Last month reports started to emerge in the press that Intel had lowered prices for its server CPUs. Digitimes broke the news and Tom’s hardware has a good write-up that’s not behind a paywall. We had been hearing rumblings about this all summer, and now it has broken out into the open. This is fairly significant, and not a great sign for Intel.
Intel’s server products make up roughly half of its profits, on roughly a fifth of revenue. As that ratio implies, Intel’s server chips command pretty healthy profit margins. This is the result of Intel’s decades long near-monopoly on this multi-billion dollar “niche” market. Intel has been one of the biggest beneficiaries of everything moving to the cloud; cloud data centers do a lot of heavy compute and for a long time that meant x86 Intel Xeons. That monopoly position has eroded fairly seriously in recent years with Intel giving meaningful share to AMD, and a lot of cloud scale demand shifting to GPUs and AI accelerators, including internally built chips at Amazon and Google (and soon Facebook). We would also add that privately-held Ampere has apparently been pricing its flagship below comparable Intel SKUs, something that we suspect they are capable of doing because they can provide a very powerful total cost of ownership advantage over Intel’s Xeon.
Going back decades, Intel has built a solid playbook for confronting share loss. This is not the first time they have seen some rival (often AMD) gain some share for one reason or the other. Intel then rolls out this playbook to claw back that share. However, this time is different. In the past, Intel’s share loss could easily be seen as temporary, just long enough until they spun up the next generation of their chips. Now, their manufacturing process is not firing on all cylinders and it is unclear when their core products will be competitive again, likely not until 2023 when (if?) they get x86 into TSMC’s 3nm process, at the earliest.
Still it is worth walking through this playbook to understand just how vulnerable Intel is. The quick summary is that “cutting prices” is the last chapter of that playbook, so this is starting to feel like end game maneuvering.
The playbook runs something like:
- Spread Fear, Uncertainty and Doubt (FUD) about competitors
- Tug at the channel
- Push the software moat
- Cut prices
Not so long ago, every large Intel customer was afraid to even to think about using an alternative. If Intel found out that a customer was testing samples of a competitive product, Intel would quickly send angry messages to that customer. (And Intel always found out.) Customers were worried about losing preferential pricing, early access to chips and a host of other “customer service” factors; and so they would quickly back off that testing. Then Intel would spread further FUD – leaking performance data (real or hypothesized) showing Intel’s advantages and generally casting doubt on the competitive product. All of this is immensely powerful. It made it very hard for customers to even consider alternatives clearly.
If that failed, Intel could then push harder on the channel. All the distributors, ODMs and systems integrators do a lot of business with Intel’s products, and would be reluctant to engage fully with competitors. These companies have to look out for their best interests, so they would often do some work with non-Intel parts, but almost always this work would be done by the back-up team, under-resourced and generally ambivalent.
Intel would also push out a series of announcements and optimizations for various bits of software. Intel has had 20+ years to build up a formidable software moat around its products. Software companies invested heavily in making sure their products ran as fast as possible on Intel, because that is what their customers needed. To this day, this remains Intel’s best defense against competition, even against AMD. That being said, the distance between Intel and the software ecosystem has grown considerably in recent years. Many customers, especially the big cloud service providers, build much of their own software themselves and are perfectly capable of doing the requisite optimization work themselves. These customers were deeply ambivalent about testing out Intel alternatives because of the amount of work required, but with Intel’s growing performance gap they are now much more motivated in assigning the teams needed to do this work. We should also point out that every time we raise this subject we usually have several people reach out to us and say “it is easy to port software today”. We maintain that there is still some complexity involved, but the point remains – Intel’s software moat is drying up.
Finally, once all those avenues were exhausted, Intel would cut prices. Often this was done very tactically, everyone NOT testing competitive products got those discounts first. And as we said, these price cuts would usually only create temporary margin pressure for Intel.
We still maintain that it is too early to write Intel off. They have a lot of money, and a lot of influence (not least in Washington), and a lot of very smart, talented people. They may find a way out of this, but price cuts on their flagship products are not a good sign.