For ARM, It’s Server Side Up


Ian Drew, executive vice president of marketing at ARM Holdings (s armh), a Cambridge, U.K.-based company that makes semiconductors powering a majority of the smartphones, tablets, 70 percent of world’s hard drives and half the world’s printers, is on a whirlwind tour of Silicon Valley. And what everyone (including me) wants to talk to him about is servers, or rather low-power server chips that can power the data centers of tomorrow.

With this foray into the fast-growing data center business, the company, which can trace back its roots to Apple Computer, is slowly becoming a bane to Intel’s existence. And the funny thing is, it does so not by making chips, but instead it develops technology and licenses it to all comers.

Some special companies — Qualcomm (s qcom), Marvell (s mrvl) and Infineon (s ifx) — get access to all of ARM’s entire technology stack, so they can build their own chips which in turn compete with Intel. The latest company to sign-up: Microsoft (s msft), the long time partner of Intel (s intc) and the other half of the ‘Wintel’ monopoly. What Microsoft will do with ARM’s technology remains to be seen, but it is a notable move nonetheless. It is a sign that the company is killing it.

Check out their American Depository Receipts: at the close of trading they were up nearly 165 percent over past twelve months. In the second quarter of 2010, its revenues jumped 42 percent to $150.3 million versus $105.5 million in Q2 2009. Earnings per share leaped 147% percent to $2.34 a share versus 95 cents a share it earned in a year ago quarter. From the looks of it, things seem to be going well for these guys.

The main reason ARM has done so well is because it has focused on developing low-power technologies that it licenses in turn to others, thus eschewing the headaches that come with manufacturing and selling your own chips. More importantly, ARM as a company sits at a unique position in the technology ecosystem. Because it licenses its chip technology to others, it has a good idea of what kind of products are coming to market, how well certain type of products are doing and more importantly where the industry is headed. In short, it talks to chip makers, device makers and these days even software companies that are developing software for these consumer devices.

And what Drew and his cohorts are seeing is a radical revolution in the data centers. “While the x86 world focused on pure megahertz, we have focused on the megahertz per milliwatt,” Drew said during our conversation earlier today. “We focus on quarter-to-half milliwatts as a key metric.” Most of the new devices such as the iPhones don’t have heat sinks in them, he joked.

“If you look at our heritage (of low power chips) it makes perfect sense for us to be looking at the servers and the data centers,” said Drew. With “cooling” making up nearly half the capital expenditure and almost two-thirds of the operation expenses, Drew said power is going to be a bigger part of the conversation.

“Everyone is using the Web and the Web is more demanding today which means all of the stuff is going to run through data centers,” he noted. “Two things are very clear: there is going to be a lot of data and need for less power.” By getting the world to buy more edge devices (iPhones, iPads etc.), ARM is at the same boosting demand for back-end computing infrastructure. Now by diversifying into the data center server business, it can make more money selling its low-power chip technology to server makers. In other words, ARM wins on both sides of the trade.

?Stacey, in a post earlier this year noted:

The news shouldn’t come as a surprise to our readers, since I profiled Smooth-Stone, one company trying to build low-power servers earlier this month, and in that same post pointed to ARM’s server ambitions. And it’s not just startups that are interested in using the low-power ARM architecture inside data centers, either. Google recently acquired a secretive startup called Agnilux that was rumored to be making a server with the ARM architecture.

We also reported on a Microsoft job listing that sought a software development engineer with experience running ARM in the data center for the company’s eXtreme Computing group. For the last couple of decades, Intel’s x86 chips have gained dominance in the data center, but as power considerations begin to outweigh the benefits of a cheap, general purpose processor, other chip makers have started to smell blood. Nvidia is pushing its graphics processors for some types of applications, while Texas Instruments is researching the use of DSPs inside servers.

But don’t expect this to happen overnight, Drew cautioned. “We are going to see some pilots over next year, but this is a long term initiative.” He believes that this long, continuous transition to lower-power server chips is going to take between three to five years. When I asked Drew what are those pilots, he declined to comment. From our reporting, we can easily tell you Microsoft, Smooth Stone and Marvell are experimenting with ARM-based server processors.

And while ARM tries to build a server business, the company, Drew said has plenty on its plate. For instance, the upcoming/next generation Cortex-A class processor codenamed “Eagle”, which is likely to help redefine the smartphone landscape again.

Related content from GigaOM Pro: (sub req’d.): ?Are Green Enterprise IT Pastures Within ARM’s Reach?


Tarun Dua

Low powered chips are good but they need to implement the x86_64 compatibility. Power consumption is only one part of game. To take advantage of power efficiency they also need to be available in smaller form factor e.g. 4 or 6 boxes in 1U space with shared electronic SMPS power strip consuming less than 150 VA overall including SSDs+RAM+Motherboards.

We ran some numbers for ourselves, Intel Atom’s are fairly useless in a server farm for us in India and virtualization using dual Quadcore boxes is far more capital and power efficient. In India running anything but the latest greatest Xeon quadcores is simply too power in-efficient.


x86_64? Only if you’re hell-bent on using Windows. On a server, you’re far better off running Linux.

ARM is already several times more power efficient than Intel’s offering and they power most mobile devices, so the small form factor is already there.

Intel is severely lagging them and has for some time.

Tarun Dua

A vast majority of our customers run CentOS, there doesn’t seem to be an ARM port for that as yet, but Debian/Ubuntu which is a fairly large proportion of our customers seems to have good support for ARM architecture. I guess we’ll have it on our roadmap to try out and validate ARM architecture.



The only thing I can imagine with AgniLux is that the high-speed communication interface could be implemented with optics. Otherwise, nothing much has transpired, but this will definitely impact Google’s bottom-line positively if they manage to spend less for their servers in the massive data centers.

You could also mention Apple’s acquisitions of PA Semi and Intrinsity as two ARM licensees. Intrinsity worked to break the 1Ghz barrier.


The only thing I can imagine with AgniLux is that the high-speed communication interface could be implemented with optics. Otherwise, nothing much has transpired, but this will definitely impact Google’s bottom-line positively if they manage to spend less for their servers in the massive data centers.

You could also mention Apple’s acquisitions of PA Semi and Intrinsity as two ARM licensees.


Sorry for the double posting, Om.

No, unfortunately, I have been trying to find out more about AgniLux, but count the company as one of those very stealthy ones, and as soon as Google bought them out, the website became a skeleton, so it’s very difficult to know what is happening.

The only thing I know for sure are these:
1. AgniLux is an ARM licensee
2. They have built upon the ARM IP to make chips for servers. At least they admitted that much. They also said they wouldn’t say more because it was quite radical
3. Google spends a lot of time thinking about how to reduce their expenditure, including that of their servers. Google is always optimizing its business (they’re data-driven internally too for business decisions)
4. One benefit of being an ARM licensee is that you can build on the IP but are free to design the high-speed communication interfaces.
5. For Cortex A8, it proved challenging for some to break the 1GHz barrier (that’s what Intrinsity managed, and Qualcomm with the SnapDragon platform), so maybe one aspect could be a heavily improved GHz rating, but how radical can that be?
6. Agni = Fire in Sanskrit & Lux = Light in Latin. So my premise is that they probably have a way of using light for the high-speed communication interfaces.

I wish we could have more real information instead of being forced to speculate on the name!

Om, I emailed you concerning your search for strategists a few weeks ago but didn’t get any reply. Should I send it again?

Comments are closed.