TrendForce: AWS to give Arm a leg up to 22% of datacenter servers by 2025
Cloud giant's Graviton forcing the hand of rivals, according to analysts
The increasing adoption of Arm-compatible processors by cloud service providers is expected to grow the CPU architecture's penetration rate in datacenter servers to 22 percent by 2025, according to a report by Taiwan-based research firm TrendForce.
The report, released Tuesday, pointed to Amazon Web Services' growing footprint of Arm-based cloud instances as a major catalyst for growth of the British chip designer's CPU blueprints in servers, and said it's forcing competing cloud providers to play catch-up with their own in-house chip designs. There's no mention in the report of how Nvidia's failure to acquire Arm, which fell through last month, and Arm's subsequent plan to re-enter the public market could impact server adoption.
Anyhow, these homegrown Arm-compatible chip projects are giving cloud providers more flexibility as they face greater demand in areas like artificial intelligence and high-performance computing. This is one big reason why TrendForce believes Arm adoption in the datacenter will continue to grow.
"If testing is successful, these projects are expected to start mass introduction in 2025," the report said.
TrendForce said AWS's roll out of Arm processors reached 15 percent of its overall server deployments in 2021, and is forecasted to surpass 20 percent this year. Whether that share is completely made up of AWS' homegrown, Arm-based Graviton chips was unclear, but the research firm said that's not the only factor impacting Arm's growth in the datacenter market.
- Nvidia reveals 144-core Arm-based Grace 'CPU Superchip'
- Nvidia CEO Jensen Huang talks chips, GPUs, metaverse
- Microsoft targets multicloud with Arc-enabled Azure Stack HCI
- SiFive bags $175m to further challenge Arm with RISC-V
The research firm said geopolitical concerns and the increasing need for countries to keep data within their borders are also working in Arm's favor, driving the development of Arm-based "micro datacenters" by telecommunications firms and cloud providers. Arm's Neoverse server CPU designs are well-suited for edge computing environments like this, as well as "ultra-large-scale datacenters" that are stood up by cloud providers, according to TrendForce.
However, the firm added, with these two focus areas and the x86 architecture of Intel and AMD still dominating the larger datacenter market, the introduction of Arm CPUs into enterprise datacenters is expected to move at a slower pace. For this reason, TrendForce said it doesn't expect competitive Arm-based servers in the enterprise market until 2025.
Helping hand from AWS
AWS first introduced cloud instances running on its Graviton chips in 2018. It claims it's a more efficient alternative to x86-based cloud instances that are dominated by Intel but increasingly powered by AMD.
Originating from AWS' 2015 acquisition of Israeli chip startup Annapurna Labs, the latest generation of its homegrown Arm processors, Graviton3, entered preview mode last December, promising 20-80 percent higher performance and 35 percent less latency than Graviton2 chips in EC2 instances.
But Graviton isn't the only type of Arm-based CPUs used by AWS. The cloud provider recently started previewing a new EC2 instance for virtualized Macs powered by Apple's M1 CPU. However, these instances likely haven't made that big of a dent in AWS' datacenter footprint, given how new they are.
TrendForce didn't say which of AWS' competitors are playing catch-up in the homegrown chip design department, but we do know that, as of December 2020, Microsoft was reportedly designing its own Arm-based chips for Azure datacenters and Surface PCs.
More definitively, Google said last year it was broadening its custom chip efforts beyond the homegrown Tensor Processing Unit to work on server chips, although we don't know if Google plans to use Arm's instruction set architecture.
Last fall, it was reported that Chinese retail and cloud giant Alibaba introduced its own Arm server processor. This will support China's desire to own more of the supply chain for high-performance datacenters, as our sister site The Next Platform has discussed. Chinese tech giant Huawei is developing Arm-based server chips too, as we pointed out back in 2019.
Alibaba is also part of a handful of cloud provider that have turned to Arm-based CPUs designed by other companies. In this case, we're talking about chip startup Ampere Computing, whose Altra CPUs have also been picked up by Oracle and Equinix in the US. Ampere's chips have also been adopted by Tencent Cloud, JD Cloud and UCloud in China.
It's important to point out that there isn't complete agreement on how big Arm could get in servers in the next few years. For instance, research firm Omdia said last August that it expects Arm to account for just 14 percent of servers by 2025.
The TrendForce report does not mention RISC-V, an alternative, open-source chip instruction-set architecture that competes with Arm. But that makes sense, given that SiFive, the largest RISC-V chip designer, recently told The Register that it doesn't plan to see RISC-V server chips in market for at least a few more years. Another RISC-V chip designer, Ventana Micro Systems, is also planning to make server chips in the coming years. ®