On Monday I laid out the case for platform monopolies: that they provide firms with incentives to create new products by allowing them to recoup their fixed costs. Yesterday I had two posts arguing that closed platforms can harm consumers by preventing gains to interoperability.
The question is, Is a platform monopoly an effective way to promote the creation of new devices? And is this benefit sufficient to outweigh the opportunity costs of reduced interoperability?
In considering these questions, it’s vital to distinguish between creating a device and creating a platform. Obviously, we want to create incentives for firms to produce more and better devices. But we don’t necessarily want firms to create new platforms. In fact, we only want firms to create new platforms to the extent necessary to enhance the functionality of new devices. If an existing platform will do the job as well, we should prefer the firm to use it, both because that saves the costs of developing the new platform, and because it permits gains to interoperability with compatible devices.
To illustrate this point, I want to offer a brief history of one of the world’s most successful platforms, the x86 computer chip architecture. I’ve discussed Intel’s x86 chip architecture (which powers almost all modern PCs) before as an example of beneficial reverse engineering. What I didn’t talk about explicitly was the role of network effects on the fortunes of the x86 architecture.
To put it bluntly, by any objective standard, the x86 architecture sucks. It was designed in the 1970s, and many of the design limitations built into the first chips are still hobbling today’s fastest CPUs. Predictions of the architecture’s imminent design have been common for two decades. In the early 1990s, it was widely believed that so-called RISC chips (such as the PowerPCs used in Macs between 1994 and 2006) would leave Intel’s chips in the dust. In the mid-1990s, even Intel thought the x86 architecture’s days were numbered, as they laid plans for a new processor architecture that became the comically ill-fated Itanic.
But x86 wouldn’t die. It turned out that by the mid-1990s, the network effects of the platform had given it unstoppable momentum. The size of the x86 market was so large that Intel and AMD could afford to simply throw more money at finding ever more elaborate ways of overcoming the platform’s limitations. Although RISC chips did often out-perform x86 chips in the mid-1990s, they just couldn’t compete on price, because they had far fewer customers over which to spread their R&D costs.
Over the last decade, x86 chips have gradually cannibalized the rest of the desktop and server chip business. Apple succumbed last year when it abandoned the PowerPC for Intel chips. Sun has begun selling servers based on x86 chips alongside its proprietary SPARC architecture for a couple of years now. Traditional PC makers like Dell have seen increasing success selling x86-based servers.
What this story demonstrates, I think, is that the technical details of the x86 architecture were almost completely irrelevant to its subsequent dominance of the desktop computer industry. Intel’s product didn’t conquer the world because they made spectacularly good design choices when they built the first x86 chips in the 1970s. Rather, they succeeded despite the handicaps of their architecture, thanks to some savvy business decisions, fantastic products, and enormous economies of scale.
Many firms, including IBM, HP, Toshiba, Motorola, DEC, Sun, and Intel themselves developed processor architectures that were technically superior in various respects to the x86 platform. The dominance of x86 certainly wasn’t due to a lack of worthy competitors. But the architecture itself appears to have been minor factor of the platform’s success.
I think this lesson applies to other technological architectures as well. The interfaces that define a platform tend not to be a major differentiator for device value. Generally speaking, the interfaces are far simpler than the products that implement them, and many of the details of the interface flow straightforwardly from the functionality required by the device. Once an interface is “good enough,” there’s little to be gained from investing additional resources in improving the interface.
All of which is to say, there’s very little to be gained from creating incentives to create new platforms, as such. If new platforms are needed to support new devices, firms will create them along with their devices. And once an adequate platform is created, it’s usually better to use the existing one that design a new one.
So our focus should be on incentives for the creation of new devices. If those are adequate, platforms will take care of themselves. Next I’ll consider if platform monopolies are a good way of encouraging the creation of devices.