(Reuters) — Amazon’s cloud computing unit has designed a second, more powerful generation of datacenter processor chip, two sources familiar with the matter told Reuters, the latest sign that the company is pouring money into custom silicon for its fastest-growing business.
The new Amazon Web Services chip uses technology from SoftBank Group-owned Arm Holdings, the sources said. One of the sources familiar with the matter said it will be at least 20% faster than Amazon’s first Arm-based chip, named Graviton, which was released last year as a low-cost option for easier computing tasks.
If Amazon Web Services’ chip efforts are successful, it could lessen the unit’s reliance on Intel and Advanced Micro Devices for server chips.
An Amazon spokesperson declined to comment on future products or services. Arm likewise declined to comment.
In cloud computing, businesses rent out servers from Amazon instead of running their own datacenters. Analysts expect Amazon’s cloud unit to generate $34.9 billion in sales in 2019, according to IBES data from Refinitiv.
Cloud computing has become big business for datacenter chipmakers. Intel controls more than 90% of the server processor market, with AMD controlling most of the remainder. Intel’s datacenter group generated almost half of the company’s overall operating profit last year.
And most server chips go to the cloud. In 2018, almost 65% of Intel’s datacenter chip sales were from cloud and communications service providers, its executives have said.
Cost of ownership
Chip designers using Arm technology want to challenge Intel’s dominance. Arm chips power mobile phones today. But several companies aim to make them suitable for datacenters, including startups run by former executives from Intel and Apple.
Amazon’s first Arm chip did not appear to have an impact on Intel’s datacenter business, which continued to grow over the past year, said Bernstein analyst Stacy Rasgon. But Rasgon said major technology companies, which spend billions each year with Intel and AMD with few alternatives, have the resources to make more powerful Arm chips.
“Arm by itself I’m not worried about, but Arm in the hands of an Amazon or a Google who could potentially invest in it, that becomes potentially more problematic [for Intel],” Rasgon said.
Amazon’s Arm effort appears to be making progress, one of the sources said. The new chip’s speed gain “sends a message to the market” that Amazon is serious about investing in Arm-based chips, the person said.
Both sources familiar with the matter said the new chip is not expected to be as powerful as Intel’s Cascade Lake or AMD’s Rome chips.
Though less powerful, Arm chips are cheaper and consume less electricity than Intel’s top-end chips. Intel’s most powerful chips can cost several thousand dollars, while barebones Arm-based server chips can cost less than $1,000.
In a datacenter that houses tens of thousands of servers, chip buyers often focus on a mix of factors — speed, chip size, power consumption, and cooling costs — called the “total cost of ownership.” That is where Arm-based offerings hope to one day compete with Intel.
New arm technology
Amazon’s first Graviton chip used Arm’s older Cortex A72 technology. The forthcoming Amazon chip is expected to use newer Arm technology, most likely Arm’s Neoverse N1 technology, one of the sources familiar with the matter told Reuters. Another source familiar with the matter said the chip is expected to have at least 32 cores versus the Graviton’s 16.
The new chip will also use a technology called a “fabric” that will allow it to connect with other chips to speed up tasks like image recognition, one of the people familiar with the matter said.
To take advantage of the new chip, cloud customers will likely need to use software written for Arm-based chips, which is less common than software for Intel and AMD chips.
“The hardware is only part of the equation,” the second source familiar with the matter told Reuters.
(Reporting by Stephen Nellis in San Francisco, additional reporting by Jeffrey Dastin in San Francisco, editing by Greg Mitchell and Lisa Shumaker.)