Last month Google Cloud became the latest public cloud provider to reveal it was deploying chips from one of the newest names in the semiconductor market. Ampere Computing has seen its profile rise sharply in the past 12 months, with some of the biggest names in the cloud adopting its silicon, which is specifically designed for data centres.
Ampere’s Arm-based Altra processors can be found in Google Cloud’s new Tau T2A Virtual Machines. The company says these will offer a wide range of network-attached storage options, and can be used for scale-out workloads including web servers, containerised microservices, data-logging processing, media transcoding, and large-scale Java applications.
“Developers now have the option of choosing the optimal architecture to test, develop and run their workloads,” said Sachin Gupta, vice president and general manager for infrastructure at Google Cloud.
Google is far from the first company to take a punt on the potential of Ampere chips, with several other big names in the sector having already installed them in their data centres. And though this popularity with the hyperscalers may prove to be short-lived, with many developing silicon of their own, the company, which is heading for IPO later this year, is well placed for success.
Who is Ampere Computing?
Ampere Computing is little more than four years old, having been founded in Silicon Valley in February 2018 by a group of Intel engineers led by Renee James, the chipmaking giant’s former president.
The company was set up with a remit of designing processors to underpin cloud computing. “We have an opportunity with cloud computing to take a fresh approach with products that are built to address the new software ecosystem,” said James at the time. “The workloads moving to the cloud require more memory, and at the same time, customers have stringent requirements for power, size and costs. The software that runs the cloud enables Ampere to design with a different point of view.”
Despite its roots, Ampere zeroed in on using the chip architecture designed by British company Arm, rather than Intel’s rival x86 framework. Smaller, more efficient and, for certain workloads, more effective than their x86 counterparts, Arm chips have many of the qualities required by cloud providers looking to run increasingly complex and specific operations for their customers.
“Arm chips are good at efficiently carrying out small tasks, and the software ecosystem transitioning to containerised and serverless workloads has presented an opportunity to leverage the Arm architecture,” says Manoj Sukumaran, principal analyst for data centre computing and networking at Omdia. “The majority of cloud workloads, such as web service interfaces, can be executed on an Arm core much more efficiently than an x86 core.”
Ampere’s first “cloud native” processor, the Altra, was unveiled in 2020, and since then the Altra Max has followed. Boasting 80 and 128 cores respectively, the processors first caught the eye of Oracle, which has been using them in its data centres since last year. In April, Microsoft Azure announced its first Ampere-based instances, while Chinese companies Tencent, Alibaba and ByteDance, owner of TikTok, are among its other customers.
With interest running high in its products, Ampere filed for an IPO in April, and expects to go public later this year. The filing revealed that Oracle has become one of the company’s biggest backers, with Larry Ellison’s company investing some $426m in Ampere. The company has made a confidential filing, which means details of its income and other financial information remain undisclosed.
“Ampere is designing for a completely different goal,” to its rivals, says Dylan Patel, a chip industry analyst from SemiAnalysis. “Intel is targeting a wider net of people, versus what Ampere is going for, which allows them to make certain sacrifices and gain benefits from that,” he says.
Though the company’s chips have a “weaker and smaller CPU core” than some of the x86 processors designed by Intel and AMD for servers, Patel says this means the chips themselves are smaller, and that as a result power usage is lower and the chips are more efficient. “That’s a big deal” for data centre operators, he says.
Patel adds: “Renee realised this needed to be done much earlier than most of the cloud providers themselves. Amazon figured out they needed to build CPUs like this, but the others did not, and as a result, most of them are now in a position where they need to buy the technology from someone else. That’s why you find Ampere in every major cloud now.”
The changing face of the server chip market
Indeed, the cloud market leader, Amazon’s AWS, is the only hyperscale cloud provider to eschew Ampere in favour of its own Arm-based chips, the Graviton series. The company has invested heavily in semiconductor R&D since its $350m acquisition of cloud chip start-up Annapurna Labs in 2015. Market analyst Trendforce says 15% of AWS servers ran on Arm-based chips last year, and this figure is expected to exceed 20% in 2022.
In the wider server chip market, traditionally dominated by x86 semiconductors from Intel and AMD, Arm processors are likely to account for 22% of cloud workloads by 2025, Trendforce says, having occupied less than 1% of the market in 2020.
Aside from the power and efficiency advantages, Trendforce research manager Mark Lui says Arm devices have two additional advantages compared to x86. “Arm-based processors provide higher customisation for different niche markets with a more flexible ecosystem,” Lui says, adding that “physical footprint is relatively small which meets the needs of today’s micro data centres”.
Indeed, geopolitical tensions have heightened the importance of data sovereignty – ensuring that citizens’ cloud data is kept safe and secure within national or continental boundaries – to governments. Tech Monitor has reported on efforts by cloud providers to offer heightened levels of protection, with Microsoft last month launching its Cloud for Sovereignty product to try and offer fresh assurances to public sector clients.
Lui says this push for cloud sovereignty will lead to more micro data centres being constructed, which are perfect environments to run chips such as Ampere’s. “Influenced by geopolitics and the strengthening of data sovereignty in various countries, major cloud service providers and telecom operators are actively developing micro data centres which will further drive the penetration of Arm-based processors,” he adds.
Beyond IPO: What does the future hold for Ampere Computing?
With demand for Arm chips in the data centre on the rise, Ampere is unlikely to be short of customers any time soon. Google, Microsoft and Tencent are all now working on their own custom silicon for servers, but Patel says the first iterations of these processors are unlikely to hit the mark.
“Even for the major, established, companies, many chips don’t work out,” he says. “Amazon had a first generation Graviton chip which no one used. And even if they can prove their in-house efforts work and can be cheaper or more effective for certain types of computation, they also need to prove the software around them works.”
Because of this, Patel says the companies will need back-up plans. “If [the cloud providers] first-generation chips don’t work out, they’ll use Ampere,” he says, adding that the company’s future could hinge on the success of its second-generation chips. While the Altra and Altra Max have been based on Arm’s Neoverse1 blueprints, the upcoming AmpereOne, due to be released later this year, has been designed from scratch by the company.
“AMD and Intel are now working on their own ‘cloud native’ designs, Bergamo and Sierra Forest respectively,” Patel says. “They’ve woken up and decided to try Ampere’s strategy, so the question is can Ampere out-engineer them? It will come down to whether their team is good enough and can make a success of that second-generation chip, and if they can do it – and I think there’s every chance they will – then we could see them gain market share and become a significant player in the server market.”
Omdia’s Sukumaran says that while Ampere’s popularity with the hyperscalers may not last, it will stand the company in good stead. “By the end of 2023 it’s likely both Microsoft and Google will have their own CPUs,” he says. “So the opportunity for Ampere is likely to be limited because they will prioritise their own chips.”
But by working with the public cloud hyperscalers, Ampere has proved the quality of its technology, he argues. “I admire the company’s leadership because they wanted customers that could deploy their CPUs straight away, and the hyperscalers are well equipped to do that,” he says. “These wins at the moment are very strategic for Ampere, because they show the OEMs that the technology works. And if you look at a three-five year horizon, the Arm architecture will be much more mature than it is today [for data centres].”
This, Sukumaran says, will open up new opportunities for Ampere. The company announced it had landed its first OEM server maker in June when HPE revealed it was building a rack server utilising the Altra Max processor, and it is likely to be targeting more similar deals in future.
“People won’t be buying their server chips from Microsoft or Google,” Sukumaran says. “They’ll buy them from Ampere, and being an independent vendor could open up the broader enterprise or telco markets, as well as edge computing. So when I look at Ampere I see a company with a strategy that’s very well lined up.”