Advanced Computing in the Age of AI | Monday, May 27, 2024

Cray – and the Cray Brand – to Be Positioned at Tip of HPE’s HPC/AI Spear 

More so than with most acquisitions of this kind, HPE’s purchase of Cray for $1.3 billion, announced last week, seems to have elements of that overused, often abused term: transparency. Another surprise: HPE apparently isn’t paying the usual lip service when it says Cray will live as an entity under the HPE umbrella and that the venerable Cray brand will continue to exist.

We say all this for several reasons – for one, industry observers agree that giving Cray semi-autonomy makes sense; for another, senior executives at the two companies have already provided details on how Cray will operate within HPE’s management and business structure.

Here’s what we know: two companies of major importance to HPC and to the evolution of AI have joined forces. We know that HPE paid a somewhat surprising premium for Cray (3X revenues, 20X earnings and a 27 percent premium above the average Cray stock price the previous 30 days), indicating that HPE perceives Cray as an asset with a healthy future life. We also know that the two companies have different – and, the companies said, complementary – HPC and AI product portfolios, with Cray’s market focus on government/high end and HPE’s on the commercial mid-tier.

Regarding the cultural and organizational fit, Cray CEO Pete Ungaro corroborated HPE CEO Antonio Neri’s statements last week that Cray and the Cray brand will remain entities within HPE. Acknowledging it is, “early in the process and many aspects have yet to be determined,” Ungaro stated in a letter to the company’s 1,300 “Crayons” that “They (HPE) see value in the technologies we’ve built (and are building). In fact, they have decided to combine their HPC business into ours and are planning to keep the Cray brand alive in how we market our supercomputing products.”

Further, Ungaro said he will “lead this new organization, to work with their team and make sure we preserve the best of both companies as we bring our two companies together.”

The “combined organization” will exist within HPE’s Hybrid IT business unit, which Ungaro said is the largest segment of HPE and is led by Phil Davis, president of Hybrid IT and HPE’s chief sales officer, who reports to Neri.

Piecing together Neri’s and Ungaro’s statements, it appears that the Cray brand will carry forward HPE’s product and marketing efforts within the high-end HPC – i.e., supercomputing – business.

We know that words and sentiments of this kind are commonly mouthed, with varying degrees of sincerity, by acquirers, and we know why: to avoid disrupting customer and employee relations. But in this case, the fact that Ungaro has made the statements lends them credence. And as mentioned, it makes sense to use the Cray brand as HPE’s marketing spearhead at the zenith of supercomputing-class technology, given Cray’s unique position as a driver of this segment; as a company that historically helped create and define supercomputing as a technology and a market in its own right; and because well more than half of Cray’s business is with government agencies and with supercomputing centers and universities, much of whose funding comes from government and have been Cray customers for decades.

By contrast, HPE, though the market share leader in HPC servers, does most of its business in commercial HPC sectors selling midrange HPC products.

In addition, Cray is executing on contracts to build two exascale-class systems, for Argonne and Oak Ridge national labs, based on Cray’s “Shasta” supercomputing platform and its Slingshot interconnect technology, and it seems unlikely HPE will do anything to disrupt delivery of those systems and realization of their attendant revenues (Neri pointed out last week that exascale computing is a $4.3 billion market opportunity over the next five years).

Breaking down the overall HPC compute market, Neri sizes the supercomputing/exascale sector at between $2.5 to $5 billion, while the sub-supercomputing HPC sector at $8.5 billion.

Also, while it’s not clear what will come after exascale-class computing, which could serve as the high end supercomputing architecture-of-choice for the next decade or more, there is additional opportunity in building systems that attain the next level of supercomputing performance – whatever that may be.

Given the expected, continued health of the zenith-class supercomputing market, preserving the iconic Cray brand makes sense, Hyperion Research SVP Steve Conway told us.

“Our studies show that requirements for the enterprise and hyperscale and cloud markets are pushing up into the HPC competency space, especially the leading edge where Cray lives,” said Conway. “HPC is indispensable today at the forefront of AI and machine learning, and the evolving requirements are beyond any vendor’s capabilities today, so Cray is well positioned to help lead the charge.”

Karl Freund, consulting lead for HPC and deep learning at industry analyst group Moor Insights & Strategy, shared a similar view.

