FPGA world: Intel is weaker than Xilinx
Editor | Bao Yonggang
Leifeng.com: With the acquisition of two giants, Altera and Xilinx, the competition in the FPGA industry has undergone subtle changes. In the near future, it may become more difficult to understand this market.
As AMD's $3.5 billion acquisition of FPGA company Xilinx nears completion, it's natural to think about the prospects for FPGAs and how AMD will compete with its main rival Intel - specifically its Programmable Solutions Group (PSG).
PSG's predecessor was the independent company Altera, which was acquired by Intel for US$16.7 billion in June 2015.
Xilinx shareholders should probably be grateful to Intel for its acquisition of Altera, which has led to an increase in the market value of Altera and Xilinx. But in fact, Xilinx's market value increase and significant profit growth over the past five years are its own achievements.
Since its founding four decades ago, Altera and Xilinx have been at the forefront of the industry in terms of chip manufacturing processes, FPGA architectures, and business scale, which is healthy for the FPGA market. But in recent years, Altera has come to rely on Intel's chip manufacturing technology and became Intel's first large foundry partner until Intel decided to acquire the entire Altera company. As we all know, Intel has encountered difficulties in its 10nm manufacturing process and is currently lagging behind its competitor TSMC's 7nm and 5nm processes in advanced processes.
Some people therefore compare Altera to a frog whose legs have been cut off and cannot jump away. Industry insiders believe that if there is no acquisition by Intel, Altera will eventually turn to TSMC or even Samsung to use their most advanced technology to manufacture chips (just like IBM once did).
Today, the leapfrogging has stopped, so it’s no surprise that Xilinx’s revenue and profit trajectory is roughly the same as Intel predicted for Altera when it acquired the company seven years ago.
At the beginning of 2022, there is still a significant competitive relationship between the two companies, but it is more important to realize that it would be extremely unwise to exclude Intel from the technology or marketing competition at this time.
Since we know very little about Intel's PSG division, we'll talk about it first. This is especially important because Intel is reorganizing its groups and divisions.
As part of the reorganization plan, the FPGA department will be merged with the data center CPU department to form a new data center and artificial intelligence department. At the same time, the IoT product department and Ethernet switching department, which are closely related to the original data center, will be merged into the network department and the Edge department. Finally, like AMD, Intel will place the GPU computing business in the data center of the graphics computing department and call it the accelerated computing and graphics department.
Mobileye and IFS will operate independently, while the flash memory business will be separated. It is not clear where the Optane 3D XPoint memory business will go, but it is also possible that it will belong to the new data center or artificial intelligence division. Intel will further disclose the plans of these new divisions at its "Investor Day" on February 17.
It is foreseeable that we will soon lose our understanding of Intel's PSG department. Similarly, once AMD swallows up Xilinx, we may also soon lose our understanding of Xilinx's FPGA business. Because this business is likely to be incorporated into AMD's embedded and semi-custom department (after the acquisition of Xilinx is completed, AMD may set up a new data center department and integrate CPU, GPU, FPGA and other departments).
In any case, PSG's quarterly sales for the period ended December 2021 were $484 million, up 14.7% year-over-year, and revenue was $51 million, up 18.6% year-over-year. Based on Intel's revenue and net income ratio, we estimate that PSG earned approximately $47 million in net income.
Here's how the Altra/PSG business has fared since the Great Recession of 2009:
Intel PSG has an average quarterly sales of about $450 million. On average, net income is about one-fifth of sales. But in the past twelve months, Intel PSG's sales were $1.93 billion, up 4.4% year-on-year, and net income is expected to reach $302 million, up 33.4% year-on-year. The department's revenue accounts for 15.6% of total revenue, which is lower than the historical average since the Great Depression.
It is worth noting that during a conference call with Wall Street investors on the fourth quarter 2021 results, David Sinzner, the new Intel CFO who just resigned from Micron Technology, said: "If it were not for external supply constraints, we believe that the PSG business would bring in more than $500 million in revenue in 2021."
We can’t make a comparison without knowing exactly how constrained Intel was in FPGAs in 2020. But if Intel hadn’t been constrained by its Stratix and Agilex product lines in 2020, its growth would have been more impressive than the 31.4% growth and $2.34 billion in sales for the FPGA division for the full year 2021.
But all this is just a fantasy. The NextPlatform believes that Xilinx can take advantage of Intel's shortcomings in FPGA sales because it has gained the support of TSMC's wafer manufacturing packaging process in its cooperation with TSMC. In the past year, Xilinx's products and services brought it $3.68 billion in sales, a year-on-year increase of 20.4%, and its net profit increased by 49.6% to $929 million. If $500 million in FPGA sales were transferred from Intel to Xilinx, Xilinx's revenue growth last year might have been only 4.1%, about $3.18 billion. (We are not saying that the revenue transfer is so simple that it goes directly from Intel's sales to Xilinx's sales. In fact, Intel has been able to pass on part of the loss of this $500 million in sales to 2022.)
In the third quarter of fiscal 2022 ending January 1, Xilinx announced its first "unicorn" quarter. In the quarter, sales reached $1.01 billion and net income reached approximately $300 million, accounting for 29.7% of sales. Xilinx has had better performance in the past quarters, but at that time Xilinx was a much smaller company than it is now. The NextPlatform believes that Xilinx is benefiting from AMD's acquisition of it and Intel's chip shortage. At the same time, based on the architectural and process advantages it brings to the "Mount Everest" versa FPGA hybrid chip, it sells it on its own advantages.
Clearly, Xilinx was able to grow, while Intel PSG had been pretty mediocre in recent years. This resulted in Xilinx having more than twice the sales of Intel PSG, and about 6 times higher earnings. With the Great Recession, Xilinx was suddenly about 50% larger than Altera, and it remained that size advantage until Intel acquired Altera. After Intel acquired Altera, Xilinx's growth took a hockey stick shape to the right and up.
Xilinx has burned some money on acquisitions. In 2018, Xilinx acquired DeePhi Technology. In 2019, Xilinx acquired Solarflare, a large smart network card manufacturer. But as you can see, now they are putting the coins back into the piggy bank.
One of the factors driving Xilinx's growth is that the manufacturing process used in its chips is ahead of Intel's PSG division. As shown in the figure below, Xilinx's increasing revenue comes from FPGA hybrid chips (called hybrid because they have programmable logic blocks surrounded by hardware-coded blocks for computing and networking) using more advanced processes.
Xilinx's data center business performed particularly well in the most recent quarter, growing 28% sequentially and 81% year-over-year to $111 million in the second quarter of fiscal 2022, which ended on October 2. To keep in line with the chart Xilinx has used for years, the data center and communications portions of Xilinx's industry segment revenue stream are combined in the chart below:
The following table shows the data for the most recent quarters in this chart:
All three areas that Xilinx is pushing its hybrid FPGA chips into are growing. But given that both the communications and data center businesses have been volatile in recent years, sometimes they reach a relative high and sometimes they reach a relative low. Like other forms of high-performance computing, selling hybrid FPGA devices is a long-term battle.
One more thing to add: I hope AMD can give us a glimpse into Xilinx's business, just as the FPGA maker has done for Wall Street investors for decades. Especially in recent years, as FPGAs have become more and more important in data centers and widely used in other devices, we highly doubt that AMD has any reason not to do so. However, AMD can always surprise us.
Reference Links:
https://www.nextplatform.com/2022/02/07/xilinx-benefits-from-intel-fpga-shortages/