The rising prominence of field-programmable gate arrays (FPGAs) for training artificial intelligence (AI) models has opened up a huge opportunity for Xilinx (XLNX) as it controls the majority of this market. Energias Market Research estimates that sales of these chips could increase at a CAGR (compound annual growth rate) of 9% through 2023 to almost $13 billion.
Xilinx, however, needs to stay on top of its game in the FPGA market as it faces competition from a big fish like Intel (INTC 4.45%). Chipzilla realized the importance of FPGAs early on, forecasting that a third of cloud service providers will be deploying these chips in data centers by 2020. As a result, Intel bought Altera for $16.7 billion back in 2015 to make its entry into this space.
But Xilinx's technology lead has kept Intel from eating into its market share, and it won't be relinquishing this advantage anytime soon thanks to its latest chip design -- codenamed Everest.
Project Everest: The lowdown
Xilinx has reportedly invested four years, employed 1,500 engineers, and incurred $1 billion in research and development costs for Project Everest. But what is it. It's a chip design that packs together an FPGA, a CPU, high-bandwidth memory, and fast connectivity to the network. Usually, an FPGA can be reprogrammed to perform a variety of tasks as compared to traditional chips that can only perform the task they were originally designed for.
This is why these chips have been gaining traction in artificial intelligence applications as compared to traditional chips such as GPUs and CPUs, which are restricted in their capacity due to their fixed architecture. And, the addition of a CPU makes the FPGA platform even more versatile as it can be used to execute both specific and customizable tasks, speeding up the data center and making it more suitable for AI applications. In fact, Xilinx claims that this integrated chip package is 20 times faster than its existing offerings while running AI applications.
Now, Xilinx hasn't created something new by fusing together an FPGA and a CPU. Intel has launched such hybrid chips in the past, and it is reportedly working on a new generation of a hybrid FPGA-CPU device. But then, Xilinx seems to have beaten Chipzilla to the market with its own offering this time, and this could spell trouble for Intel's programmable solutions business.
Why Intel should be worried
Xilinx's Everest chip is based on the 7-nanometer (nm) architecture, while Intel is currently still making the jump from a 14nm to a 10nm platform. Intel previewed its 10nm chips in September last year, but it hasn't revealed when it will commercially launch them. By comparison, Xilinx is expected to start shipping its Everest-based chips from next year.
A lower nanometer number means that the chip platform is based on a smaller node, so it should ideally have lower manufacturing costs and deliver more performance for each watt of power consumed.
Intel's Programmable Solutions Group (PSG) business grew 14% in 2017 to $1.9 billion, accounting for 3% of the company's total business. So, FPGAs don't move the needle in a big way for Intel, but this business has got great potential. In fact, if Intel is able to maintain its current market share of 42% in the FPGA space, its revenue from the PSG segment could almost triple in the next five years if sales become as big as Energias Market Research estimates.
Xilinx, however, is looking to capture as much as 65% of the FPGA market in the next three years, and it is making great progress on this front. Sales of the company's advanced products, which include chips based on advanced processing nodes such as 20nm and 16nm, increased 30% year over year last quarter.
These products now supply 56% of the chipmaker's top line as compared to 47% in the prior-year period. The introduction of more advanced products based on the Everest platform should help Xilinx sustain its impressive momentum.
Why Intel could continue losing to Xilinx
Intel reportedly held 48% of the FPGA market back in 2011, so it's lost 6 percentage points over the ensuing years as Xilinx has moved from 52% to 58%. It won't be surprising if Xilinx continues to consolidate its lead on the back of a superior chip platform in the form of Everest.
If it could get 60% of the market, based on Energias' estimates, Xilinx would more than triple its trailing-12-month revenue of $2.48 billion by 2023. This would seemingly relegate Intel to second place in the FPGA industry despite its multibillion-dollar bet on Altera, potentially adding yet another name to Chipzilla's string of failed acquisitions.