and accelerated software stacks for AI, networking, analytics and video transcoding.
Combined company capabilities
I see opportunities in AMD bringing its IP and solutions to Xilinx’s embedded markets and vice versa, but I see the biggest opportunity in the datacenter where both companies have different kinds of strengths.
AMD has strength in server CPUS and has a line of datacenter GPUs while Xilinx offers IP, products, and expertise in SmartNICs, AI inference, and AI analytics. Marvell invented the SmartNIC, AWS took it to another level and now AMD will be going after that market competing with Marvell, Intel, and NVIDIA. With new AI inference capabilities for the datacenter, I could see AMD having much stronger offerings to compete with both NVIDIA and Intel. AMD, technically, has already been targeting the datacenter AI training market with GPUs but it has not resulted in a high degree of market share success. Xilinx has been focused on AI inference, and their software stacks are more mature and I think will bring much needed software resources and expertise to the party. With NVIDIA’s market cap of $600B, Wall Street values AI leadership.
I believe the future of datacenter processors are flexible and accelerated processors with the right balance of programmability and the most efficient, fixed function blocks in very sophisticated packaging. AMD deserves a lot of credibility for its in-market chiplet and interconnect technologies. While I have not seen details on AMD’s future packaging (a peek with 3D stacked V-Cache), AMD plus Xilinx brings an endless possibility of programmability and ultimate efficiency. Xilinx has quietly built an impressive array of high-performance packages and interconnects for its adaptive SOCs.
As for the datacenter edge, I believe companies need to bring more than a horizontal approach to the market to see success. Success in datacenter will not make you successful in the datacenter edge. Companies need to get vertical and better understand use cases and industries. Companies need corresponding distribution, sales and marketing who get these markets. This is where Xilinx will help a lot, particularly with carriers who are becoming a force on the edge.
Combined IP products in 2023
So when will we see the combined fruits of the AMD-Xilinx combination? Early than anyone might have imagined. Lisa Su told me we should see the first AMD processor with Xilinx AI IP in 2023. Apparently, AMD has had a long-term development agreement with Xilinx and was able to work together even as the deal was getting approved. I believe it took AMD and ATI many years after the deal completed for combined IP products.
Business-wise, you have to like the AMD-Xilinx combination.
First off, the deal is accretive to non-GAAP margins, non-GAAP EPS and free cash flow generation in the first year. Psychologically this serves as a deterrent to mess with the classic Xilinx business. In fact, current Xilinx CEO, Victor Peng will become President of AMD’s new Adaptive and Embedded Computing Group (AECG) that combines the classic Xilinx business with AMD’s embedded business. Peng will report into AMD CEO Lisa Su.
While 16 months ago, scale was one of the drivers for AMD acquiring Xilinx, AMD has increased its revenue to such a point that that Xilinx can benefit a lot more from it than AMD can. I can see Xilinx getting wafer and packaging cost reductions as well as ATMP cost reductions, improving classic Xilinx gross margins.
I like the stable, reduced peaks and valleys kind of business Xilinx brings to AMD, that has had a history or feast or famine over its existence. I lived in peaks and valleys at AMD for 11 years and it is disruptive. While I think AMD CEO Lisa Su has created a much more stable AMD , AMD markets are highly competitive and it will be interesting to see how AMD performs if and when Intel starts hitting on all cylinders or if NVIDIA goes all-in on Arm-based PC or datacenter processors. Many of Xilinx’s customers commit to the company for a decade in its designs, unlike the faster cycles in the PC, gaming, and server processor markets.
While AMD did not say much about G&A capture in its disclosures, it did announce its 2022 Financial Analyst Day is scheduled for June 9th where it will shed even more light on how the two entities will mesh together.
While I do not think either company needs each other for a positive future, I do think AMD and Xilinx can be better off together. After over 30 years of marriage, I can say that equation worked well for my wife and I. Xilinx helps fills some of AMD’s holes and vice-versa. AMD and Xilinx both bring unique capabilities, and I can envision combined capabilities that end customers will want and can profitably drive revenue. The tight rope CEO Lisa Su will need to walk is not disrupting Xilinx’s market while getting value from its people and assets for the entire company. Having watched Lisa Su in action the past decade, you’d have to be a fool to bet against her.