Rackable Systems has just completed its acquisition of Silicon Graphics, and intends to change its own name to SGI going forward. Rackable becomes the name of a product line sold by SGI. There’s life in the SGI brand yet, it seems. Rackable bought SGI’s assets in a bankruptcy deal for a paltry $25m, and is also assuming certain liabilities.
SGI’s future has been in some doubt for several years now. I wrote a fairly critical piece back in the Summer of 2006, shortly after SGI had declared bankruptcy protection (not for the last time, it turned out) asking whether SGI would, as I put it, “Become the Next Data General?”
The piece was based on an extensive interview with the incoming CEO, Dennis McKenna, and I think it serves fairly well as a quick overview of why SGI found itself treading water, when once it had been a real powerhouse in the graphics-intensive server and workstation markets (click continue reading to see that piece if you can’t already).
CBR magazine, July 2006
Will SGI Become the Next Data General?
Remember Data General (DG), the server and storage vendor that, despite some great technology, ultimately failed to capitalize on it and was sold off to EMC back in 1999? Well SGI’s new CEO Dennis McKenna was adamant in an interview with me that, despite the company recently filing for Chapter 11 bankruptcy protection in the US, his company is not now simply looking for an exit strategy as DG once did.
SGI’s then CEO, Dennis McKenna.
As I noted in a recent blog, SGI’s McKenna was in London last month meeting some key customers and a few hacks. It perhaps says a lot about the current enthusiasm for SGI — or rather the lack of it — that despite the fact SGI had just weeks earlier announced its bankruptcy, not to mention the offer of free champagne and canapés in a trendy London venue, only three journalists turned up to hear what he had to say for himself.
The question in my mind was whether the game is up for SGI. Founded in 1982 as a maker of graphics display terminals, SGI has largely failed to capitalize on its technology heritage, and lost focus in the nineties when it made a series of — ultimately botched — acquisitions.
BOTCHED M&A STRATEGY
In 1995, it bought Alias Research and Wavefront Technologies and merged the companies into AliasWavefront. In February 1996 it bought supercomputer manufacturer Cray Research for $740m, and in September 2000 it acquired the Zx10 series of Windows workstations and servers from Intergraph Computer Systems. Since then Alias and Cray have been sold off again, and the Zx10s discontinued.
More problems ensued in the late nineties as the company transitioned from its own MIPS processors to the Intel Itanium, which was severely delayed. Despite having ported its clustering technology from its Origin Unix servers to Linux-based servers based on Itanium for its Altix servers since then, it has still failed to regain momentum. It’s also faced stiffer competition from PC-based 3D engines and commodity open source clustering technologies.
It all adds up to a perceived technology obsolescence that rightly or wrongly saw the demise of DG, Digital Equipment Corp (or DEC, which was sold to Compaq) and Sequent (which was sold to IBM). All of those companies, like SGI, had at least some good technology. But they also relied on that heritage too much, and did not move with the times fast enough to stay viable – let alone attractive to new customers.
Things came to a head for SGI in early May when it was forced to file for bankruptcy protection in the US. While the company had not posted an annual profit since 1997, its third quarter results saw a net loss of $43m, on revenue of $108m, down 32% from the year-ago quarter. It was left with $54.3m in cash, $40.6m in short-term restricted investments, and crippling debts.
So is it all over for SGI? Not according to McKenna. Asked whether he had any idea that things had become quite so bad financially at SGI before he took the reins in January this year, McKenna said: “Hell no,” and added, “They don’t train you for this kind of thing.”
But he told me that the filing of Chapter 11 bankruptcy protection in early May should be seen as a “positive event”.
“Before the Chapter 11 there was a high level of uncertainty around SGI,” he said. “Customers were saying that we could have the best strategy but how were we going to deal with all of this debt?”
While McKenna conceded that employees and customers might find the filing of Chapter 11 bankruptcy protection “unsettling”, he was adamant that, “As we have explained what we are doing and what this all means, we have gone from uncertainty to certainty.”
Since then the company has been focused on cutting costs, renegotiating its debt agreements and raising cash from divestitures where it can. It filed a Joint Plan of Reorganization, and announced that the court had approved a real estate settlement that saw Google paying SGI $319m to buy SGI’s headquarters property and another building in Mountain View, California, which Google had already been leasing.
