Cray CEO responds: "Yes, we can make money at the high end of HPC."

Print Friendly, PDF & Email

Cray logoAt the end of October Cray announced financial results for their fiscal third quarter, ended Sep 30. Revenue was up, but the company reported a loss for the quarter. As I noted in that piece this is not unusual: Cray is nominally a $200-$300M per year business with a focus in the high end of HPC. In that rarefied market, individual deals can be 25% or more of a Cray’s yearly income, and a slipped acquisition or acceptance schedule can rapidly turn a single quarter from a profit to a loss. For Cray it makes more sense to track the company’s performance on an annualized basis. On the face of it the news here is not good either though: the income statement shows a loss in each of the past three years.

My writeup on the financial results got a few comments, including one who challenged the fundamental business model of a company like Cray

Yes they are in the high end but why make excuses for the fact that they can’t make money – maybe greater diversification is needed or their business model is flawed. Almost every quarter there is some write down, order/product delay, higher cost followed by the CFO saying if we didn’t have these we would have been profitable. Last night I missed the lottery by one number therefore I should consider myself a millionaire.

That got me to thinking about whether Cray and company’s like it that focus exclusively on the high end are doomed to starve in a market too small to support the R&D they need to do to compete. I certainly hope this isn’t true, because the country needs high end HPC in the same way that it needs submarines, another critical technology to our nation’s global stature that has no commercial market. And HPC today actually has more going for it than submarines: today’s supercomputers are built from commodity parts and the technologies can often be scaled down to sizes that are right for workgroups and businesses, creating potential downstream markets for technology at the high end.

But I’m a writer and a consumer of HPC goods and services, not a producer of either, and rather than spilling 1,000 words navel-gazing about the HPC business I wanted to get the company’s own take on its business model. So I talked with Peter Ungaro, Cray’s CEO and President, to try and understand how he sees his company’s business and the market he’s chosen. I opened with a direct question — is it possible to make money in the high-end of HPC? Ungaro’s answer? “Yes.”


Peter UngaroWhat’s behind that “yes” is the interesting story. Ungaro sees his company’s jobs in much the same light that every business leader does, regardless of industry: develop a great product, create a following, and grow revenues. Creating the following has been the problem for pure-play HPC companies. The HPC market is very demanding, and acts swiftly to punish companies and technologies that fall behind (witness the rise, fall, and rise again of Intel on the Top500). Companies in HPC have to innovate continually, which creates a huge R&D burden.

Each year Cray spends about 30% of its gross revenues directly on R&D, and according to Ungaro over half of Cray’s employees are directly involved in research. Companies like IBM and HP can leverage other market segments to support that kind of cost burden, but pure play HPC companies cannot.

So the key question for Cray whether — even assuming a compelling product that customers will buy — the revenue large enough to support all the R&D it takes to stay competitive. Again, Ungaro says yes.

Ungaro puts Cray in a $3B segment of the roughly $10B total HPC market today. At $300M, the company has plenty of room to grow, and if you dive into the financial statements you’ll see that they are actually making money where they are today. Cray has been one of the few positioned to take advantage of a strong cash position going into the economic downturn to buy back all of its outstanding debt at a significant discount to book value, while maintaining its line of credit and about $65M in cash at the end of Q3’09. This means they have options: they can run the business out of cash on hand, or invest in the future and maintain an ability to service large orders by drawing on short term credit facilities.

In addition to its balance sheet strength, the company is putting cash into the checking account. Following a significant realignment of the company during the first several years of Ungaro’s turn in the captain’s chair, the company actually made “real” money in 2008, putting about $23M into the bank. I qualify that statement because Cray was forced to reduced stated earnings by $54.5M for goodwill impairment (that would be “fake” money) in a move that impacted the balance sheet, but not the checking account.

Still, the point about diversification my commenter made in his comment is a reasonable point. Most businesses diversify themselves to insulate earnings from one-time fiascos, slipped schedules, and the like. The supercomputing market — the very high end of HPC — is a hard business to be in technologically, and requires some very dedicated focus. This means either doing one thing, and doing it very well (Cray’s approach), or it means being big enough to focus on many things very intently (like IBM and Intel, for example).

In our discussion, though, Ungaro pointed to the company’s efforts to diversify into businesses that would help damp oscillations in the revenue stream while continuing to focus on the high end. This leads to a strategy of finding new downstream markets for work Cray is already doing. Enter the mini and custom engineering.

Diversification based on the high end

In the past 12 months Cray has introduced two new lines of hardware.

The first, Cray’s CX1 deskside HPC system, is a Cray in name only: Cray concentrates on the design and go-to-market aspects and leaves the rest to others.This makes the CX1 almost a branding exercise with very little downside for the company. A smart way to experiment with a new market.

The mini however, the system officially designated the XT5m, is all Cray. “The mini leverages our high-end R&D down to $0.5-2.0M systems buyers,” says Ungaro. “We are still innovating with this system, but in a different way. Mini buyers run much more ISV codes than our high end buyers, and so the work we do to facilitate that use on the XT5m will find its way up to the XT5 and improve the value for customers there in the same way that our architecture and system management innovations at the high end formed the basis of this machine.” How significant is XT5m business today? “There are a handful of systems around the world, but we look for that to grow significantly and become a double digit percentage of revenues in the future.”

Custom engineering, on the other hand, is a different story. The custom engineering business was announced by Cray back in April of this year, headed up by previous services head Chuck Morreale. The idea behind the business was simple: Cray has a lot of smart people who have solved incredibly challenging problems in bringing large scale supercomputers to market. Why couldn’t they sell those smarts to others as a service?

The idea has been well-received. “Custom engineering has grown to a $30+M business from $0 in just over one year,” according to Ungaro.

In the abstract this is essentially a bodyshop business, and bodyshops have notoriously low margins; there is always someone else willing to take a loss on a job “just this one time” to get the business. In order to avoid competing against that crowd, however, Cray has positioned its custom engineering team to take advantage of technology and competencies it has to develop in-house for itself anyway, and use those to drive new revenue. For example, Ungaro highlights the company’s ECOphlex cooling technology developed for its own high-end supers, and now being tweaked by custom engineering for a variety of different customer applications. Other growth areas for the team are data management and knowledge management, both areas that rely on the company’s high-end HPC business.

Ungaro hopes that the recipe for growing revenue as a high end HPC company without simultaneously ceasing to be a high end HPC company is to grow around the needs of the high-end market. “I don’t only think of us as a supercomputer company,” he says. “I think of us a supercomputing innovation company. If we can remove the barriers to people using Cray technology, we can continue to make gains in the market.”


Of the persistent undertone that the only reason the company continues to exist is thanks to the charity of three-letter agencies in the federal government, Ungaro says (almost wistfully) that just isn’t the case. “We are past the days when customers, any customers, were willing to buy a nameplate. High-end customers are very technical, and very logical. If our machines didn’t stand on their own technical merit, we just wouldn’t be successful. We might sell 3 systems, but not 30.”


  1. […] talked about Cray’s CE business before, most recently in an exclusive interview I did with Cray CEO Pete Ungaro about the practicalities of making money at the high end of HPC. […]