The negotiations have been running quietly for months, and on Wednesday the details began surfacing across Asia. BYD, Google, AMD, and Tesla have all been in discussions with Samsung Electronics about manufacturing chips at its foundries, according to people familiar with the talks, as the AI boom strains the production capacity of the industry’s dominant player, TSMC.
The interest is a direct consequence of the AI buildout. TSMC produces more than 90 percent of the world’s most advanced semiconductors, and its capacity is booked solid, with Nvidia, AMD, Google, and others competing for the same advanced nodes and packaging slots. Customers that need capacity in a hurry have begun looking for a second source, and Samsung, despite its yield problems in recent years, is the only realistic alternative at scale.
The numbers show the momentum. Samsung’s foundry and system LSI business generated 6.9 trillion won, roughly $5 billion, in first-quarter revenue, up 15 percent from a year earlier, according to the company’s earnings report. The overall company had a record quarter — 133.9 trillion won in revenue and 57.2 trillion won in operating profit, an eightfold jump from a year earlier, driven by memory chips whose prices have surged on AI demand. Foundry, by contrast, has been the weak spot, losing money for four consecutive years. The new customer discussions have revived expectations that the division could return to profitability this year.
The most notable win is Neuralink. Samsung has taken on development of the fourth-generation brain implant chip for Elon Musk’s neurotechnology company, according to people familiar with the project, marking the first time Samsung’s foundry has worked on a Neuralink chip. The chip, code-named O1, is being developed on Samsung’s 4-nanometer process, with test chips expected in the first half of next year and mass production possible by late next year. Neuralink has relied on TSMC for its earlier chips, and the move to Samsung reflects its desire to diversify suppliers amid the capacity crunch.
The Tesla relationship is the foundation. Samsung already manufactures Tesla’s AI5 autonomous driving chip and has secured the follow-on AI6 chip, an order valued at roughly 23 trillion won, or about $15 billion, according to people familiar with the arrangement. Tesla’s decision to split its chip production between TSMC and Samsung, which Musk has publicly praised as a way to combine the strengths of both companies, gave Samsung a reference customer that other potential clients take seriously.
Google is evaluating Samsung for components of its next-generation AI hardware. The company is considering Samsung for input-output chips for its 10th-generation Tensor Processing Unit, code-named Icefish, according to reports, and has discussed Samsung’s participation in manufacturing its Axion server processors around 2028. TSMC is still expected to make the main compute die for the TPU, but Samsung could take supporting chips and advanced packaging work. AMD, meanwhile, has held talks with Samsung about producing certain CPUs beginning in 2028, according to people familiar with the discussions.
The list of prospective customers extends to China. BYD, the electric vehicle maker, has been in talks with Samsung about manufacturing autonomous driving system-on-chips using its 2-nanometer and 4-nanometer processes, according to people familiar with the matter. Chinese automakers have limited access to TSMC’s most advanced capacity, both because of the shortage and because of export restrictions, making Samsung an attractive alternative for automotive chips.
The strategy carries real risk. Samsung still trails TSMC in manufacturing yields, the percentage of usable chips produced per wafer, and its customers know it. The deals under discussion are structured accordingly: Samsung is taking on supporting roles, older nodes, and automotive chips rather than the most advanced compute dies, which continue to go to TSMC. The company’s 2-nanometer process, its best hope for closing the technology gap, is still ramping, and analysts say its yield performance over the next year will determine whether Samsung becomes a genuine second source for frontier chips or remains a overflow option for capacity-constrained customers.
The shift has geopolitical dimensions as well. Advanced chipmaking capacity is concentrated in Taiwan, and the AI boom has made that concentration a policy concern in Washington, Seoul, and Tokyo alike. Samsung’s foundry operations span South Korea and a plant in Taylor, Texas, giving customers a non-Taiwanese source for advanced chips — a point Samsung executives have pressed in meetings with U.S. officials courting semiconductor investment. Governments have responded with subsidies, and Samsung has been among the largest recipients of U.S. CHIPS Act funding.
For the broader industry, Samsung’s gains mark a structural shift. The AI boom has made advanced chip capacity a strategic resource, and governments and companies alike are seeking redundancy. Samsung’s foundry, with capacity in South Korea and the United States, offers a hedge against the concentration of leading-edge production in Taiwan. The company has said it plans to reach full utilization of its advanced-node lines in the second quarter and is expanding 2-nanometer customers. Whether the new orders translate into profit will depend on yields — the same variable that has humbled Samsung’s foundry ambitions before.
