Intel’s data center dominance may be slipping as hyperscale and cloud data centers increasingly opt for competing chips from AMD and Arm chipmakers. Google Cloud Platform in particular has become increasingly reliant on chips from team red in recent years. The cloud provider continued that trend today with the launch of its latest family of virtual machines (VMs) powered by AMD’s EPYC 3 CPUs.

The cloud provider bills the EPYC-based Tau VMs as a “price-performance optimized” option that's ideal for scale-out workloads like Kubernetes clusters. The VMs will be available for use in Google Kubernetes Engine from day one, which is planned for the third quarter.

Each VM can be configured with up to 60 virtual CPUs and up to four gigabytes of memory per core. This, Google asserts, enables Tau VMs to provide 40% better price to performance than any other public cloud provider, a claim that Oracle founder and CTO Larry Ellison would, no doubt, contest.

Ellison made similar claims about Oracle Cloud Infrastructure’s Arm-based data center chips during the company’s quarterly earnings call earlier this week.

“Ampere delivers much better compute cost performance than either Intel or AMD, and by far the lowest energy usage of any server microprocessor in the world,” he said.

Founded in 2017 by former Intel President Renee James, Ampere has emerged as a disruptive force in the data center computing space. To date, the fledgling chipmaker has won several high-profile contracts with major cloud and hyperscale providers including Oracle, Microsoft, Equinix, and Cloudflare.

Google’s Tau VMs still deliver an attractive combination of performance and price compared to competing cloud providers, according to Matt Eastwood, IDC SVP of enterprise infrastructure, cloud, developers, and alliances, in a statement. “The initial testing results for Tau VMs demonstrate the ongoing commitment Google has to innovating in scale-out platforms.”

Google, unlike its competitors, has yet to embrace Arm CPUs in the data center, and software compatibility may be a culprit.

“The AMD EPYC processor-based VMs also preserve x86 compatibility so that customers don't need to waste precious technical resources and time redesigning applications and instead can immediately take full advantage of x86 processing speed and ecosystem depth,” the company said.

However, software support for Arm-based data center chips is rapidly improving thanks in part to the open source community. VMware this week also signaled its intent to bring its ESXi hypervisor to Arm platforms.

Despite growing competition from rival chipmakers like AMD and Ampere, Intel remains a formidable force in the data center and hyperscale markets.

According to S&P Global, Intel retains the lion's share — north of 90% — of the data center CPU market.

Intel’s data center stronghold may under pressure, even declining “a bit,” but people shouldn’t be interpreting that as a death knell for the chipmaker, Glenn O’Donnell, VP and research director of infrastructure and operations at Forrester Research, told SDxCentral. “Don’t count Intel out, especially since I expect Pat Gelsinger to make Intel much more competitive than it has been in recent years,” he said.

Intel is well aware of the competitive landscape and appears to be taking steps to mitigate those threats, added Wayne Lam, senior director of research for the Americas at CCS Insight. “I believe they are attempting to build up a competitive solution for data centers and target applications like high performance compute and artificial intelligence,” he said, pointing to a rumored bid to buy RISC-V chipmaker SiFive.

The pressure on Intel, however, is unlikely to let up as cloud providers like Amazon continue to develop custom silicon to accelerate workloads and cut costs.

Last December, AWS CEO Andy Jassy, who is slated to take over as Amazon CEO on July 5, said the cloud provider’s future would be paved in custom silicon. At the company’s AWS re:Invent conference last winter, Jassy announced a series of new Graviton2-based VMs alongside a partnership with Intel’s Habana Labs.

However, in the same breath, Jassy made it clear the partnership would be a stop-gap measure until Amazon's own custom AI training processor, dubbed Trainium, was ready for prime time. “Trainium will be even more cost effective than the Habana chip,” Jassy said.