What Chip War Taught Me About How Technology Becomes an Industry

After finishing Chip War, the first image that came to mind was not a chip. It was a moth.

In 1947, Harvard’s Mark II was an electromechanical computer built from roughly 13,000 electromagnetic relays. When the machine produced a wrong result, the team traced the problem to relay number 70. A moth was stuck between its contacts and had shorted the circuit. Someone removed it, taped it into the logbook and wrote that they had found the first actual “bug.” This is the story most often told about the origin of “bug” and “debug.”

I initially believed the common version: Grace Hopper had discovered the bug and, almost incidentally, invented the word. A little checking showed that both parts were wrong. “Bug” had already been used for technical faults in Edison’s era; the moth was simply the first time the word had acquired such a literal meaning in a machine log.

I put this small episode at the beginning because it unexpectedly points to what the book is really about. Relays and vacuum tubes were mechanical switches that generated heat, oxidized and attracted insects. Their unreliability was one of the direct reasons engineers wanted transistors to replace them. The beginning of the chip war was therefore not an abstract grand strategy. It was a group of engineers becoming tired of repairing moths and contacts and wanting a switch that would not fail at random.

The book changed how I think about the way technological progress happens.

1. A breakthrough is only the beginning

The history of Soviet semiconductor development and production made the strongest impression on me. The book describes Soviet weapons designers in the 1960s deliberately reducing the number of electronic components in their weapons because domestically produced semiconductor devices were not stable enough in performance and reliability.

My first explanation was simple: this was the familiar problem of a planned economy. The Soviet Union could concentrate resources to build nuclear weapons, but could not push forward a technology that required continuous iteration. The explanation was so neat that I became suspicious that I was merely applying a ready-made conclusion.

What helped me understand the issue was not the book but my own experience. Earlier in my career, I spent several years in bearing manufacturing, working on product development and process SPC, or statistical process control. New-product trial production and mature-product volume manufacturing are different things. The first is often about selecting qualified parts from a batch. The second is about making the manufacturing process itself more stable and raising the yield over time.

The first relies on screening. The second relies on iteration.

I suspect Soviet semiconductor production followed the same logic. It may have been possible to select a very small number of usable devices for the most important weapon systems, but the system did not successfully build the cycle in which yield, cost and stability improve generation after generation.

Technical breakthroughs are never the whole story. The harder task is making a system capable of moving itself forward: an initial demand creates manufacturing capability; manufacturing capability lowers cost; lower cost opens a larger market; and the larger market supports another round of R&D. Once that cycle starts turning, it is harder to catch than any isolated technical breakthrough.

2. The state and the market do different jobs

The book also changed how I think about the relationship between the state and the market. Early semiconductor orders in the United States came substantially from the military and NASA. Civilian demand could not yet support a technology that was expensive and uncertain. The government became the first customer willing to pay.

Japan’s later catch-up involved a different form of state participation. Its banking system and industrial policy allowed companies to tolerate several years without profit, reduce yield and cost, and then compete for the market.

My earlier view was simpler: markets are good and planned economies are bad. After reading the book, I think the more useful formulation is that the state and the market have different jobs.

The state can take on costs that the early market is unwilling to bear: basic research, first demand and shared testing infrastructure. The market does something else: it tests real demand, selects among technical routes, eliminates mistakes and drives cost down.

The difficult part of a technological revolution is that the state can help start the process, but the industry eventually has to build the self-reinforcing cycle itself. It cannot depend forever on external transfusions.

I could not help wondering what this principle would look like in the diamond industry I follow every day. I will leave that direct comparison for another article; here it is enough to keep the question in view.

3. Specialization grows from trusted interfaces

The third lesson concerns industrial organization. The semiconductor industry gradually moved from the IDM model, in which design and manufacturing were integrated, toward a division that included fabless design and dedicated foundries. Around them grew specialized companies focused on materials, equipment, EDA software and packaging.

I used to think that this division of labor had been designed by someone sitting around a table. I now think a more accurate description is that specialization grows after manufacturing capability becomes standardized enough to be trusted by outsiders.

Only when there are common design rules, common process windows and common acceptance procedures can two unfamiliar companies risk putting their critical interests in each other’s hands.

The strongest feature of a mature industry is not that one company can do everything. It is that many companies can each do one thing and combine through standards and interfaces to make something extremely complex.

Once industry standards and interfaces are well developed, companies in the chain can focus on what they do best. They no longer need to renegotiate the interface with every upstream or downstream partner for every new product.

4. Strategy is not the same as predicting the future

The case of Intel declining to manufacture chips for the iPhone was the part I had understood least clearly. The familiar version says that Intel was blinded by short-term profit and failed to see the mobile-internet opportunity. I initially accepted that explanation and extended it into the familiar lesson that complacency is the real risk.

I later looked up Paul Otellini’s own account in an interview with The Atlantic around the end of his tenure. The situation was more complicated. His explanation was that Apple’s target price was below Intel’s predicted cost, and that this was not a business where volume could simply spread the cost. In hindsight, he said, the cost prediction was wrong and the volume turned out to be 100 times anything people had imagined at the time.

This looks more like a specific forecasting error than a collision between “profit-seeking” and “failure to see disruption.” I wrote this down not to reject the idea that complacency is dangerous. I still agree with it. The point of emphasis has changed, however.

The core of strategic vision is not whether a company can predict the future correctly. It is whether it can recognize that its old logic may be starting to fail while that logic still works, and test the new path in a way that allows it to stop without catastrophic damage.

That does not mean a company should test every imaginable path. Companies have to survive, and employees need stable work. Experiments have to be conducted while cash flow and normal operations remain protected.

5. Efficiency becomes more complicated under geopolitical competition

The second half of the book follows the semiconductor supply chain. Whoever controls an irreplaceable link has leverage, and the supply chain becomes a central instrument of national competition.

Technology may originally allocate resources according to cost, efficiency and profit. Once it becomes part of geopolitical competition, the evaluation criteria also include security, supply continuity and control. Decisions can no longer be made simply by asking who can do something best and most cheaply.

That does not mean efficiency has stopped mattering. It means that the meaning of efficiency itself has become more complicated.

The lens I took away

The most practical result of finishing the book is a way of looking at industries. A technology-only view is not enough. Technology, capital, markets, organizations and the state have to be considered together.

That is why I kept comparing what I was reading with the diamond industry I know. Many of the lessons from Chip War can be useful there too: a technical breakthrough must become repeatable manufacturing; the market must distinguish real demand from a demonstration; standards and interfaces must make specialization possible; and public support must eventually turn into an industry that can sustain its own learning cycle.