The Foundry Wars

Section 01

The Bet America Made

When Congress passed the CHIPS and Science Act in 2022, the diagnosis driving it was stark. The world's most advanced semiconductors — the chips running every AI model, every smartphone, every advanced weapons system — were manufactured almost entirely in one place: Taiwan.↗ A single miscalculation in the Taiwan Strait, a blockade, a natural disaster, could sever access overnight. COVID had already shown the world what a disrupted chip supply looked like: idle car factories, delayed laptops, hospitals scrambling for ventilator controllers.

The legislation committed $52 billion to rebuilding American semiconductor manufacturing. Intel, TSMC, and Samsung all announced new U.S. facilities in quick succession. Samsung chose a greenfield site in Taylor — a small city east of Austin with cheap land, favorable tax treatment, and a state government that treated the deal like a moon landing.

The investment that started at $17 billion would eventually balloon to $44 billion, making it one of the largest foreign direct investments in American history. Samsung would receive $6.6 billion in CHIPS Act grants, with the grants tied to hitting production milestones. The strategic logic was straightforward: build it, and they will come. Customers would follow capacity.

They didn't. Not at first.

Section 02

The "No Customers" Crisis

Samsung's foundry division — the part of the company that manufactures chips designed by other companies — had been struggling globally before the first shovel broke Texas ground. Its 4-nanometre process technology, the node the Taylor fab was originally built around, had suffered from persistent yield problems: too many defective chips per wafer, making the economics unreliable for volume customers who need predictability above almost anything else.

The major buyers of cutting-edge foundry capacity — Apple, Nvidia, AMD, Qualcomm — had already committed the bulk of their production to TSMC, which holds roughly 67% of the advanced foundry market and has spent a decade building a reputation for reliability that Samsung's foundry division could not yet match.↗ Samsung held approximately 8%.

By mid-2024, the situation at Taylor was being described inside the company with unusual bluntness. Reports cited sources who summarised the problem in five words: there are no customers.↗ The opening date slid from 2024 to 2025 to 2026. There were internal discussions about whether to pause the entire U.S. investment programme. The facility sat physically near-complete, its cleanrooms humming, waiting for orders that weren't arriving.

Section 03

The Deal That Changed Everything

In July 2025, Elon Musk announced that Tesla had signed a chip supply agreement with Samsung valued at $16.5 billion, running through December 2033.↗ It was the anchor customer Samsung's Taylor facility had been waiting for — and then some. The deal covers two generations of Tesla AI chips: the AI5, which powers Tesla's Full Self-Driving driver assistance system, and the more advanced AI6, which Samsung will manufacture at the Taylor facility at scale.

The announcement had cascading effects. Samsung upgraded the Taylor fab's target process from 4nm to 2nm — a significant technological leap that places it in direct competition with TSMC's most advanced capabilities.↗ Equipment move-in at the facility began in April 2026. Commercial production is expected to start in the second half of 2026, ramping toward full volume into 2027.

For a facility that eighteen months earlier had been described internally as a problem without a solution, the transformation was striking. Samsung's Taylor fab had gone from cautionary tale to strategic asset — not through gradual improvement but through a single partnership that repositioned its entire trajectory.

Section 04

The Geopolitical Wildcard

But the story of Samsung's Taylor facility is larger than one company's fortune reversal.

Taiwan's government recently implemented what chipmakers are calling the "N-2 rule" — a regulation that prohibits the transfer of the island's most advanced semiconductor technology overseas. In practice, this means TSMC cannot immediately export its 2nm manufacturing process to its new Arizona facility. American companies — AI developers, defence contractors, automakers — that need 2nm chips made on American soil currently have exactly one viable near-term domestic option.

Samsung's Taylor, Texas.

This is the context that transforms the chip race from an industrial competition into something more fundamental. The transistors on a 2nm chip are etched into spaces smaller than the diameter of a strand of DNA. Fabricating them at volume, reliably and cheaply, determines who builds the AI systems, the autonomous vehicles, and the weapons platforms of the next decade. For thirty years, the United States designed those chips and Asia manufactured them. The CHIPS Act was a bet that this arrangement could not survive the geopolitical pressures building in the Pacific.

Samsung's $44 billion Texas factory is that bet made physical. Whether it pays off now depends on whether a fab that spent eighteen months with almost no customers can scale fast enough to matter — and whether the Tesla deal is a floor or a ceiling for what comes next.

The cleanrooms in Taylor are filling up now, one machine at a time. The filtered air moves through them the same as before — particle-free, precisely controlled, waiting. This time, there is work to do.