Xilinx versus Intel
A fast mover in the FPGA (field programmable gate array) space, Xilinx (XLNX) has developed software support to encourage adoption of its FPGA technology and is already working with several cloud companies in their machine learning efforts. Specifically, Xilinx has partnered with ARM Holdings (ARMH) and TSMC (TSM) to develop 7 nm chips.
Meanwhile, Intel (INTC) hasn’t accomplished nearly as much, though it has been busy integrating Altera, catering to its customers in the defense, communications, networking, industrial controls, and instrumentation markets. In October 2016, Altera launched its first FPGA, Stratix 10, since its integration with Intel. The new FPGA is built on Intel’s 14 nm node.
One year ahead of Intel
While Intel’s 14 nm is more efficient than TSMC’s 16 nm node in terms of both power and performance, customers still prefer Xilinx solutions because Xilinx’s 16 nm products came onto the market one year ago, while Intel’s 14 nm FPGAs have just arrived. This one year gap gave Xilinx a first mover advantage and helped it win over 200 customers for its 16 nm chips.
Xilinx is now preparing its 7 nm chips, which should have better performance and power efficiency than Intel’s 14 nm chips. It thus appears that Xilinx’s 7 nm will arrive at least one year before Intel’s 10 nm, but this will depend on the progress made by TSMC toward its 7 nm node.
TSMC plans to start production on the 7 nm node in 2018, while Intel plans to start production on 10 nm node in late 2017.
Xilinx more profitable than Intel’s Programmable Solutions
In calendar 3Q16, Xilinx revenues rose 10% YoY (year-over-year) to $579 million, while Intel’s revenues from PSG (programmable solutions group) grew only 6% YoY to $425 million. On a sequential basis, Xilinx’s revenues rose 1%, which Intel’s PSG revenues fell 9%.
To be sure, Xilinx is more profitable than Altera because the former has an operating margin of 30.5%, as compared to latter’s 18.3%. In fact, Intel’s Altera segment posted losses in the first two quarters of calendar 2016 due to acquisition-related charges.
Xilinx may grow faster than Intel
Xilinx has been tapping growth in cloud service companies, supplying FPGAs to Baidu (BIDU) for machine learning and now to Amazon Web Services for FPGA-backed cloud services. Xilinx has also collaborated with Advanced Micro Devices (AMD), ARM, Huawei, IBM, Mellanox, and Qualcomm on new CCIXs (cache coherent interconnect for accelerators), a single interconnect format that would allow different CPUs (central processing unit) to share data with FPGA accelerators without the need for special software.
Still, Intel dominates the data center market, with more than 99% share. It will be hard to beat such a behemoth. But data centers have several sub-segments, and Intel’s GPUs (graphics processing unit) are not very competitive when compared to Nvidia’s (NVDA) discrete GPUs in the data center space. This means that if it doesn’t catch up with the changing landscape, Intel could lose ground in the FPGA market.