The coming of the megacomputer
Here's an incredible, and telling, data point. In a talk yesterday, reports the Financial Times' Richard Waters, the head of Microsoft Research, Rick Rashid, said that about 20 percent of all the server computers being sold in the world …
Context & Ripple Effects
Writing on Rough Type in March 2009, Nicholas Carr seizes on an FT-reported talk by Microsoft Research head Rick Rashid: roughly 20 percent of all server computers being sold worldwide are going toward building what he calls megacomputers — warehouse-scale clusters assembled from commodity machines. The stat lands at a pivot point for Microsoft itself: its Windows monopoly is visibly shaking, and the company has just begun pushing its server stack onto third-party clouds like Amazon EC2 in an expansion of Windows Server beyond its own datacenters.
The significance is structural, not just statistical. If a fifth of server shipments flow to a handful of megacomputer projects, the center of gravity of the industry is moving from millions of enterprise buyers to a few operators running search, mail, and web services at planetary scale — with Microsoft Research, fresh off showcasing more than 100 future-computing innovations, positioned squarely inside that buildout. It is an early snapshot of the concentration question that resurfaces a decade later, when researchers warn that AI advances increasingly require datacenter-scale computation few companies can afford.
First-order effects
- Purchasing power over the server market concentrates: with ~20% of units going to a handful of megacomputer projects, vendors' product roadmaps and pricing increasingly answer to a few hyperscale buyers rather than the broad enterprise base.
- For Microsoft specifically, the statistic marks an internal rebalancing — resources and research attention tilt from selling boxed software toward operating and optimizing datacenter-scale systems.
Second-order effects
- Competing web platforms — Google, Amazon, Yahoo — must match warehouse-scale buildouts or lose on cost-per-query and service latency, while server and component makers begin designing product lines around hyperscaler requirements instead of traditional IT procurement.
- Software economics start bending toward services delivered from these megacomputers, compounding the pressure on the packaged-license model whose cracks were already showing as Windows' dominance eroded.
Third-order effects
- If the pattern holds, compute consolidates into a utility-like layer owned by a small set of platform companies — the same dynamic that by 2019 had the field worried that frontier AI research would be limited to big tech, and that eventually pushed Microsoft to design its own silicon like the Maia 100 rather than depend entirely on merchant chips.
- Bargaining power structurally migrates up the stack: whoever owns the megacomputer sets terms for software vendors, chip suppliers, and eventually the researchers and startups who need access to that scale — a dependency relationship that defines today's cloud-AI economy.
The trend: This is one early data point in the long consolidation of computing from distributed PCs and enterprise servers into warehouse-scale megacomputers controlled by a shrinking set of platform companies.