The Lessons Japan Keeps Forgetting to Teach

In 1986, the United States and Japan signed the Japan-US Semiconductor Trade Arrangement — a bilateral agreement that, among other things, required Japan to guarantee that American chips captured at least 20 percent of the Japanese domestic market. It was a remarkable intervention: the world’s largest economy using trade policy to compel a nominally free-market ally to purchase specific quantities of a foreign product. It was also, by the standards of what came before, quite modest.

In the preceding decade, the American semiconductor industry had watched Japan’s DRAM producers — Fujitsu, Hitachi, NEC, Toshiba, Mitsubishi — capture an increasing share of the global memory chip market through a combination of genuine manufacturing excellence and government-supported dumping. The US industry’s response involved antidumping complaints, congressional pressure, trade negotiations, and eventually the semiconductor agreement. American DRAM producers still largely exited the market.

The story is more complicated than either the “Japan cheated” narrative or the “US industry lost because it couldn’t compete” narrative. Both contain partial truths. The lesson it offers for the current US-China competition depends on which parts of the story you read carefully.

The parallel with China begins with the structure of the competition. In both cases, a US adversary (or at least rival) used state support — directed lending, domestic procurement preferences, subsidized infrastructure — to accelerate the development of a domestic semiconductor industry that was genuinely more efficient at certain types of manufacturing than the market would have naturally produced. In both cases, the US industry was surprised by how quickly the gap closed, and its initial response was to attribute the competition’s success primarily to unfair practices rather than to genuine capability development.

The difference is political orientation: Japan was a US security ally, and the economic conflict was explicitly contained within the alliance framework. China is simultaneously a major trading partner and a defined strategic competitor, which makes the conflict harder to manage through the kind of bilateral negotiated arrangements that the Japan case produced. You can negotiate market share guarantees with an ally. You can negotiate with an adversary too, but the trust requirements and enforcement mechanisms are entirely different.

The more instructive parallel is in what happened to the US semiconductor industry after the Japan semiconductor wars. American companies largely abandoned commodity DRAM manufacturing, where Japanese companies had genuine cost advantages, and concentrated on the higher-margin, design-intensive segments — microprocessors, digital signal processors, graphics chips — where American engineering talent had distinctive advantages that were harder to replicate through manufacturing scale and government subsidy.

The Intel that emerged from the DRAM retreat was a company focused on microprocessors, and that focus produced the x86 dominance that ran the personal computer revolution. Qualcomm, founded in 1985 at the nadir of US DRAM competitiveness, built its business on wireless chip design rather than manufacturing — a model that did not exist in the 1970s because the wireless industry did not exist. The US semiconductor industry’s retreat from DRAM was a defeat in a specific battle that preceded the construction of a much larger advantage in a different part of the field.

The historical reading that is both correct and often misapplied is that American semiconductor success after Japan was not primarily the result of export controls, trade policy, or government industrial programs. It was the result of American engineers, universities, and venture-capital-funded companies finding new problems to solve that Japanese manufacturing excellence could not address. The design-centric model of semiconductor competition — fabless companies that design chips and contract out manufacturing — was an American innovation that the Japanese industry was structurally unable to replicate, because its competitive advantage was in manufacturing culture rather than design culture.

Whether an analogous retreat-and-rebound is available in the current competition is the strategically important question, and the honest answer is: maybe, with qualifications.

The qualifications are significant. The Japan competition was in memory chips — commodity products where manufacturing scale and cost are the primary competitive dimensions. The current competition is in AI accelerators — complex products where architectural design, software ecosystem, and manufacturing precision are all relevant, and where the design software (CUDA) and the software ecosystem (the AI research community’s accumulated code) represent an American advantage that is harder to replicate than DRAM manufacturing.

The retreat-and-rebound option would require identifying a segment of AI-relevant semiconductor competition where American design excellence is decisively superior to Chinese manufacturing scale — and then concentrating American resources there while accepting loss in the segments where Chinese scale economics are dominant.

This is, arguably, exactly what Nvidia’s competitive position represents. CUDA is not a manufacturing advantage. It is a software ecosystem advantage — the result of twenty years of developer investment that is as difficult to replicate as TSMC’s forty years of process knowledge. If Nvidia can maintain its software ecosystem advantage while gradually losing hardware manufacturing share to Chinese alternatives in the domestic Chinese market, the outcome might look more like the Intel microprocessor victory than the DRAM defeat.

The lesson Japan keeps not quite teaching is that industrial competition in semiconductors is never resolved by a single round. The 1986 semiconductor agreement and the subsequent trade pressures changed the landscape, but the outcome was determined by what each side did with its accumulated advantages in the subsequent decade. Japan’s DRAM manufacturers had a genuine manufacturing advantage that trade policy could not erase. They chose, partly under US pressure and partly due to their own strategic assessments, not to extend that manufacturing advantage into the next wave of semiconductor products.

South Korea’s Samsung and SK Hynix did extend their manufacturing advantage into the next wave — into DRAM generations that Japanese companies also could have made but chose not to prioritize — and those two companies now dominate global memory production while Japan’s DRAM industry is essentially gone.

The parallel for China is that current success in building Ascend-class AI chips for the domestic market is a first-round result. Whether it translates into enduring competitive position depends on whether Chinese chip companies can extend their capability across the technology generations that will define the next wave of AI hardware. The export controls are designed to make that extension harder. Whether they make it impossible is unknown and probably unknowable from the current vantage point.

What the Japan analogy does most clearly is challenge the binary framing that dominates current analysis. The question is not “will China catch up to US AI chip capability?” The question is “which capabilities will China develop, in which timeframes, and what will America have built in the same period?” The Japan case suggests that catch-up in specific defined capabilities is real and achievable; that American advantages in design, software, and ecosystem can outlast manufacturing advantages; and that the most important determinant of long-run outcome is what each country builds in the generation after the current competition.

The 1980s semiconductor wars ended without a clear winner. The US kept its most valuable position (design-intensive chips), Japan lost market share in the segment it had won (DRAM), and both industries are still operating today. The resolution was not dramatic. It was an ongoing negotiation about which capabilities each country’s industrial structure was best suited to develop.

The current competition will probably resolve similarly: not with a winner and a loser, but with a division of the field that reflects each country’s accumulated capabilities and industrial choices. The export controls shape, but do not determine, where that division falls.

That is the lesson Japan keeps not quite teaching. Probably because it is not the lesson anyone wants to hear.