Field-programmable gate arrays (FPGAs) are in strong demand thanks to their versatility. The programmable nature of these chips means they can be wired to tackle specific applications after they are manufactured, making them ideal for deployment in data centers to accelerate artificial intelligence workloads. This is one of the reasons FPGA sales are expected to jump from just under $6 billion last year to almost $10 billion in 2023, according to one estimate.
The secular growth of this market should be great news for Xilinx (XLNX +0.00%), which has dominated bigger rival Intel (INTC +1.14%) in the FPGA market. FPGA pure-play specialist Xilinx controlled nearly 60% of the market at the end of 2017, with Intel commanding the rest. But it looks like the status quo has started changing.
Image Source: Getty Images.
Intel is turning the tables
Intel's programmable solutions group (PSG) revenue shot up 18% year over year during the second quarter to $517 million thanks to strong demand for its FPGAs in the data center. In fact, Chipzilla claims that its data center sales within PSG have now more than doubled for two quarters in a row.
As such, Intel is now clocking an annual revenue run rate of over $2 billion in FPGAs, which is pretty close to Xilinx's trailing-12-month revenue of $2.5 billion. So the gap between the two companies isn't much right now, and it might not be long before Intel overtakes Xilinx in this market. This seems a bit surprising as Xilinx reportedly enjoys an 18-month technology lead over Intel in FPGAs, but the latter has been outgrowing its nemesis of late.
Xilinx's revenue in the latest quarter grew 14% year over year, trailing Intel's growth. What's more, a closer look indicates that Intel's FPGA business has been growing at a faster pace than Xilinx in recent quarters.
Data from Xilinx and Intel quarterly reports, Chart by author. PSG = Programmable Solutions Group.
So Intel has managed to upstage Xilinx despite being behind on the FPGA technology curve. However, Xilinx has a new weapon up its sleeve in the form of Project Everest. The FPGA specialist could start shipping chips based on the more efficient 7-nanometer manufacturing node as soon as next year.
Intel, meanwhile, will be stuck with the 3-year-old 14nm process for a while, provided it doesn't delay its latest target of launching 10nm chips by the end of 2019. Intel's chief engineering officer, Murthy Renduchintala, believes that the company's 10nm data center products will hit the shelves "shortly after" the PC chips, which means that it might be close to another two years before we see a new-gen Intel FPGA.
But Chipzilla won't let this disadvantage hurt its FPGA momentum as it has found a new way to attract customers.
Intel pulls a rabbit out of the hat
Intel recently acquired semiconductor manufacturer eASIC, which makes customizable chips for deployment in cloud computing and other wireless environments for an undisclosed sum. This move is expected to help Intel address a major problem with FPGAs.
The fact that FPGAs can be customized to perform specific tasks has boosted their popularity as they are more efficient at executing AI applications. But they need to be transformed into ASICs once an application doesn't need customization.
In simpler words, once an application has stopped changing, it has to be run using chips that are more structured in nature to deliver better performance and consume lower power. Application-specific integrated circuits (ASICs) are well-equipped to handle the post FPGA stage because of their power efficiency and ability to be configured quickly.
This is where Intel's acquisition of eASIC comes into play. Dan McNamara, the general manager of Intel's Programmable Solutions Group, told VentureBeat that eASIC's software will enable a smooth transition from FPGAs to ASICs. As such, Intel is looking to offer an end-to-end solution to customers that begin with FPGAs and finally use ASICs to tackle data-heavy applications in the cloud or aid the deployment of faster wireless networks.
The big picture
The bottom line is that Intel is looking beyond FPGAs to lock in customers from the beginning to the end of the deployment. This is a smart ploy as an integrated solution could help it score more customers over Xilinx despite the technological disadvantage.
AMD is busy making inroads into both PCs and servers, two markets that Intel has traditionally dominated, so Chipzilla needs to find new ways to grow the business, and FPGAs present one such opportunity as they are expected to overtake graphics cards for accelerating AI workloads.
