Samsung Electronics has started pilot production at its Taylor, Texas fab as it prepares for mass production of 2-nanometer chips early next year, according to industry reports from South Korea. The facility, originally designed as a general-purpose line focused on 4nm technology, has been retooled to add 2nm gate-all-around production and advanced 3D packaging as AI chip demand reshapes the foundry market.
The pilot lines now running in Taylor have one job: demonstrate real mass production yields to major clients who have already confirmed orders, and pass final quality tests. Samsung has secured contracts for next-generation AI chips with Tesla and for custom 2nm AI chips based on Arm architecture. Whether the first test chips meet customer standards for yield, heat generation and power efficiency is treated internally as a pivotal moment for the foundry business’s recovery.
Why Taylor matters to Samsung’s foundry push
Samsung’s foundry operation has spent two years losing ground to TSMC, which dominates advanced-node orders from Nvidia, Apple and most of the AI chip industry. The Taylor fab, announced in 2021 with a planned investment of roughly $17 billion, is the centerpiece of Samsung’s attempt to win US-based customers who want production outside Asia. The US CHIPS Act subsidy program has tied support to Samsung’s American capacity, adding political weight to the site’s success.
The pivot to 2nm reflects how quickly the AI buildout changed the calculus. A 4nm general-purpose line looked reasonable when the fab was designed. Now the money is in leading-edge AI accelerators, and Samsung’s internal process stability indicators are reportedly favorable. Industry sources cited in Korean reports say Samsung’s 4nm yield has stabilized above 80 percent, a level the company struggled to reach in earlier years.
One industry insider told Aju Press that achieving 80 percent yield for 4nm and improving 2nm gate-all-around yield into the 70 percent range indicates the processes have entered a stabilization phase. If Samsung can prove mass production capability by passing quality tests for custom chips from major clients, it would give the foundry business substantial momentum after years of underperformance.
The competitive picture
Samsung remains a distant second in foundry revenue. TSMC’s lead in yield and ecosystem maturity has kept the largest AI chip customers on its side of the Pacific, and Nvidia’s next-generation roadmap is locked to TSMC processes through the decade. Samsung’s realistic path is to win customers who specifically need US production, plus those seeking a second source for supply-chain reasons. Tesla’s AI chip order fits that pattern: diversification matters as much as price or yield for companies burned by single-supplier dependence.
China’s SMIC, ranked third globally, increased its second-quarter revenue by 20 percent compared with the previous quarter and now trails Samsung by just half a percentage point in market share. SMIC cannot access extreme ultraviolet lithography equipment due to export controls, which caps its leading-edge ambitions, but its momentum in mature nodes adds pressure from below. Samsung’s foundry market share, once above 15 percent, has slid toward the low double digits as TSMC’s dominance in advanced nodes concentrated the most profitable orders elsewhere.
The pressure is not only commercial. Samsung’s memory business, its traditional profit engine, has been riding the AI boom through high-bandwidth memory sales, which has softened the criticism directed at foundry losses. But HBM supply agreements with Nvidia are competitive against SK hynix, and the company cannot rely on memory profits indefinitely to fund a foundry operation that has required billions in annual support. Samsung’s leadership has publicly committed to making foundry profitable, and the Taylor pilot is where that commitment gets tested against physical reality.
Timeline and risks
The official pilot operation for the full line was postponed to later this year, with a target set for the third quarter. That delay is interpreted as a strategy to maximize initial yield through equipment setup and pilot runs before establishing a complete supply chain in time for mass production early next year. The sequence matters because early yield numbers get baked into customer decisions for multi-year supply agreements.
The risks are familiar ones. Advanced packaging, now central to AI chip performance, is a discipline where Samsung trails both TSMC and Intel. Power delivery and thermal constraints at 2nm make packaging choices as important as transistor density. And the fab needs a steady stream of US-based engineering talent, which has been a bottleneck for every foreign chipmaker building in Texas and Arizona. TSMC’s Arizona plants have faced the same staffing challenge, and both companies have flown in experienced engineers from Taiwan to bridge the gap.
There is also the question of follow-through demand. Winning Tesla and one or two Arm-based AI clients would validate the line, but Taylor’s economics assume a pipeline of customers behind them. Fabless US companies have historically been cautious about qualifying a second foundry at the leading edge, since each qualification costs months of engineering work and millions in test silicon. The US government’s push for domestic capacity has changed some of that calculus, and CHIPS Act incentives reward exactly the kind of second-sourcing Samsung is offering.
For now, the pilot runs give Samsung its clearest signal yet on whether the Taylor bet will pay off. Customers with confirmed orders are waiting on the same numbers, and the next few months of quality tests will decide whether Samsung’s US foundry push finally gets traction or joins the list of expensive delays that have marked the project since 2021.