AMD is said to be in talks to buy Xilinx in a move that would be worth an estimated $30B and would give AMD a foothold in the FPGA market. Such a move could be a good one for the CPU manufacturer, as it would open new markets for the company where it doesn’t currently compete.
Xilinx is the inventor of the Field-Programmable Gate Array, which are integrated circuits designed to be reconfigured at-will (hence “Field Programmable”) and can be reprogrammed for a huge range of logic functions, including any function that an ASIC (Application-Specific Integrated Circuit) can perform. CPUs, GPUs, FPGAs, and ASICs are part of a continuum of hardware stretching from CPUs (most flexible, least efficient) to ASICs (least flexible, most efficient), as shown in this diagram from Intel below (ignore the “core counts,” as they’re not relevant for our purposes). Intel bought Altera several years ago, so AMD acquiring Xilinx would make a certain degree of sense.
AMD’s attempt to purchase Xilinx could be a move towards the AI market, which is the one major computing area where AMD is currently only semi-active. While the company has its ROCm platform for Linux, it serves as a wrapper for CUDA and isn’t Windows-compatible. ROCm also lacks Navi support at the moment, which is something that’s expected to change once AMD launches its CDNA computing architecture. Still, this is one area where Team Red doesn’t have much presence. Intel has pushed ahead with AVX-512 on CPUs and its own OneAPI for GPU and other device support, Nvidia has CUDA, and AMD has been focused on the general computing market, where Zen has been remarkably successful.
Still, it’s only a matter of time before the company turns its attention to AI, and owning Xilinx would give it a leg into that space. Xilinx has previously partnered with companies like Mipsology to create simple methods for developers to port code from GPUs to FPGAs. For its fiscal year 2020 (ended April 22, 2020), Xilinx reported $3.16B in sales and net income of $793M.
This kind of revenue and positioning would definitely give AMD a new market. Intel hasn’t put much emphasis on the idea of shipping Xeons with integrated FPGAs, but the adoption of chiplets creates new opportunities for different types of silicon to share the same physical chip. It’s not hard to imagine AMD mixing and matching different types of chiplets — FPGA blocks, GPU blocks, and CPU blocks — to build products that target different markets.
The original article by the Wall Street Journal notes that AMD is now worth more than $100B, with its stock up 89 percent this year. Xilinx is worth about $26B. With premium, that brings the value of the deal to ~$30B, which AMD would presumably pay for in stock, since it doesn’t have a war chest quite that large to buy the other company with. Xilinx sales have been hit by the Trump Administration’s restrictions on sales to companies like Huawei, with that firm accounting for an estimated 6-8 percent of its revenue, according to the report.
Any acquisition would need to be managed properly and hopefully better negotiated than AMD’s purchase of ATI in 2006. AMD drastically overpaid for that company, wrote off 50 percent of the purchase price within two years, and spent years struggling to fully integrate its CPU and GPU businesses. Any acquisition of a major firm like Xilinx would need to be careful not to fall down a similar post-acquisition well.