One of the inherent problems with IBM’s Power Systems Computing effort was its focus on the processor. This makes sense for companies like AMD and Intel which, at their heart, are processor companies but for IBM—a solutions company—this effectively crippled the effort. It simply was not set up to compete against firms like AMD, Intel, or Qualcomm, being designed to compete with firms like Oracle and HPE instead. In addition, when going after a market segment dominated by one company—Intel—you just can’t out-resource it broadly yet that was what it looked like IBM initially was trying to do and the end result wasn’t pretty.
But that was then, and IBM has recently pivoted its Power Systems Computing group to Cognitive Computing which effectively addresses both mistakes. It focuses IBM’s effort on solutions, not parts, and it targets an area where Intel is comparatively weak, AI (artificial intelligence). (And just in time because Intel is ramping up its efforts in AI at an impressive rate).
Let’s talk about IBM’s brilliant strategic change.
Solutions Companies Can’t Sell Parts (at Scale)
Ironically, I know of one exception to this rule and that was with IBM. Back in the 1980s, IBM controlled 95 percent of the enterprise storage business and it owned most of the critical patents for hard drives. As a result, even those that built products that competed with IBM bought IBM’s parts. But IBM, at the time, was massively dominant in technology and not chasing another dominant firm.
IBM tried to recreate this advantage four times, first with Token Ring, then with its own x86 parts, then with OS/2, and finally with Power—all of which failed. The reason is that competitors generally won’t buy a critical part from a company they are competing with. This is for two reasons, fear of the competing company pulling the parts rug out from under them, and fear that the competing company will be able to argue they are better with the technology sourced from them.
This means that a firm like IBM is best served by focusing on technologies that allow them to differentiate but don’t need the volume of a cross vendor solution to be competitive—and typically processors wouldn’t be on that list thanks to a broadly dominant parts firm like Intel. You just can’t get to the economies of scale.
However, large companies like Intel that try to build products for everyone often leave gaps where a focused vendor can come in with a better point product. In the broad market that would be a company like Qualcomm which has successfully prevented Intel from penetrating personal technology devices and has—along with AMD—found that parts focused on cloud loads can at least gain beachheads.
IBM, which is the leader in enterprise AI with Watson, focused on where its strength was and that is AI. Not just Watson AI either, but developing a platform that would better host a cognitive analytics offering.
IBM’s powerful secret sauce comes from a relationship with long time Intel rival NVIDIA and its unique NVLink technology. When it comes to deep learning machines, GPUs—rather than CPUs—have been king, but a winning solution needs both capabilities optimally balanced. By working with NVIDIA both firms have created what appears to be the leading AI hardware platform and the effort is enough removed from anything Intel is doing that its path to catching up is relatively long and ugly. Don’t get me wrong, Intel is fully capable of funding what it will take to close this gap but it won’t be overnight and catching both IBM and NVIDIA from behind is far from easy.
Wrapping Up: The Big Lesson
The big lesson here is that it often looks far simpler to compete with a firm larger and ahead of you by emulating it, but—unless you are like Google and have more money than sense—that virtually never works. Far better is to use your unique advantages to address a part of the market the dominant vendor is underserving. That is what IBM is doing with its Cognitive Computing Effort and it already has been vastly more successful than Power alone had been showcasing the real power of this approach.