Lessons Learned from Platform Wars

Earlier this month, HP and Microsoft kicked off 2010 (perhaps this will be the “Year of the Cloud”) with a pact to invest $250M towards cloud computing services. In other words, they’re teaming up so as to remain relevant in the cloud age… and, of course, to compete with their perpetual nemesis, IBM.

They should know. They’ve made bets on platforms, profited handsomely from their choices and had their positions eroded by not seeing the next wave sooner. IBM on the mainframe, HP on Unix, Microsoft on Windows.

It’s a story that’s been played out again and again over the last 40 years. Four platform waves have risen (the fourth based on Linux/Open Source on Intel), with the cloud the fifth wave in the making. It always starts with developers looking for a better and cheaper way to develop programs. And ultimately, where developers go en masse, deployment follows.

If we’ve learned anything from the progression of mainframes to Unix to Windows to Linux, economies of scale and talent ultimately determine the successor platform.

The movement away from the mainframe to Unix started with developers who found development on workstations to be better, more flexible and cheaper than timesharing slices on the mainframe. But Windows came along, and offered an even cheaper development experience on commodity Intel computers. Windows’ popularity, of course, has been famously usurped by Linux (although some may still disagree). And while earlier platforms have not disappeared, new investments – be it in architecture or university education – are now being made in newer platforms.

Which brings me to cloud computing. While Linux on Intel offers a significantly less expensive model (in comparison to its predecessors) for deploying clusters of mass produced server architecture, the capital costs involved are still non-trivial, especially if you have to scale. In contrast, cloud offers attractive economies-of-scale from both a hardware and software deployment perspective. Up front capital costs are near zero. In deployment, new users essentially pay for just memory, CPU power and storage at variable capacity. Even if you already have capital equipment and data centers, cloud allows you to design for average load, not peak which is much, much more expensive. With cloud, you can contain your data centre capital costs for predictable workloads (and revenue!) and use a very low variable cost structure to grow in the cloud… until you can justify capital expenditure based on a “new normal”. For those into mathematical proofs, read Joe Weinman’s recent blog post here.

In my opinion, it is no longer up for debate as to whether cloud computing is, as Andrew Hickey would have it, hype or real, according to Bob Evans. It’s when it will peak as a platform, and what’s next after cloud.