Big cloud winner will be based on who can make the best custom ARM chip

In the past, AMD and Intel made all the server chips. There's very little to differentiate between AWS, Google Cloud, Azure except for scale and execution. They all ran the exact same Intel Xeon and AMD Epyc chips.

That all changed when Amazon started prioritizing their Graviton2 ARM chip. In 2021, 50% of all new EC2 instances were based on Amazon's own CPU, not Intel or AMD. I'm willing to bet that Graviton2 will have an even higher marketshare in 2022. Graviton2 offers 34% better price/perf ratio than x86 chips. Amazon's own claim but it's credible.

Microsoft and Google have also started designing their own custom server CPUs. It makes a lot of sense. Intel and AMD mark up their server CPUs by 5-10x the cost to manufacture them. In the past, Intel was a monopoly and Amazon, Microsoft, and Google had no choice but to pay high prices for Xeon. If AMD wins, then AMD will just replace Intel as the monopoly. Big cloud providers want to control their own destiny and not have either Intel or AMD grab them by the balls. And they want to lower cost and differentiate based on performance. They also have enough money and scale to design and manufacture their own chips.

This will have profound impact on big cloud and semiconductor companies:

  • Whoever makes the best custom ARM CPU will likely win the cloud war
  • Right now, AWS is in the lead for custom server CPUs because they started first and is successful with their own chips
  • x86 dominant days in the server are numbered
  • Many analysts are pricing in AMD's expected increase in server marketshares. But what they don't really understand is that big cloud providers don't really want another x86-monopoly. Custom ARM server chips are disrupting AMD before AMD can finish disrupting Intel.
  • Intel, wisely, understands this. That's a primary reason they opened up their fabs to third parties. Intel wants to still manufacture Amazon, Google, Microsoft's chips. Yes, the margins are lower than selling them 5-10x marked-up chips, but at least Intel will still be in the game.
  • TSMC will be another winner from this because these companies will bid for wafers
  • AMD and Intel are the losers. But I expect AMD to lose more and Intel to lose less.

Disclosure: I own Apple stocks, a small amount of Amazons stocks, a very small amount of Intel stocks. I'm waiting to see how Intel's manufacturing will progress before buying more or selling what I have. I'm also waiting to see how Microsoft and Google's custom CPUs perform before going all in on one of the big cloud companies.

PS. Keep in mind that Wallstreet is very late to react to how chips perform. For example, it was abundantly clear that AMD's Zen2 was going to beat Intel's ass in the servers from the moment AMD announced it during CES 2019 (Jan 8 2019). But Wallstreet didn't fully react until 1+ year later. Also, it's abundantly clear that Intel's newest Alder Lake chip is going to beat AMD's ass in the consumer laptop and desktop market. But Wallstreet hasn't reacted at all. This means you can wait for benchmarks on Amazon, Microsoft, Google's custom CPUs before Wallstreet catches on.


Leave a Reply

Your email address will not be published. Required fields are marked *