When the Cycle Stopped Being the Story

One of my coaching clients asked me about semiconductors last week.

He was not expecting what I said.

He is a credit analyst. His job is to evaluate whether a company can service its debt through good times and bad. He looks at cash flow, margin durability, leverage, and the question of what happens when conditions turn against the business. Applied to semis, that lens has historically produced a particular kind of caution. The companies are capital intensive. Their cash flows swing hard with the cycle. Their debt gets tested at exactly the moments they can least afford it.

The institutional consensus on semis has split. Some investors are deep into the AI story and have repositioned aggressively. Others, especially outside of growth, are still working from the framework I learned in the 1990s. The headlines have moved faster than the reflexes.

I learned my framework from an old-school semiconductor analyst who had spent years inside the industry before coming to Wall Street. He taught me to recognize the patterns that defined the sector then. PCs were the dominant end market. Cycles lasted about eighteen months. Memory chips were commodities. The best companies struggled to hold their advantages for long because competitors caught up so quickly. Trade them off the trough. Do not own them through the cycle. The companies cannot get a lasting edge.

I did the work. Year after year. Eventually I could recognize within minutes what kind of semi situation I was looking at. The reps build judgment. The judgment gets faster. The pattern recognition becomes the edge.

A junior analyst sees data points. A senior investor sees the pattern. You walk into a meeting with a chip company management team, and within ten minutes you know whether you are looking at a cyclical trade or something structural. You read a sell-side report and the picture forms before you finish the first page. The recognition is fast. The speed is the value. The speed is also what makes the recognition invisible to its own operator.

That last part is what most experienced investors never name out loud.

The recognition gets so fast that the thinking stops being conscious. The pattern stops being something you notice and becomes the way you see. You are not applying a framework anymore. You are looking at the industry, and the industry is telling you what it is. Except it is not the industry telling you. It is the pattern you were trained to see, doing its work in the background, faster than you can interrupt it. Cyclical. Commodity. No lasting edge. The recognition fires in milliseconds and feels like simple observation.

This is how a framework becomes invisible. The recognition gets so reliable that the assumptions underneath it never have to surface.

My semi pattern recognition was built on specific assumptions. Chips were sold mostly to PC companies. Cycles were short. Design advantages were temporary. Capital intensity was a recurring drag on the business. I did not hold those as opinions. I held them as facts about the world. The recognition had been calibrated against those conditions for so long that I had stopped seeing them as conditions at all.

Then the conditions changed.

The customers diversified. The cycles stretched. The design wins started to stick. The capital intensity, which had always been a cost, became a barrier that kept new competitors from entering. The patterns I was trained to see were still firing, fast and confident, on an industry whose foundation had quietly shifted. By the time the framework visibly failed, the patterns had been pointing at a world that no longer existed for years.

This is the hardest part of being experienced. The edge that took a career to build is the same mechanism that makes you slow to see when the world has changed. The pattern recognition is doing work it cannot tell you it is doing. It is producing answers that feel like observations but are actually inferences from a model that may no longer fit.

The place I have seen this fail most often is the dynamic between a portfolio manager and a research analyst.

I sat in the PM seat for years. The analyst comes into the room with an observation that does not fit the pattern. The framework says this kind of company cannot win for long. The analyst is seeing something in the data, the customer conversations, the competitive dynamics, that does not match what the framework predicts. They put it on the table and say some version of I think this time it is actually different.

In that moment, my pattern recognition fired.

It fired fast and with confidence. Every cycle, someone says this time is different. They are usually wrong. The framework reasserts itself. Stay disciplined. I had been right about this reflex many times. The reps reinforced it. The dismissal felt like good judgment.

I had many moments like this. Some of them I got right. The framework did come back. The analyst was responding to a head fake. The discipline held. Those moments do not stay with me.

The moments that stay with me are the ones where the analyst was right. Not because the framework had been wrong before, but because the conditions underneath it had actually shifted. The analyst, who had not yet built the trained skepticism I had spent fifteen years developing, was the only one in the room who could see what was actually in front of us.

