Custom silicon is fast becoming one of the key differentiators between public clouds and private data centers. As more public cloud vendors continue to expand their custom chip selections, private data center operators could be at a disadvantage.
Public clouds and hyperscale data center operators are increasingly turning to the development of specialized compute chips tailored for specific priorities and use cases. But will these custom processors prove to be a big enough draw to lure more businesses from private data centers into the public cloud? That remains an open question.
This article explores it by explaining what custom processors are, why they're important, and how much of an impact they may have on the data center market.
What Is Custom Silicon in the Public Cloud?
Custom silicon – also called custom chips, custom processors, or custom CPUs – refers to specialized processors that large public cloud vendors like Amazon, Microsoft, and Google build for use in their data centers.
Custom silicon chips are distinct from the standard x86 processors made by Intel or AMD that have conventionally powered most servers in both public clouds and private data centers. Most custom chips use an ARM architecture, and they are tailored for specific use cases and priorities – such as hosting AI/ML workloads or reducing power consumption.
Amazon has been at the forefront of the custom silicon movement, starting with the debut in 2018 of its line of Graviton processors. But Microsoft recently joined the fray by introducing a chip customized for AI workloads on the Azure cloud. Google, too, is reportedly working on custom silicon, which it says will become available in 2024.
What Are the Benefits of Custom Silicon?
Traditional x86 chips remain widely available for hosting workloads in the public cloud. However, opting for a virtual machine (VM) instance that uses custom silicon instead of, say, an Intel or AMD processor, offers a range of potential benefits, including:
- Lower cost – VM instances that are based on custom silicon are often priced lower than comparable x86 instances.
- Better performance – Because custom chips are optimized for specific types of workloads, better performance can sometimes be achieved.
- Reduction in energy consumption – This may help businesses improve their energy efficiency and sustainability outcomes.
So, while public cloud vendors aren't forcing their customers to use special chips, they are increasingly giving them incentives for doing so.
How Custom Silicon Impacts Private Data Centers
Currently, the custom silicon offerings from public cloud vendors are only available in the public cloud. (Amazon says that support for Graviton instances via AWS Outposts is "coming soon," which would make it possible to run Graviton-based hardware in your own data center, but only as part of an AWS-centric private cloud using servers purchased from the company.)
In this sense, custom silicon gives public cloud vendors an edge that most private data centers can't deliver. If you want to take advantage of the potential cost, performance, and sustainability benefits of custom CPUs, you must move your workloads to the public cloud.
Will Custom CPUs Drive More Migration to the Public Cloud?
Will the benefits of custom silicon drive even more workloads to the public cloud? To date, no one has collected data on this topic, and even the public cloud vendors haven't said much about which percentage of their workloads use custom chips as opposed to x86 options.
It's hard to say definitively how much of an impact custom silicon may ultimately have on private data centers. But it's not hard to envision at least some businesses deciding that migrating to the cloud is worth it, particularly if custom silicon allows them to achieve lower overall costs or better performance than they would when using conventional CPUs.
Amazon says that its custom chips can deliver cost-performance enhancements on the order of 40%, which is certainly big enough to make the public cloud compelling for businesses that have been reluctant until now to go all-in on the cloud.
On the other hand, it's possible to take advantage of non-standard CPUs – just not ones developed by public cloud vendors – in private data centers. If you are deploying your own servers in a data center you own or in a colocation facility, you can choose to use any of the wide variety of ARM chips available today. In addition, a handful of ARM-based chips (such as this one from Equinix Metal) are available through bare-metal IaaS services offered by colocation providers.
Whether more organizations abandon private data centers for the public cloud due to custom silicon depends on how much unique value the custom CPUs developed by cloud vendors deliver – as well as how hard businesses are willing to work to source non-standard CPUs for their own data centers. Launching a Graviton instance on Amazon EC2 is no harder than launching an x86-based instance, but finding an ARM-based server to deploy in a data center can be tricky.
Conclusion
The bottom line is that it's too early to say whether custom silicon will exert a major impact on the data center industry. But this is certainly an area worth watching as cloud vendors continue to expand their custom silicon offerings in a bid to differentiate themselves further from private data center facilities.
