The Fundamentals Didn’t Change. Your Strategy Should.

The Fundamentals Didn’t Change. Your Strategy Should.

A veteran global CIO breaks down ninety years of computing history to expose the flawed assumption driving today’s AI capital decisions — and what senior leaders should actually be watching before the next correction.

30 years of computing history says more about your AI roadmap than any vendor deck will

Every board I sit in front of this year asks some version of the same question: Is AI a new era of computing, or the same era moving faster? The honest answer is both, and most leadership teams are making expensive decisions because they haven’t separated the two. The mathematics underlying computing has not changed in ninety years. The economics of deploying it have changed every decade. Confuse those two layers, and you will either underinvest in something durable or overpay for something that was never going to last. This article is about telling the difference, before your balance sheet does.

The Pattern Nobody in the Boardroom Is Naming

I have sat through five distinct technology cycles as a CIO. Mainframes to client-server. Client-server to web. Web to mobile and cloud. Cloud to data platforms. Now this. Each one arrived with the same promise: this changes everything. Each one was, underneath the noise, the same logic running on cheaper, faster infrastructure.

Binary representation. Stored-program execution. Information theory. These ideas are from 1936 to 1948. Turing, Shannon, von Neumann. Nobody has replaced them. Nobody is close to replacing them. What changes every decade is not the logic — it is whether the economics finally make that logic cheap enough to deploy at scale.

That distinction sounds academic. It is not. It is the single most useful filter I have for separating a real capital decision from a fashionable one.

When the transistor matured, it did not invent new logic. It made existing logic absurdly cheap. When the internet matured, it did not invent new computation. It made existing computation reachable from anywhere. AI is no different. The architecture behind today’s models has existed in pieces since the 1980s. What changed between 2012 and 2017 was not the mathematics. It was GPUs, internet-scale data, and one paper solving a parallelization bottleneck. Old logic. New economics. None of this implies AI is merely another infrastructure cycle. Transformer architectures, reinforcement learning, and agentic systems represent genuine advances in how software is built and deployed. The lesson from history is not that AI is unimportant. It is that transformative technologies still obey economic realities.  That is the entire pattern of computing history in one sentence, and it is the lens every leader in this room should be applying to their AI budget right now.

Every Major Technology Shift Arrives as a Burst, not a Slope

Boards love a smooth growth chart. Technology does not deliver smooth growth charts. It delivers long, quiet plateaus, then sudden jumps, then plateaus again.

The reason is simple once you see it: progress runs on multiple independent curves at once — compute cost, data availability, algorithmic technique — and nothing visible happens until two or three of those curves cross a usability threshold together. Computers got cheaper for fifteen years before anyone noticed. Data accumulated for two decades before anyone monetized it. The moment they converged; the world called it sudden. It was not sudden. It was finally legible.

This matters for how you plan capital, not just how you understand history. If you wait for the visible jump before you act, you have already missed the window where the smart money is positioned. If you chase every visible jump as if it is permanent, you will overbuild the way telecom overbuilt fiber in 1999 — real infrastructure, real capital, a decade ahead of real demand.

Your job is not to predict the jump. Your job is to know which curve you are actually betting on, and whether that curve is close to its threshold or nowhere near it.

Unprecedented Adoption Is Not the Same as Durable Demand

Here is the belief I want to challenge directly, because almost every leadership team I talk to is quietly operating on it: massive AI user growth proves massive AI value.

The more useful signal is enterprise value creation. Across software development, customer operations, compliance, and knowledge work, organizations are already reporting measurable gains in productivity, cycle-time reduction, and automation. The question is no longer whether AI can create value. The question is where value is durable enough to justify long-term capital allocation.

It does not. It proves a massive AI trial. Those are different things, and the gap between them is where capital gets destroyed.

The overwhelming majority of consumer AI usage today is unpaid. Free tier usage dominates the headline numbers you see quoted in every strategy deck. Free-to-paid conversion across the industry sits roughly where freemium software has always sat — a single-digit percentage. One of the largest AI consumer companies in the world is already projecting an 80% decline in its primary paid subscription tier this year, planning to replace that revenue with cheaper, ad-supported access instead. Read that again. The company itself does not believe most of its current paying users will keep paying at today’s price.

That is not a company in crisis. That is a company being honest about freemium economics — the same economics every SaaS category has lived with for twenty years. But it should stop every CIO in this room from citing "a billion weekly users" as evidence of enterprise-grade demand. Consumer curiosity and enterprise willingness-to-pay are not the same signal, and treating them as the same signal is how strategy decks get the multiplier wrong.

The more reliable signal sits one layer down, in enterprise retention and contracted usage, where the willingness to pay is tied to a measured cost reduction or productivity case, not discretionary spend. If you want to know whether AI is durable in your organization, stop watching the user-growth chart. Start watching whether the business case survives a renewal conversation.

What This Means for Capital Allocation, Not Just Curiosity

Three decisions follow directly from separating the logic layer from the economics layer.

1. Distinguish infrastructure bets from application bets. 

The current AI buildout has two very different risk profiles stacked on top of each other. Physical infrastructure — compute, data center capacity, power — behaves like the telecom fiber cycle of the early 2000s: real capital, financed partly through debt and circular vendor arrangements, betting on a demand curve that has to keep compounding to justify the spend. Application-layer software sitting on top of that infrastructure behaves more like the dot-com era: thin, fast-moving, and far more exposed to being replaced the moment the underlying platform adds the same feature natively. Know which one you are funding. Underwrite them differently.

2. Stop pricing vendor promises against an extrapolated trend. 

Every overbuild in computing history shares one fingerprint: capital expenditure justified by assuming today’s growth rate continues indefinitely, rather than by today’s actual unit economics. Ask your own organization the uncomfortable question directly: Does this initiative work if adoption merely plateaus rather than keeps compounding? If the business case only survives under the optimistic curve, you do not have a business case. You have a bet on someone else’s forecast.

3. Watch the architecture layer, not just the capability layer. 

The deepest principles of computing — what is computable, what information theory permits — are not moving. But the engineering default that has held since the 1940s, an explicit human-written instruction executed deterministically, is genuinely being challenged for the first time by systems that learn their own internal logic rather than having it specified. That shift changes where your governance, audit, and accountability structures need to sit. It does not change whether the underlying computer is still a computer.

What to Do Monday Morning

Separate what is permanent from what is merely current

Before approving the next AI initiative, ask which layer it depends on: the ninety-year-old mathematics, which is not going anywhere, or this decade’s economics, which has reversed before and will again. Capital deployed against the first is infrastructure. Capital deployed against the second is a trade, and should be sized and governed like one.

The Discipline Boards Actually Need

History does not ask you to predict the cycle. It asks you to know which part of it you are standing in.

I have never once seen a technology cycle that rewarded the leader who got the timing exactly right. I have seen many who punished the leader who could not tell a durable capability from a fashionable one. That is the discipline this moment demands — not prophecy, just clarity about what you are actually betting on, and whether your organization survives if the optimistic curve does not show up on schedule.

The fundamentals have outlasted every cycle I have led through. Your strategy should be built on the part that outlasts the cycle, not the part that is merely loud this year.

AI will likely become embedded in every major business process over the next decade. The risk is not investing in AI. The risk is confusing durable capability with temporary enthusiasm. History suggests the winners are rarely the organizations that adopted first. They are the organizations that understood where value would ultimately accrue and allocated capital accordingly.

Where is your organization placing its bet — on the mathematics, or on the moment? 

I would like to hear how your board is drawing that line.

#Leadership #CIO #DigitalTransformation #ArtificialIntelligence #BoardroomStrategy