I missed some of those calls. Not because I was being arrogant. Because my pattern recognition was firing faster than my willingness to interrogate it. The reflex was protecting me from a hundred wrong calls. It was also protecting me from the one I most needed to hear.

This is where serious investors lose the most. Not on the analytical question itself, but on the openness to a junior person whose pattern recognition is not yet calibrated to the patterns the senior person cannot stop firing.

What I learned, over time, is that the tell is in the speed of your own dismissal. When you are batting back the analyst's observation before they have finished describing it. When the framework's answer arrives before the analyst's case has fully landed. That speed is the signal. Not that you are wrong, but that you have stopped interrogating the recognition.

Semiconductors are the cleanest place to watch this play out right now, because the change has been so substantial and the old patterns are so deeply trained into a generation of investors.

The demand base broadened. Mobile, cloud, automotive, and AI each added a new source of growth without replacing the last. The industry consolidated into a smaller group of scaled franchises with real pricing power. Business models specialized. Fabless designers captured returns on capital that the old vertically integrated structure could never deliver. The chips themselves got smarter, with custom silicon and software stacks doing work that pricing power alone could never accomplish.

The industry that had no lasting edge began to develop one. Longer customer commitments. Higher returns on capital. Real switching costs in the parts of the chain where the design work mattered most.

Repeatability.

Of a kind that did not exist before. Not the kind anyone should call permanent. No category of business gets to be permanently repeatable. Software is being asked that question right now, and the answer is no longer obvious.

But here is where the framework has to update carefully, not flip to the opposite extreme. The industry is more repeatable than it was. It is not free of cycles. It has become what I would call a growth cyclical.

A true cyclical is a business whose cycle is driven primarily by the broader economy. Housing. Autos. Heavy industrials. Demand contracts when the macro slows and expands when it accelerates. A growth cyclical is different. It has cyclical features such as capital intensity, inventory swings, and lumpy customer orders. But the cycle is not primarily driven by the macro. It is driven by dynamics inside the industry's own demand structure. Hyperscaler capex timing. Product cycles. The way design wins translate into volume two and three years later. The underlying secular growth is real. The cycles happen on top of a rising baseline rather than around a flat one.

The investor who applies a true cyclical framework to a growth cyclical underestimates the structural growth and trades too actively. The investor who applies a pure secular growth framework gets blindsided by corrections that are still part of the underlying business. 2022 was the clean reminder. Demand air-pocketed. Customers had double and triple ordered. Inventories had to clear. Gross margins compressed across the sector. The good companies survived intact, which is itself evidence of the structural change. The correction was real. The cycle was real. The secular growth was real. All three were true at the same time.

Look at where we are now. Semis are leading the market. The AI buildout is the largest capital cycle the industry has ever seen. The structural change is real. The cycle is also real. Some of the current AI capex enthusiasm will be revealed, in retrospect, as the kind of overshoot every capital cycle produces.

The investor who decided in 2002 that semis were dead and the investor who decides in 2026 that semis are software are doing the same thing. They are anchoring on a single narrative instead of holding the more complicated truth.

The structural truth that everyone agrees on can quietly stop being true while everyone is still agreeing.

You cannot leave anything for dead. Not an industry. Not a company. Not a category of investment you wrote off years ago.

And you cannot trust the pattern recognition that has served you well to keep serving you in conditions it was not calibrated for. The framework is what you can see. The patterns are how you recognize. The assumptions are the conditions everything rests on.

The framework is the last thing to break. The patterns can keep firing reliably long after the assumptions underneath them have started to shift. The investor who waits for the framework to fail before updating is always late. The investor who interrogates the assumptions on a cadence, quietly and regularly, before any failure makes it urgent, sees the change coming.

The investors who decided long ago that semis were uninvestable were not wrong about the past. They stopped paying close attention and missed one of the most important shifts in the modern market. The investors who conclude today that semis are software will get caught by the next correction.

Both mistakes have the same root. They stopped interrogating their assumptions.

This week's question:

Which of your assumptions have you stopped noticing are assumptions? Not the conclusion you draw from them. The conditions underneath the conclusion. The things you took for granted when you built the framework, that you have not interrogated in years. What would change if you let any of them be questions again instead of settled facts?