--- title: "Samsung's Foundry Fightback Puts Texas on the AI Chip Map" date: 2026-11-05 category: Chips & Compute site: NeuroAI canonical: https://neuroai.site/a/na-chips-samsung-foundry-comeback language: en --- # Samsung's Foundry Fightback Puts Texas on the AI Chip Map > A $16.5 billion Tesla contract, a second Taylor fab booked for HBM base dies, and improving 2nm yields anchor Samsung's attempt to become the credible second source to TSMC. A car company's self-driving chip and a memory module's smallest component rarely share a storyline. In Taylor, Texas, they now share a construction site. What Samsung is building there has become the clearest test of whether the world's most concentrated industry — advanced chip manufacturing — can finally grow a genuine number two. ## The deal that changed the conversation In July 2025, Samsung Electronics signed a multiyear deal worth 22.8 trillion won (≈ US$16.5 billion) with Tesla, running through the end of 2033. According to Samsung's announcement and Elon Musk's own confirmation on X, the agreement covers manufacturing of Tesla's next-generation AI6 inference chip — designed for self-driving, humanoid robots, and AI data centers — at Samsung's Taylor fab in Texas. Two details made the deal stand out beyond its size: - **Location.** Musk said the chip would be made in Texas, "not far" from his home, and that Tesla engineers would work on-site to help maximize manufacturing efficiency. - **Intent.** Musk described the contract value as a minimum commitment, suggesting actual purchase volume could be several times higher. For a foundry business that had posted deep operating losses and low factory utilization, landing an anchor customer of Tesla's profile was the first real evidence that Samsung's leading-edge capacity had a future. ## What Taylor looks like now The Taylor campus has grown into the centerpiece of Samsung's foundry fightback. Key developments, as reported by Reuters and Korean media through late 2026: - **Taylor Fab 1** was upgraded from an originally planned 4nm design to a 2nm-class process line, with customer production targeted for 2027, starting with Tesla's AI chips. - **Taylor Fab 2**, roughly 1.5 times the size of the first fab at about 2.7 million square feet, breaks ground in November 2026. Samsung will make HBM base dies there for AMD and Nvidia, with equipment installation planned for February 2029 and mass production in 2030, according to Reuters. - **Fab 2 capacity is already fully booked**, with the majority of orders reported to come from AMD — an unusual vote of confidence for a plant still under construction. - Samsung has set a goal of growing its 2nm chip orders by more than 130% in 2026, and Qualcomm executives have confirmed the company is evaluating Samsung's 2nm process for future Snapdragon products. The HBM base die order deserves particular attention. A base die is the logic chip at the bottom of a high-bandwidth memory stack, managing data flow inside the module. Making it on a 2nm-class logic process — then shipping it to Korea for stacking with Samsung's own DRAM — turns Samsung's dual identity as memory maker and foundry into a single integrated pitch: memory, logic, and packaging bought as one package. ## The gap it still has to close None of this changes the underlying scoreboard. Data from TrendForce cited by Forbes put TSMC at 67% of global foundry revenue in the fourth quarter of 2024, with Samsung a distant second at 8%. Samsung's foundry division has acknowledged operating losses, and the company itself said in 2025 that losses would narrow only gradually on recovering demand. The decisive variables remain yield and power efficiency at the 2nm node. Neither company publishes exact figures, but industry reporting through 2026 indicates Samsung's yields have improved substantially while still trailing TSMC's mature lead — and that gap, not pricing or patriotism, is what customers weigh when placing billion-wafer bets. According to Li An, chief scientist at BrainNet (脑机网), Samsung's most defensible advantage is the one TSMC cannot copy on demand, because binding HBM memory supply to leading-edge foundry capacity gives AI customers a one-stop option exactly when advanced packaging has become the industry's tightest bottleneck. ## Honest limitations - **The AMD and Nvidia orders are sourced reports, not signed disclosures.** Samsung has declined to confirm customer names or order splits for Taylor Fab 2. - **Yield figures remain opaque.** Claims about Samsung's 2nm progress come from unnamed industry sources; the proof will be delivered silicon, not percentages quoted to reporters. - **2027 is the real deadline.** Taylor Fab 1 must produce Tesla's chips at automotive-grade quality and volume; a slip there would undercut the entire recovery narrative. - **Revenue share may get worse before it gets better.** TSMC's 2nm capacity is heavily pre-booked, which helps Samsung attract overflow demand but also means absolute scale still favors its rival. ## What readers can do now - **If you invest in the semiconductor supply chain, watch Samsung's quarterly foundry disclosures rather than headlines.** Narrowing losses and utilization rates tell you more than contract rumors. - **If you source AI hardware, evaluate second-source options early.** A functioning Samsung leading-edge line changes negotiation leverage with every supplier. - **If you study industrial policy, track how the Taylor campus performs against its US CHIPS Act support.** It is the largest test case for whether America can host leading-edge memory-logic integration. ## Key takeaways - Samsung signed a 22.8 trillion won (≈ US$16.5 billion) deal with Tesla in July 2025 to make the AI6 inference chip at Taylor, Texas, running through 2033. - Taylor Fab 1 was upgraded to a 2nm-class line with customer production targeted for 2027; Fab 2, breaking ground in November 2026, will make HBM base dies for AMD and Nvidia with mass production planned for 2030. - Samsung remains a distant second to TSMC, which held 67% of foundry revenue in late 2024, and the 2nm yield gap remains the central open question. - The recovery case rests on execution: delivered automotive-grade silicon in 2027 and confirmed HBM base die production by 2030. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-chips-samsung-foundry-comeback Free to quote with attribution and a link to the original.