SMIC in 2005: The Bet That the Door Would Stay Open

Twenty years on, what stands out about SMIC in 2005 is everything the people living through it could not have seen coming. The company had gone public the spring before, in 2004, on the Hong Kong and New York exchanges at once, one of the largest share sales of that year, four years after a Taiwanese-American engineer named Richard Chang started building fabs on farmland outside Shanghai. China was buying itself a chip industry in public, with foreign money and mostly foreign technology, and nobody in the building treated the foreign part as a risk. From here in 2026, that dependence is the entire story. In 2005, it barely rated a mention, which is exactly why what came later blindsided everyone.

Chang had left Texas Instruments after two decades, and before SMIC he founded Worldwide Semiconductor in Taiwan, the foundry TSMC absorbed in 2000. He knew how to stand up fabs, and he stood up several around Shanghai faster than the industry thought possible, hiring engineers from Taiwan and the United States and sourcing equipment and process recipes wherever he could. By the end of 2005, SMIC was running 0.18- and 0.13-micron processes and finally dragging 90-nanometer (0.09-micron) into production, late, after a year of telling analysts it would land sooner. Two or three generations behind TSMC and the pure-play foundry model it had taught the industry, on nodes the leaders had shipped years earlier. The thing that rattled people was not the gap; it was the rate of climb. No mainland foundry was supposed to be this close to the frontier five years in, and at some point in 2005, the rest of the industry stopped laughing.

Chartered Semiconductor in Singapore handed over a 0.18-micron logic process in exchange for capacity and equity. Toshiba supplied 0.15- and 0.21-micron flows. Fujitsu brought an FCRAM process. Infineon went deepest, swapping 0.14-micron DRAM technology for guaranteed memory supply, then widening the deal again and again to 0.11-micron, to 300-millimeter wafers, and in 2005 to a 90-nanometer DRAM trench transfer, with 70-nanometer dangling as the step after. SMIC climbed the node ladder by buying tickets, one technology-for-capacity contract at a time, filling its lines with other companies’ processes in exchange for the recipes. Nothing illegal in any of it. What it amounted to, structurally, was a national champion whose every advance was on loan from somewhere else.

The first 12-inch line in mainland China was switched on in Beijing in the spring of 2005, running 0.11- and 0.10-micron DRAM for Infineon and Elpida, and ramping from a few thousand wafers a month toward twenty-odd thousand by year-end. DRAM was a brutal, low-margin commodity, and that was exactly the point: it filled the fab and generated the volume that kept the lines loaded while SMIC chased the logic nodes that actually mattered long term. A leading-edge wafer size, on Chinese soil, roughly two years behind the leaders who pioneered it. That fab was the line item the state cared about, far more than any single quarter’s order book.

Broadcom anchored the order book, alongside Texas Instruments, STMicroelectronics, Samsung, and Fujitsu, the same fabless and IDM names that kept TSMC and UMC busy. A Chinese national champion whose customers were almost entirely not Chinese, fighting for those accounts in 2005 in a foundry market still bleeding from a price war it was widely blamed for starting. Its edge was structural: a first-mover position in China and a tax code that levied a higher value-added tax on imported chips than on domestic ones. TSMC and UMC had clocked that years earlier, and both spent 2005 maneuvering toward the mainland themselves. The home-field advantage came with a clock.

The trade-secret war with TSMC ended, or looked like it ended, in January 2005. SMIC agreed to pay $175 million over six years to settle the patent and trade-secret claims that TSMC had filed in California, with a patent cross-license running through 2010. It had spent eighteen months denying the worst of it, calling the suit a bully’s move, and then it wrote the check. The settlement granted no license to TSMC’s trade secrets, which is not the paper a winner walks away holding. Underneath the legal language sat the same dependency the technology-for-capacity deals already spelled out: SMIC’s speed came substantially from knowledge, tools, and people that originated somewhere ahead of it, some of it, TSMC was convinced, carried out of Taiwan in the heads of engineers SMIC had hired. From 2026 we know this was only round one; TSMC sued again in 2006 and won a far costlier settlement in 2009. In 2005, it read like closure.

The reason Beijing wanted its own foundry had nothing to do with any quarter’s wafer orders. Chips sat under everything with a processor in it, from a phone to a missile’s guidance package, and China imported almost all of its advanced ones from companies in countries that were friendly that year and might not be the next. That was the vulnerability the state was paying to close, with money and, more than the money, with patience, because catching up in this business does not happen on a timeline any ordinary company would tolerate. The grim irony, from 2026, is that the fear driving the entire project was the right one. Beijing only misjudged how soon the bill would come due.

A roughly $90 million profit in 2004 became a $30 million loss in the first quarter of 2005. A $40 million loss in the second, and the year closed more than $100 million underwater as foundry prices collapsed and SMIC’s heavy DRAM exposure turned from ballast into a hole in the boat. In July, the board removed Chang from the chairmanship. It installed Wang Yangyuan as chairman, keeping Chang on to run operations, the kind of reshuffle that happens when a founder’s story needs a steadier institutional face next to it. The fab construction never paused through any of it. Spending that hard on new capacity while bleeding cash only makes sense if the capacity was never meant to pay off in that year’s accounts, and it was not. It was meant to pay off in a decade that Beijing was willing to wait for.

Everyone in 2005 took one thing for granted: the tools and the talent would keep flowing. The chip business was getting more global every year, the old Wassenaar export rules looked like a Cold War relic nobody had the appetite to enforce, and the idea that the most advanced equipment might one day be deliberately withheld from China was not a scenario anyone serious was planning around. SMIC’s entire strategy, every technology-for-capacity deal and every imported tool in it, rested on that door staying open. Perso, I find the confidence almost unbearable to read now. Within fifteen years, the United States would put SMIC on the Entity List, the EUV machine it needed for the leading edge would be blocked before a single unit shipped, and by 2022, the controls would widen into the deliberate cutoff nobody in 2005 thought worth worrying about. SMIC clawed its way to 7 nanometers anyway, on machines it was never supposed to be able to use that way, which is its own kind of answer. The door everyone assumed was permanent turned out to be the whole game. They could not see it from inside the yard; it was still open.