The Greatest Threat To Intel In Data Centers Comes From IBM, Not ARM

While Intel’s (NASDAQ:INTC) PC business revenue has been posting declines (-7% y/y last quarter), Intel’s Data Center Group revenue has been growing (up 12% y/y last quarter). Although much attention has been paid to the development of ARM-based server systems to compete with Intel, such systems are a long way from being adopted in larger numbers. The greater near-term threat to Intel comes from a company that already has an established position in the data center, IBM (NYSE:IBM).

Divesting the Past

Despite having sold off its Intel-based server business to Lenovo, IBM still ranked third in IDC’s Q2 report on the Worldwide Server market that came out in August:

Vendor 2Q15 Revenue 2Q15 Market Share 2Q14 Revenue 2Q14 Market Share 2Q15/2Q14 Revenue Growth
1. HP (NYSE:HPQ) $3,427.7 25.4% $3,181.7 25.1% 7.7%
2. Dell $2,364.3 17.5% $2,233.4 17.6% 5.9%
3. IBM $1,994.4 14.8% $2,972.0 23.4% -32.9%
4. Lenovo (OTCPK:LNVGY)* $949.2 7.0% $144.7 1.1% 556.1%
4. Cisco* (NASDAQ:CSCO) $866.7 6.4% $726.7 5.7% 19.3%
ODM Direct $942.6 7.0% $834.9 6.6% 12.9%
Others $2,930.5 21.7% $2,602.6 20.5% 11.9%
Total $13,476 100% $12,696 100% 6.1%

The 33% y/y revenue decline was a consequence of the divestiture of the Intel server business. According to IBM’s earnings reports, IBM’s Power Architecture Server business grew revenue in both quarters, by 5% y/y in Q2 and by 2% y/y in Q3.

The fact that IBM chose to dump its Intel server business but keep the Power Architecture (PA, the heir to the PowerPC) servers should pique the curiosity of investors. Didn’t Apple (NASDAQ:AAPL) kill off the PowerPC when it switched the Mac to Intel back in 2006? Not quite. IBM shifted its focus for the PowerPC to servers as a complement to its Z series mainframe business.

IBM’s decision says a lot. IBM looked out over the semiconductor business landscape and realized that processors based on ARM Holdings (NASDAQ:ARMH) processor IP had become dominant in mobile computing, and seemed to be moving beyond mobile devices into mainstream personal computing and into servers as well.

The distinctive feature of ARM architecture is that it’s an example of so called Reduced Instruction Set Computing (RISC), as compared with so called Complex Instruction Set Computing (CISC) processors, of which x86 is an example. IBM’s Power Architecture is also RISC.

I doubt there’s any significant performance advantage for RISC vs. CISC. I’ve argued that the key difference between x86 and ARM is in cost. The ARM architecture is simpler and lower cost to implement than x86 for equivalent performance and power consumption.

The main substantiation for this thesis is the fact that Intel has not been able to build a cost competitive and performance competitive smart phone processor compared to typical ARM processors, despite having the advantage of superior process capability.

The Face Off

It remains to be seen whether IBM’s Power Architecture is also more cost effective than Intel. It probably is, by virtue of being RISC, and having a similar instruction set architecture to ARM.

What PA has mostly lacked is a competitive process node compared to Intel. This comes across in a recent review by Anandtech of the IBM Power8 S822L rack mount server. The IBM system was deficient in the key area of power efficiency. Although the IBM system was reasonably competitive in performance, it was by far the more power hungry, by a factor of two or greater, especially at idle. Power efficiency counts for a lot in the data center.

The Power8 chip in the IBM system is fabricated on a 22 nm process, whereas the Xeon system used for comparison runs Haswell derived Xeons that uses Intel’s 22 nm FinFET process. It’s Intel’s process, combined with other power reduction techniques that give it the power efficiency advantage over Power8.

Anandtech also found that the Linux based software stack of the IBM system lacked maturity. The testers found that the ports of Linux to Power Architecture were in some cases incomplete. For IT managers who prize reliability and maintainability, this lack of maturity is a deadly sin.

Looking to 2016

Despite the fact that the IBM system fell short in a head-to-head match up against Intel, there are good reasons to believe 2016 will be better for IBM. The signs of the haphazard software ports to Power Architecture in the Linux OS and associated software will probably have been resolved by then. Most importantly, IBM will be able to move it’s next iteration of Power8 (Power8+) to 14 nm FinFET, probably provided by Global Foundries. This should do a lot to lower power consumption and make the IBM PA systems more competitive.

In 2013 IBM formed the OpenPower Foundation to foster third parties involvement with developing PA systems. There is every reason to believe that this will further reduce the cost of PA based systems. IBM has license the Power8 design to a Chinese startup, Suzhou PowerCore, which will start making the chips. Tyan has created a rack mount system using the Power8 chip. Other members of the Foundation include, Google (NASDAQ:GOOG) (NASDAQ:GOOGL), Nvidia (NASDAQ:NVDA), and Mellanox.

Moving the PA chip production to a more competitive process, combined with low cost Chinese production of chips and systems, combined with a low cost and more mature Linux based software stack, should make the PA system much more competitive with Intel.

And what of Intel’s much vaunted silicon photonics, which has been presumed to offer an advantage to Intel in the data center? Here I believe that technically naïve financial writers have been taken in by Intel’s clever marketing. There’s nothing new about the use of fiber optics for interconnects in the data center. Mellanox, one of the members of the OpenPower Foundation, makes 100Gb/s Ethernet cards for servers as a commercial product.

There is the promise, as yet fulfilled by no one, including Intel, of eventually being able to integrate all the lasers and electronics for fiber optics onto the silicon substrate of the microprocessor. Until that promise is realized, the fiber optic systems all need to be in separate modules from the microprocessors themselves, including what Intel offers.

Investor Takeaway

Intel’s Data Center Group, which has been a star performer up to now, could be in for a shock to the system next year. Unlike inexperienced companies offering ARM-based systems, IBM has presence in the data center. IBM has credibility. IBM has a sales and support infrastructure. And IBM has partners in the OpenPower Foundation that can further its development of cost competitive systems.

The first signs that Intel is under pressure will be a stagnation of revenue growth in the Data Center Group and reduction of operating income. Intel will fight tooth and claw to hold onto its server market share, but this will come at the cost of operating profit for the segment. I expect this will not sit well with investors.

Lack of meaningful progress in smartphones, stagnation of the traditional client and server businesses make Intel at least a Sell in my view. Given that the market still has high hopes for Intel, I believe that makes it a Short opportunity as well.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s