“I think it is likely that HPE provides support and additional sales coverage, while not interfering in the core Cray business,” said Freund. “The main-line true-supercomputer class system based on Shasta is already in a class by itself, and I don’t see that changing much other than possibly enjoying the benefits of scale in parts procurement. The US and European governments need this kind of performance for research, and will continue to buy from HPE/Cray, Dell, and Lenovo.”

Last week, Neri said a factor in favor of the acquisition is putting HPE’s global sales and marketing organization to work selling Cray gear to commercial markets, and Freund agrees this strategy has potential.

“HPE brings a very large commercial enterprise go-to-market that's truly complementary to Cray (because) they don't have that reach,” Neri said last week, emphasizing the services opportunity. “So you need to look at this first and foremost as an opportunity for HPE as a margin expansion because we leverage their technologies in through the rest of our portfolio and second is upside of revenue by not just participating in supercomputing but over time in the other subsets of the market, which are technologies that will be available over next several quarters, which today we didn't have the IP.”

On this point, and referring to the Cray “Frontier” exascale system due to be delivered to Oak Ridge in 2021, Freund said, “The real benefit to HPE will be selling ‘mini-Frontier’ systems to smaller government and enterprise institutions. Unlike the other acquisitions like Nvidia/Mellanox and Xilinx Solarflare, this is a merger of two systems companies, not foundational silicon platforms, so it puts HPE in a unique position in the market; they can and will sell a lot of silicon from all the leading vendors.” He also foresees the convergence of Cray’s CS500 cluster server line and HPE Apollo servers, “making both a stronger player in the enterprise for AI and HPC,” Freund said.

Conway said the nature of the two companies’ product lines and market strategies will result in “Cray to be given a fair amount of autonomy, all the running room needed to realize the potential value of the acquisition by capturing additional exascale and other high-end business. But I would also expect HPE to manage combined company teams to integrate the Cray and HPE product portfolios and roadmaps, and to develop next-generation technologies together. So, there will be a mix of autonomy and integration.”

Source: HPE

From a broader perspective, it’s generally agreed that the recent spate of high performance computing consolidations is driven by emerging markets centered on what could be called the “holy quintet” of integrated HPC-big data-AI-5G-IoT solutions. This is spurring the infrastructure vendors to build out a wider range of high performance storage, networking and interconnect capabilities to address the AI-infused compute-everywhere world to come.

It’s this combination of technologies that leads industry watcher Addison Snell, CEO, Intersect360 Research, to speculate on why HPE agreed to pay nearly three times Cray’s annual revenues of $456 million – Snell said the purchase price is about “three times over what I’d expect to see in a stable kind of market. So that leaves two conclusions, one is that everyone at HPE is crazy – I don’t think that’s right. The other is they see some other value here where they can take the Cray tech into some other area.”

The asset – possibly a hidden asset – that HPE may view as a revenue multiplier, Snell said, is the ClusterStor high performance storage line developed by Seagate and taken over nearly two years ago by Cray when it was given rights to brand name and hired Seagate’s ClusterStor engineers.

Noting that HPE cited ClusterStor as a key technology during its briefing last week, Snell said high performance storage “is an area where I have been critical of HPE in the past… It could be what we’re seeing is the potential for HPE to take that ClusterStor line, re-purpose it and sell it a lot more broadly than Cray has been able to do. If HPE could add 50 or 100 percent to its storage footprint in these high performance markets that would quickly start making up the gap in the expected purchase vs the current value of those assets.”

This indirectly relates to the widely noted “lumpiness” of Cray’s high-price/low-volume/long-sales-cycle business model, resulting in revenue ups-and-downs the company must weather between systems deliveries. Neri said HPE’s deeper financial resources (2018 revenue: $31 billion), and it’s focus on the less expensive/higher volume commercial HPC/AI markets smooth out the financial environment in which Cray will operate.

Ungaro acknowledged this in his letter to employees, explaining that the company wants to pursue opportunities to expand its customer base and take advantage that “the massive growth in data and convergence of modeling, simulation, analytics and AI provide us… and the chance to win a few more monster exascale systems around the world.”

However, he said, “With our current operating model, it’s challenging to make the investments we need to fully take advantage of these opportunities. At the same time, our size often inhibits us from getting the same prices on components that our competitors do. Additionally, we face the longer-term threat of the massive cloud vendors. Making the necessary investments while also delivering profitability over time is a challenge and a major risk for our business.”

The acquisition is expected to close within six to nine months; the two companies will remain separate and will work in the market independently until that time.