The real-estate settlement will reduce its facilities-related cash obligations by $15m to $20m annually. Combined with a new financing facility, this provides it with approximately $19m of additional liquidity, the company said.
McKenna said that he has helped to turn companies around in the past during his 25-year career in electronic components, semiconductors, and semiconductor capital equipment industries, and that apart from the Chapter 11 process, this turnaround is not so very different.
He said the company is looking to raise additional money either from enforcing some of its numerous patents, or from divestitures. For instance he confirmed that the OpenGL application programming interface for the development of 2D and 3D graphics applications is “considered non-core” to SGI and that he is “open to selling it”.
He said though that the intention when spinning off technologies will be to retain a stake in the company in order to recoup some of the profits that technology could deliver in the future: “We sold off technologies in the past too easily for a one-time fee,” he said.
As for the company’s troubled past, McKenna has this to say: “My analysis is that we made three acquisitions and ended up selling them. We had six transactions in a relatively short time frame and that was disruptive. More to the point, we bought high and sold low.”
But McKenna is also looking to the future – a future in which SGI not only makes more of its existing technology but evolves with the times, something it hasn’t been particularly good at in the past.
In the past, SGI tried to make a go of it selling non uniform memory access (NUMA) based supercomputer clusters based on Intel’s Itanium architecture and the Linux operating system.
But even before McKenna joined the company in January this year SGI had decided that in order to address a larger market it would look downstream at x86 chips, specifically x64 processors which are a 64-bit version of that architecture.
Sure enough at the end of last month the company duly launched its first Xeon-based machines, which use Intel’s new ‘Woodcrest’ implementation of the low-power Core architecture. There are two models in the new Altix XE line: the Altix XE210 is a 1U server that has two processor sockets while the Altix XE240 is a 2U rack-mounted server that uses a more scalable Blackford motherboard.
But isn’t looking downstream at the lower end of the high performance computing market an anathema for SGI, with its heritage in higher-end supercomputers and clustering technology?
According to McKenna, it’s all about expanding its addressable market. “We’ve been island hopping,” he told me, “but now we are going to chase entire continents.”
McKenna argues that it can compete with rival open source or commodity clustering technologies by offering pre-bundled, clustered systems: “People don’t want to have to buy their own servers and configure them to try and get the performance they need,” he said. “Our x86 clusters will now address their capacity needs at a lower price point than our higher end Altix [Linux on Itanium] supercomputers.”
“There are different price bands in the [high performance computing] market,” McKenna continued. “We are bringing bundled cluster solutions that are already tuned and benchmarked, and then we put our middleware piece of magic on top of that.”
Since so much of the action in the supercomputer space has been going on in the x86 market — with no offerings from SGI to address that space until last month — it begs the question of whether one day the company would take the decision to put all of its efforts into the x86 space and migrate away from Itanium, not least because its financial performance of late has been so dismal.
“We are going down both paths,” said McKenna, “developing technologies on the hardware side and the chip side, and we think between the high end and midrange Intel architectures these levels can continue to be differentiated. We have a roadmap for both architectures out to 2010. But we will of course have to see what takes place in a couple of years.”
A couple of weeks back SGI announced that its plan of reorganization is going to plan and that it hopes to exit Chapter 11 by the end of September. McKenna says that by that time it won’t have any debt, and that its x86 sales should be ramping up. With a number of divestitures likely the company should be better focused, and McKenna looks like he will be able to shed the less profitable parts of SGI without getting hung up by an affection for what are indubitably some innovative technologies.
But is the plan simply to get SGI healthy enough to be sold off altogether? Not if you believe McKenna: “Of course if you are successful you will attract attention from potential bidders,” he said. “But it is our firm belief that we can make SGI a successful standalone company once more. I wouldn’t be here if I didn’t think that too.”
SGI has more challenges ahead, and I still find it hard to believe that after all of the chances it has had to run a profitable server and visualization business in the past it can miraculously do so now, selling lower-end boxes on even slimmer margins. But I’m hoping that the Chapter 11 has provided the necessary wake-up call for the company to get really lean really fast, because only from a more stable financial footing does it have any hope of fighting its way back onto new technology buyers’ wish-lists. It is a real shame when companies with such a great technology heritage ultimately fail to capitalize on it, so I’m still hopeful that SGI will not become the next Data General.
This article first appeared in CBR, July 2006.