It's the work, not the machine.

Scale was the answer

We've always chased more. Software outgrows hardware as fast as we can build it. The newest processors are fast, until they aren't. Storage is plentiful, until it fills. Bandwidth is abundant, until it saturates.

The fix always seemed obvious: build bigger, faster machines. And for decades it worked. Nearly every major advance came from adding capacity, with more processors, more memory, more storage, bigger data centers. Whole industries grew up around that one idea. Moore's Law became the roadmap, the cloud became the business model, and the hyperscalers became some of the most valuable companies on earth. Scale was the answer, and it was the right one.

The work fit the machine

Underneath all of it is a simpler pattern. Every era of computing is defined by the relationship between work and machines, and when that relationship changes, a new era begins. Mainframes ran batch work, personal computers ran interactive work, the cloud ran web-scale distributed work. In each case, the way we executed matched the work of the time. The work fit the machine.

Something changed

We're building bigger and faster systems than ever, and we keep losing ground. Modern workloads run everywhere at once, across clouds, regions, clusters, and edge environments, and teams now burn enormous effort just deciding where work should run, when it should run, and how it should move between resources. The systems are bigger than ever, and the work is harder than ever to place.

The reason is the old pattern running in reverse. The work changed shape, and the execution model didn't.

The mismatch is the cause

For most of the last twenty-five years, infrastructure was built around predictable work: web apps, APIs, databases, all of it scaling in fairly straight lines. That work has since grown larger, more dynamic, and more distributed than the systems beneath it were built to run, and it demands something different from the resources that execute it. The challenge is no longer scaling machines to do the work. It's matching the work to the right machines.

When the work no longer fits, we compensate. We layer on orchestration, scheduling, and management tools until the systems grow more complex and their inefficiencies compound. So what looks like a performance problem, a scaling problem, or a cost problem is usually the same problem seen from a different angle. Complexity is the symptom. The mismatch is the cause.

The bottleneck moved

For decades the binding constraint was capacity. Capacity still expands, faster than ever, but a growing share of our effort goes to coordinating work rather than running it. None of this means scale stops mattering. We'll keep building larger processors, clusters, networks, and data centers. It means raw capacity can no longer solve for work that's misaligned with the machine. The bottleneck has moved from compute volume to compute fit.

The next era is compute fit

The next era won't be defined by who builds the largest machines. It'll be defined by who best fits the work to them. The companies that shape the coming decade won't simply add more capacity. They'll unlock the value already sitting unused in the capacity we have.

Every era of computing is a story about the relationship between work and machines. When that relationship breaks, a new era begins, and I believe we're entering one of those moments now.

The work changed shape. The execution model didn't. It's the work, not the machine.