🔬 The most advanced AI chips on Earth can't be made without a light-sensitive goo that almost nobody knows how to formulate. The world leader in that goo isn't Intel, isn't TSMC, isn't a Silicon Valley startup. It's a Japanese chemical company, founded in 1936, that got its start making battery material for the lamps coal miners wore on their helmets. Here's how Tokyo Ohka Kogyo ended up holding one of the most strategic chokepoints in the global chip supply chain.

From coal-mine lamps to silicon

In 1936, an inventor named Shigemasa Mukai set up a small lab in Tokyo, the Tokyo Ohka Research Institute. His first product had nothing to do with electronics: a high-purity potassium hydroxide for the batteries in the cap-lamps miners wore underground. Mukai's stated ambition, in the company's own telling, was to make things "useful to society that other firms wouldn't touch." The lab was incorporated as Tokyo Ohka Kogyo (TOK for short) in 1940.

The turn toward electronics came in 1955, when TOK localized production of potassium silicate, a material needed to make the cathode-ray tubes inside early televisions. That pulled the company into Japan's booming postwar electronics industry. Then, in 1968, TOK did something no Japanese firm had done before: it made a domestic photoresist for semiconductors.

A photoresist is the light-sensitive film that makes a chip possible at all. In photolithography, light is shone through a patterned mask onto a wafer coated with resist; the exposed areas change chemically, and that is how a circuit finer than a virus gets printed onto silicon. No resist, no chip.

The part of a chip almost no one can make

Today TOK is the world's No. 1 photoresist maker, with a 24.7% share of the overall market in 2024 (by projected shipment volume, per Fuji Chimera Research). But the number that matters most for the AI era is a different one: 28% of the market for EUV resist, where TOK ranks second in the world.

EUV — extreme ultraviolet — is the shortest-wavelength light used in chipmaking, and it is what makes the leading-edge processors behind AI possible. The shorter the wavelength, the finer the pattern you can print. But EUV resist is brutally hard to formulate, because it has to nail three things at once that pull against each other: sensitivity (reacting to very little light), resolution (rendering extremely fine lines), and suppressing roughness (keeping the edges of those lines clean). Improve one and the others tend to slip.

The purity bar is just as punishing. TOK describes its standard as tolerating not a single drop of impurity in a 50-meter competition swimming pool. You can't clear that with a checkbook. It is the residue of decades of trial and error in formulation, which is exactly why so few companies can play in this field at all.

Decades of recipe-tweaking, suddenly priceless

For most of TOK's history, the patient knowledge of which recipe works for which customer's process was a slow, unglamorous business. Then generative AI arrived.

The accelerators inside the data centers training AI models sit at the absolute leading edge — the kind of chips that can only be made with EUV lithography. And EUV lithography needs EUV resist. Demand for precisely the thing TOK had spent decades getting good at went vertical. The recipe optimization built up since 1968, combined with "switching costs" so high that a fab is loath to change suppliers once a resist is qualified into its process, hardened into a genuine moat. The business outlet Toyo Keizai cast the company as the firm that "broke open" Japan's once import-dependent chip-material supply with its own technology.

From ¥100 billion to ¥227 billion in six years

The numbers tell the breakout. TOK's revenue ran around ¥100 billion (about $625 million) in fiscal 2019. For fiscal 2025 the company expects roughly ¥227 billion (about $1.42 billion), more than double in six years. Sales topped ¥200 billion for the first time in 2024, a record. Electronics materials like resist now make up roughly half the business, and a large slice of revenue, around 30%, goes to a single customer, TSMC. About 80% of sales come from outside Japan.

To keep pace, TOK is building. A new plant at its Aso-Kumamoto site came online in 2025, and two new factories in South Korea are slated to start up in 2026 and 2027, close to the Samsung and TSMC fabs they will feed. The company has set a 2030 target of ¥350 billion in revenue (about $2.2 billion), up 54% from its FY2025 forecast.

The rivals — and a fork in the road

TOK's closest rival is JSR, the EUV front-runner, and the two are now betting differently on what comes next. The incumbent technology, dominant for about three decades, is "chemically amplified resist" — organic-polymer chemistry that leans on additives such as photo-acid generators and sensitizers. It is the field TOK has refined since 1968, and where it holds the widest lineup, ranking No. 1 in older KrF and g/i-line resists. The challenger is "metal-oxide resist": tin-oxide chemistry that soaks up EUV light far more efficiently, aimed at the sensitivity and resolution that sub-2nm chips will demand.

The strategies split from there. JSR moved first, buying US metal-oxide pioneer Inpria in 2021 and planning a plant for it in Cheongju, South Korea, around 2026. It also took a path TOK has not: in a roughly ¥1 trillion deal, JSR was taken private in 2024 by the state-backed Japan Investment Corporation, turning the EUV leader into something close to a national champion. TOK stays an independent listed company, developing its own metal-oxide resist with less fanfare while leaning on its conventional-resist depth, with a new EUV resist building in Fukushima due in the second half of 2026. Shin-Etsu Chemical and Fujifilm, the other EUV-capable Japanese makers, are expanding too. Whether the 30-year-old approach holds or metal-oxide rewrites it is the open question, and TOK is hedging both sides.

So who really controls the AI chip?

It's become a familiar line that ASML, the Dutch company, has a near-monopoly on the EUV lithography machines themselves. Less discussed is the other half of the chokepoint: the materials those machines pattern. And there, Japan's grip is even tighter. (It reaches the machines that apply the resist, too — Tokyo Electron holds 90%-plus of the coater/developer market, as we covered in Japan's chip-equipment makers.)

Of the top five makers of advanced-lithography photoresist in 2024 — JSR, TOK, Fujifilm, Shin-Etsu Chemical, and South Korea's Dongjin Semichem — four are Japanese, and together the five hold around half the global market. Across all photoresist, Japanese makers control roughly 80%. Narrow it to EUV resist with real mass-production capability and the field shrinks to essentially four Japanese firms — JSR, TOK, Shin-Etsu, and Fujifilm (per research firm Fuji Keizai). America's DuPont and Germany's Merck (AZ) compete around the edges, but in EUV the front-runners are Japanese: JSR first, TOK second, effectively setting the pace.

Strip the picture down and it is stark. The most advanced AI chips on the planet — the ones training the models reshaping the global economy — depend on a light-sensitive film that a handful of Japanese chemical companies, led by a firm born from coal-mine lamps, are uniquely able to make. TOK is one node in a wider materials lock that runs from Ajinomoto's packaging film to Shin-Etsu's silicon wafers (more on that here).

In Japan, that fact lands as pride for some and as a warning for others: a single fire, earthquake, or geopolitical shock near a few plants could ripple through the whole world's chip supply. What does the chip supply chain look like from where you are, and could your country build this kind of capability in under ninety years?

References