The Frugal Architect- by Design
5th June, 2026
The era of unlimited compute is over. Not as a prediction — as a fact already visible in cloud bills, AI inference costs, and the quiet unraveling of architectures that were never as efficient as they appeared. For the better part of two decades, compute was cheap enough that it became a substitute for thinking. If a system slowed down, you didn't rethink it. You paid for more of it. The engineering culture that emerged from this period didn't discourage efficiency. It simply never required it. When compute is cheap enough, inefficiency has no consequences — and what has no consequences gets no attention. What is becoming clear, slowly, is that efficiency was never optional. It was just deferred.
Werner Vogels, CTO of Amazon, published his own Frugal Architect framework in recent years — a set of laws arguing that cost should be a first-class concern in system design, not an afterthought. The laws are worth noting. Make cost a non-functional requirement from day one. Architecture that doesn't account for cost isn't finished. Systems should scale cost sublinearly with load. Coming from the man who runs the infrastructure that powers much of the internet, the message carried weight. But there is more to add. Cost awareness is a necessary discipline. It is not, on its own, sufficient. Watching the bill tells you what the system is consuming. It does not tell you whether the system was designed well in the first place. The frugal architect operates at an earlier stage — before the instance is provisioned, before the query is written, before the pipeline is laid. The question isn't "how do we reduce what this costs." The question is "what does this system actually need to do" — and the lean design follows from the rigour of that answer.
What abundance architecture actually produces, when you look at it honestly, is systems optimized for speed of delivery rather than quality of design. Enterprise data platforms are a clear example — heavy middleware that pre-aggregates data because it's easier than writing precise queries, semantic layers so thick they require their own infrastructure to run, reporting stacks that execute expensive operations on every dashboard refresh regardless of whether the underlying data has changed. The same pattern repeats across the stack: oversized models applied to tasks that don't require them, pipelines that ingest everything on the assumption that storage is cheap — until it isn't. It was a reasonable response to the environment. When compute is effectively free and time-to-market is everything, you build for now and deal with the architecture later. The cost of that deferral compounds quietly — and at scale, it multiplies. Every redundant query, every bloated pipeline, every oversized model applied to a task that didn't need it — these don't just persist as the system grows, they scale with it. The inefficiency that was invisible at one data volume becomes the dominant cost at ten times the load. AI workloads make this unavoidable. Inference costs expose what abundant compute used to absorb.
There is another category worth naming. Abundance architecture also optimizes for appearance. Software designed and showcased in rich desktop environments — where fonts render beautifully, interfaces feel premium, and every visual detail is controlled — often carries that presentational overhead into cloud deployments where it was never appropriate. Real-time processing applied where near-real-time would serve the business equally well. Features built to demonstrate capability rather than solve a problem. The aesthetic was designed for the demo, not the deployment. These choices accumulate quietly — and they scale just as the inefficiencies do.
The principles that the frugal architect follows are not complicated, but they require deliberate commitment. Fetch what the query needs, not everything that might be relevant — in a data-intensive enterprise environment, that single discipline alone can reduce processing costs significantly. Process at the right frequency for the business decision, not the maximum frequency the infrastructure can support. Choose the model appropriate to the task — the largest available model is rarely the right one, and the cost difference between appropriate and excessive compounds quickly at scale. Design for the deployment environment, not the demo — what performs beautifully on a controlled desktop often carries overhead that was never accounted for in production. None of these are restrictions on what a system can do — they are decisions about what it should do. That distinction is where good architecture begins.
Artificial intelligence hasn't changed the principles of frugal architecture. It has simply made the cost of ignoring them visible in a way that is hard to look past. Token costs, inference bills, data pipeline bloat, the overhead of embedding and retrieval at volume — every inefficiency that abundant compute used to absorb now has a line item attached to it. For organisations that built without efficiency as a design principle, the bill has arrived for decisions made a decade earlier.
The frugal architect is not a conservative one. The discipline of efficiency is not a brake on technological advancement — it is what makes advancement sustainable. Building lean doesn't mean building less capable. It means making deliberate decisions at every stage to keep the system lean, so it can take on more sophisticated workloads precisely because the foundation doesn't waste what it has. As enterprises and regulators begin to take compute consumption and carbon footprint seriously, this matters beyond the balance sheet. An architecture that processes only what it needs, runs models appropriate to the task, and avoids the overhead of abundance by design — that is also a greener architecture. Not by intention, but by consequence. Efficiency and sustainability are the same thing expressed differently.
The goal was never to spend less. It was to build better. Good design is often simpler and more frugal than assumed — precise enough to do what is needed and no more. That precision is what produces systems that are cheaper to run, lighter on energy, and more durable when conditions change. Cost will bring most organisations to this conversation. Vogels made that case from the top of the industry. But cost is the signal, not the standard. Robust technology has always been engineered this way — lean by design, resilient by consequence, and more often than not consuming less energy than the alternative. The frugal architect does not build ‘cheap’ products. The frugal architect builds efficiently designed ones.
