For years, investors and analysts have wondered about the potential impact of ARM-based servers on Intel's (NASDAQ:INTC) dominance of the server microprocessor market. ARM itself recently predicted that, by 2020, its partners would be able to capture roughly 25% of the total server processor market -- up from approximately 0% of the market in 2016.
Even though ARM-based servers don't appear to have gained much steam over the past several years, it's very important that Intel not assume it has nothing to worry about.
Indeed, even if Intel can outpace its fellow merchant processor suppliers -- Qualcomm (NASDAQ: QCOM) and Cavium (NASDAQ: CAVM) seem to be emerging as the front-runners in the merchant ARM server processor market -- there's still another threat that Intel needs to deal with: in-house chips produced by Intel's major customers.
The true threat of ARM
ARM does not design and sell complete chip designs. What it does is produce key intellectual properties that licensees can draw upon to put together chips that are then built by third-party contract manufacturers such as Taiwan Semiconductor Manufacturing Co. Ltd. (NYSE: TSM).
These intellectual properties include instruction set architectures, processor cores, interconnect technologies, physical IP, and other key ingredients required to build competent chips.
What this allows Intel's current customers to do is to license the technologies that they don't want to build themselves, build differentiating technologies in-house without needing to reinvent the wheel, and then integrate all of that into a chip customized to that customer's exact needs.
Of course, building server-grade processors is no trivial task, but if ARM is able to do a lot of the heavy lifting, doing so could go from "too hard" to "feasible." Indeed, it's important to remember that some of Intel's top cloud server customers, such as Facebook (NASDAQ: FB), Alphabet (NASDAQ: GOOGL) (NASDAQ: GOOG), and Microsoft (NASDAQ: MSFT), are much larger than Intel and have the deep pockets required to get it done -- if it makes sense for them to do so.
How Intel can respond
The bad news is that Intel's biggest customers probably have the wherewithal to invest in building their own chips if there is a long-term total cost of ownership (TCO) benefit to doing so.
The good news is that it's probably only Intel's biggest customers with the necessary scale to justify custom chips; it'd be very hard for smaller customers to make the economic case for building their own chips. Even if they tried to do so anyway, their odds of success may be much lower than it would be for one of the larger companies.
For Intel to minimize the threat of having major cloud service providers build their own chips, Intel needs to deliver products with lower TCOs than what those cloud-service providers could build themselves.
That means having the most power-efficient processor and chip architectures, the most power-efficient chip manufacturing technologies, and the ability to build processors to meet its large customers' exact needs.
More specifically, Intel needs to build better server-grade CPU cores and related intellectual properties than ARM can; create better manufacturing technologies in terms of performance, power, and area than what TSMC, the leading contract chip manufacturer, can; and foster a robust product development methodology that allows Intel to rapidly design custom chips that stitch together Intel's in-house technologies and potentially third-party technologies.
Intel's continued increases in research and development spending related to its data center group suggest that the company is investing heavily to be able to achieve all that I've outlined here. We'll see over the next five years just how effective Intel will be at keeping its large customers from wanting to roll their own chips.