The age of the data center being dominated by server CPUs from just two companies, Intel (INTC -1.63%) and AMD (AMD 1.78%), is coming to an end. Arm-based chips, largely a no-show in the server CPU market until recently, are having a moment.
Alphabet's (GOOG -3.33%) (GOOGL -3.37%) Google announced earlier this week that it had designed a custom Arm-based CPU that will be widely deployed inside the data centers powering Google Cloud. Microsoft made a similar announcement last year, and Amazon Web Services has long offered instances using its homegrown Graviton CPUs.
"Gradually, then suddenly" is a good description of how Arm-based chips have found their way into the data center. There have been many failed attempts -- Qualcomm tried and failed way back in 2017 with its Centriq CPUs. But with all three major cloud giants now embracing custom Arm CPUs, the server chip market appears to be at an inflection point.
A race to lower costs
Cloud computing is expensive. One reason is that server CPUs haven't improved all that quickly in terms of performance per dollar or efficiency in recent years. That's partly a consequence of the Intel-AMD duopoly.
Google's aim with its Axion server CPU is to provide solid performance and exceptional energy efficiency. While the upfront cost of a single server CPU can be in the thousands of dollars, the cost of energy usage over that chip's lifetime is a major component of the total cost of ownership.
Google is making some big promises. The company claims that its Axion processors, as part of a virtual server on its cloud platform, are capable of providing up to 50% greater performance and 60% better energy efficiency than virtual servers powered by comparable x86 CPUs from Intel or AMD.
This all translates into lower cloud computing costs for both Google and its cloud customers. With AWS and Microsoft already going down this road, Google risked being left behind if it didn't follow suit.
Trouble for Intel and AMD
Intel has more to lose than AMD as Arm-based server chips proliferate but has more to gain, as well. The company remains the server CPU leader with about three-quarters of the market, so it may take a bigger hit than AMD as the cloud giants expand their use of Arm-based chips.
Both Intel and AMD have or will soon offer products aimed at competing directly with energy-efficient Arm-based server CPUs. AMD already has its "Bergamo" CPUs on the market, which use cut-down versions of its standard server CPU cores and pack in as many as 128 cores in a single chip. Meanwhile, the launch of Intel's Sierra Forest server CPUs, comprised entirely of low-power efficiency cores, is happening this quarter.
While the near-term situation may be worse for Intel than AMD, Intel has a foundry business and AMD does not. AMD gave up on doing its own manufacturing long ago, while Intel has doubled down in recent years as it aims to become the world's second-largest foundry by 2030. An agreement with Arm Holdings to co-optimize the Arm architecture and Intel's upcoming Intel 18A manufacturing process will start to pay off in 2025 and beyond as that process scales up.
Intel already has a deal to manufacture a custom chip for Microsoft using Intel 18A, although it's not clear whether it's an Arm-based server CPU or something else. As Intel 18A ramps up in 2025 and 2026, Arm-based server CPUs will likely be rolling out of Intel's manufacturing plants.
For AMD, there's no real upside to the Arm-based invasion of the data center. The company could opt to design its own Arm-based chips, but that doesn't solve the core problem of the cloud giants opting for custom chips.
Google's push to install custom Arm-based CPUs in its data centers is certainly bad news for both Intel and AMD. In the long run, though, Intel has a great way to benefit from this trend with its foundry business, something that AMD is sorely lacking.
