WavebreakmediaMicro - Fotolia
HPE-Cray deal brings supercomputers to enterprises for AI
HPE's $1.3 billion deal to acquire supercomputing pioneer Cray Inc. should better position the company in the HPC market among IT shops looking for added muscle to run AI workloads.
Hoping to resurrect its fortunes in the supercomputing market, HPE purchased Cray Inc. for $1.3 billion in a deal that could change the landscape of that market.
In an early morning press conference, HPE CEO Antonio Neri said both company's respective High Performance Computing (HPC) portfolios are complementary, with each having delivered innovative technologies in different but key areas. Neri emphasized that Cray's technologies are highly modular and will be beneficial to computing layers below the HPC level.
"What Cray really brings to us is what I call the foundational technologies," Neri said. "(Cray has delivered) the next generation of interconnect and software tool chains, which are really critical to manage these data-intensive workloads."
HPE expects the deal to increase its footprint in federal business and academia, as well as selling supercomputing products to its corporate IT shops.
The HPE-Cray deal is expected to close by the first quarter of HPE's fiscal year 2020.
HPE acquires Cray's exascale edge
One analyst said the HPE-Cray deal could put Hewlett Packard Enterprise in a better position to win a number of major government contracts.
"They are gaining more credibility at the high end of the market with this deal," said Steve Conway, senior research vice president at Hyperion Research. "The big money procurements that are in the $300 million to $600 million each are the exascale systems coming up and Cray has done very well with those. HPE hasn't been able to compete well yet, but with Cray on board now that will change."
Cray's value to HPE may have increased considerably with two major U.S. contracts recently awarded for development of exascale computers in 2021. Exascale represents the next big leap in computing power, a 1,000 fold increase from the petascale, which was first reached in 2008.
These contracts include the Frontier supercomputer at the U.S. Department of Energy's Oak Ridge National Laboratory, an exascale system costing $600 million. Cray's Shasta system, as well as its interconnect Slingshot, are also being used in a second U.S. exascale system, Aurora, at the Argonne National Laboratory. That system will cost $500 million, with Cray as a subcontractor to Intel on that system.
Company officials did not comment on what the HPE-Cray deal would mean for "The Machine," a server it touted over four years ago as a system with "a revolutionary compute architecture," which has yet to find its way to market.
HPE-Cray deal comes together on AI
Both companies take aim where the HPC and commercial business markets converge, a place where HPE has some experience but not Cray, Conway said. AI and machine learning are driving the growth in those converging markets, Conway said.
Steve Conway Senior research vice president, Hyperion Research
"Clearly HPE has more experience (in commercial markets) than Cray, and Cray is pursuing the really big next-generation government computers," Conway said. "These next-generation computers will wash over the entire market because of the growing need for things like ML and AI and deep learning," he said.
This acquisition represents a major change to the HPC market. Today there are only three major HPC vendors in the United States: Dell, IBM and now the combined company of HPE-Cray. America's HPC vendors once dominated the global HPC market, but are now under intense competition, particularly from China.
The advent of exascale computing is accelerating the race to build the world's fastest computers. Earlier this year, IBM launched its Summit and Sierra supercomputers at the Oak Ridge National Laboratory and Lawrence Livermore National Laboratory; they are now ranked as the world's two fastest computers, respectively. The HPE-Cray deal must go through regulatory review, but it's unlikely to run into regulatory problems because Cray's main business is at the very high end of the HPC market and doesn't overlap heavily with HPE's business, Conroy said.
"There are enough major competitors, such as IBM and Dell, that the HPE-Cray combined company won't corner the market or eliminate the presence of competitive pricing," Conroy said.
Whether there will be questions raised about competitive antitrust issues depends on how the government views the market for large computers, said Dan Olds, HPC analyst and partner at consulting firm OrionX.
"If they see supercomputing HPC as a distinct and separate market, then they'll probably want to take a close look at this deal and could possibly object," Olds said. "If they don't see HPC as a separate and distinct market, there's a pretty good chance that this deal will not receive much scrutiny."
Supercomputing market on the upswing
The HPC market is set to rise from $28 billion this year to $35 billion by 2021, HPE said. Within that market, demand for exascale systems will put about $4 billion in sales up for grabs within five years, the company added.
Based in Seattle, Cray's history dates to 1972, when founder Seymour Cray created Cray Research. Its systems are manufactured in Chippewa Falls, Wis., where Cray grew up.
The company has struggled financially of late, which undoubtedly made it ripe for acquisition. Cray logged $456 million in revenue in 2018, but lost $72 million. The bleeding slowed compared to 2017, however, when Cray lost $134 million.
In its first-quarter earnings report this month, Cray reported $72 million in revenue and a loss of $29 million. Those results compared to $80 million and $25 million, respectively, the quarter one year prior.
Things did start looking up recently, thanks to Cray's contract with the Oak Ridge National Laboratory for its Shatsta exascale system. However, Shasta systems won't actually ship until the end of this year, and Cray executives have characterized 2019 as a transition phase for the business.
Cray's technology will join HPE's family of HPC systems, such as Apollo and SGI. Longer-term plans include HPC as a service and work on AI and machine learning in conjunction with HPE's Greenlake flexible consumption pricing service.
Greenlake could be attractive to partners and customers that want to offer or tap into HPC systems but don't have the desire or capability to make the massive financial investment owning one would require.