About a decade ago, executives with ARM said they saw an opportunity to move the company’s low-power chip designs – which had become the dominant architecture in the rapidly expanding smartphone and tablet markets – up the ladder and into increasingly dense and energy-hungry data centers.
After some stops and starts and years of building up the supporting software ecosystem, the company is gaining traction in a server space that is becoming more diversified. Top system makes like Hewlett Packard Enterprise, Dell and Lenovo are offering hardware powered by ARM-based processors, hyperscale cloud provides like Amazon Web Services (AWS) are building their own chips based on ARM architecture, and the number one system on the Top500 list of the world’s fastest supercomputers – Japan’s Fugaku, built by Fujitsu and installed at the RIKEN supercomputer center – is powered by the vendor’s ARM A64FX chips.
Over the last several years, ARM has been in demand. Japanese conglomerate SoftBank in 2016 bought the British-based company for $31 billion. Four years later, GPU maker Nvidia announced its plan to buy ARM for $40 billion, a move that would give it a processor business to better compete with Intel, AMD and other chip makers, and one that company officials said would help it grow its high-priority artificial intelligence (AI) capabilities.
Merger ‘bad for the industry’
However, the proposed deal hit headwinds soon after it was announced in September 2020. Qualcomm – which alongside companies like Samsung became a major supplier of chips for mobile devices thanks to its embrace of the ARM architecture – is worried about whether Nvidia would make changes to the design to give their own GPUs a greater advantage at the expense of other vendors. More recently, Microsoft and Google, top-tier cloud providers that reportedly are developing their own ARM-based chips, have joined Qualcomm in asking regulators in the United States, European Union, UK and China to review the proposed deal, according to reports.
Nvidia CEO Jensen Huang told the Financial Times that his company will maintain ARM’s open license model, adding that “we have no intention to ‘throttle’ or ‘deny’ ARM’s supply to any customer.” The news site said both the EU and UK are opening investigations into the deal.
Nathan Brookwood, principal analyst with Insight64, is among those who believe the acquisition will be bad for the industry.
“I’ve been opposed to the ARM-Nvidia deal since it was announced,” Brookwood told InternetNews.com. “ARM needs to remain a neutral party if it wants to continue as a trusted supplier to companies that have adopted it as a key technology supplier. That means it should not be acquired by a company that competes with its customers. End of story. I’m hoping the Europeans or Chinese [regulators] will see it my way.”
A technology that has become successful as an industry standard can’t remain that way if it becomes owned by a single company, he said. To some extent the x86 architecture used in Intel and AMD chips was viewed for a long time as something of an industry standard, but that had more to do with Intel’s dominance and the lack of viable alternatives.
ARM is now a viable alternative, and if Nvidia is successful in buying the company, it could give a boost to RISC-V, an up-and-coming open standard instruction set architecture, the analyst said.
ARM’s licensing model makes it distinct from chipmakers like Intel and AMD, which design their own x86 processors. ARM instead designs the systems-on-a-chip (SoCs) architecture and licenses the designs to a range of vendors, which build their own chips atop the architecture. A growing number of chip makers are relying on ARM designs for server chips, including Marvell, which bought Cavium and its ThunderX chips in 2018 for $6 billion, and Ampere, a startup led by ex-Intel executives, which in 2018 acquired the Applied Micro assets that include its X-Gene processors.
As those deals attest, Nvidia’s proposed ARM deal is part of a larger acquisition trend in the chip space, as manufacturers look to grow their capabilities in a rapidly changing and increasingly diversified computing market. Demand for accelerators like GPUs, field-programmable gate arrays (FPGAs) and AI chips are forcing players like Intel and AMD to ramp up their offerings by buying other companies.
Intel in 2015 spent $16.7 billion to buy Altera, giving it a programmable chip business, and in 2019 bought Habana Labs, a data center AI processor company, for $2 billion. Those chips, combined with its own x86 CPUs and its new Xe GPUs and an accompanying API, comprise the company’s larger XPU project.
For its part, a reinvigorated AMD, with its x86 Epyc server processors based on its Zen architecture as well as its Radeon GPUs, announced in October 2020 that it plans to buy FPGA maker Xilinx for $35 billion, a move company officials said will give AMD greater capabilities in the data center and the high-performance computing (HPC) space. Meanwhile, along with its purchase of Cavium, Marvell also is buying Inphi for $10 billion, building out its network semiconductor business.
Investors also are looking at the chip market, including in new areas such as AI and the Internet of Things (IoT). Oracle in 2019 said it was investing $40 million in Ampere. Spark Microsystems, a fabless semiconductor startup launched by ex-Qualcomm CEO Paul Jacobs and Sanjay Jha, former CEO of chip contract manufacturer GlobalFoundries, that has created a low-power wireless IoT platform, announced this week a $137.7 million funding round. Chinese hyperscaler Baidu also reportedly is looking to raise money to create a standalone AI chip manufacturer.
Market research firm The Linley Group this week said that there are at least five dozen AI chip makers covering everything from data centers and IoT to autonomous cars and the edge in a market that has now exceeded $7 billion.
A rapidly changing market
For years Intel has been the dominant player in the data center chip space, with IBM and its Power architecture also in the picture. However, that has begun to change. Intel still sits at the top of the market, but AMD under CEO Lisa Su is making inroads after its initial push and then decline with its Opteron processors. As noted, ARM also is gaining traction, Nvidia has made its GPU accelerators a crucial part of data center computing, and other accelerators like FPGAs and AI chips are coming into play.
This trend can be seen in the last Top500 list from November 2020. Intel continued to be the leader, with more than 90 percent of the systems running either its Xeon or many-core Xeon Phi processors. That said, 149 of them use accelerators or co-processors, and 140 of those run Nvidia GPUs. Only 21 run AMD chips, though that is double the number from six months earlier, and while only five use ARM-based processors, one of those is at the top of the list.
Can Gelsinger right Intel?
Intel in recent years has struggled at times. It has fallen behind AMD in rolling out a 7-nanometer architecture and has stumbled in its manufacturing timetable. Last summer the company announced a delay in its latest generation chips and Apple later in the year said it was replacing Intel chips in Mac computers with its own proprietary processors.
However, as its XPU initiative shows, Intel isn’t standing idly by as competitors build onto their chip portfolios with GPUs, FPGAs and other accelerators – or in Nvidia’s case, trying to add ARM SoCs to its broad lineup of GPUs.
The company announced in January that Bob Swan – the former CFO who took over as interim CEO in 2018 when Brian Krzanich stepped down and then was named permanent CEO in January 2019 – was stepping down, making way for Pat Gelsinger, who spent 30 years with Intel before leaving in 2009 to become president and COO of storage giant EMC (before it was bought by Dell). In September 2012, Gelsinger took over as CEO of VMware.
He started as Intel CEO this week, offering a brief video statement on the Intel news page.
“As the incoming CEO, I am just really thrilled that we have the opportunity to take this great icon of a company, this company that has been crucial to every aspect of technology, and have it be that leader again into the future,” Gelsinger said. “I believe that Intel has a treasure trove of technologists, of technology, and ultimately its core DNA is being that technology leader for the future.”
However, it will still have to compete with ARM, which has been embraced by system OEMs, cloud providers and hyperscalers like Facebook, which deploy ARM-based chips in their massive data centers to address particular tasks. For its part, AWS continues to roll out instances based on its ARM-based Graviton processor family. It launched its Graviton2 chips in late 2019.