TSMC's leading-edge capacity is fully locked up by NVIDIA, Apple, and Broadcom, prompting a wave of customer conversations with Samsung Foundry. According to Nikkei Asia, AMD is in talks to manufacture future CPUs at Samsung from 2028. Google is evaluating Samsung for its next-gen Axion processors and supporting components for its 'Icefish' 10th-gen TPU co-developed with MediaTek. Tesla has already confirmed its AI6 chip will be produced at Samsung's Texas facility. BYD is discussing autonomous driving SoCs on Samsung's 2nm and 4nm nodes. Groq, the NVIDIA-backed LPU maker, already manufactures at Samsung and may extend the relationship to future products. Intel Foundry is also picking up overflow orders. Samsung still trails TSMC on yields, but available capacity at leading-edge nodes is driving these discussions.
Background: TSMC's Capacity Crunch
TSMC's leading-edge nodes - particularly 3nm and 2nm - are fully committed through 2028, according to Nikkei Asia. Priority goes to NVIDIA, Apple, and Broadcom, leaving limited room for other customers. A Chinese automotive chip executive told Nikkei that TSMC prioritizes advanced production because it is more profitable and in short supply, forcing customers to explore alternatives.
AMD: CPUs at Samsung from 2028 AMD is in discussions with Samsung to manufacture certain future CPUs beginning in 2028, per Nikkei. AMD currently relies on TSMC for GPU production; this would mark its first reported CPU shift to Samsung, targeting Samsung's 2nm process ahead of planned H2 2026 mass production.
Google: Axion and TPU Components Google is evaluating Samsung for two separate engagements, per Nikkei and Reuters. The first covers next-generation Axion data-center processors expected around 2028. The second involves components for Google's 10th-generation TPU, codenamed 'Icefish,' co-developed with MediaTek. TrendForce notes TSMC is still expected to manufacture the main Icefish compute die, while Samsung discussions center on supporting chips and advanced packaging - a split-foundry arrangement that allows supply diversification without fully departing from TSMC.
Tesla and Groq: Confirmed and Ongoing Orders Tesla has confirmed its next-generation AI6 chip will be produced at Samsung's Texas facility. Groq, the NVIDIA-backed language processing unit maker whose chips power LLM inference services, already manufactures LPUs at Samsung and may extend to future products per Nikkei. NVIDIA has not yet decided whether next-generation Groq chips stay at Samsung or move to TSMC.
BYD and Automotive AI Chips BYD, the world's largest electric vehicle maker, is in discussions with Samsung about manufacturing autonomous driving SoCs on 2nm and 4nm nodes, per Seoul Economic Daily. Samsung is also in preliminary talks with NVIDIA to produce the Drive AGX Thor autonomous driving chip.
Samsung's Position and Yield Challenges Samsung's 2nm node has reached approximately 50% yields ahead of planned H2 2026 mass production. TrendForce notes yield improvement and manufacturing stability remain critical to converting discussions into confirmed orders. Multi-foundry strategies carry significant engineering overhead - chip designs require tailoring to each process - making the approach realistic only for well-funded developers. Despite these hurdles, rising utilization and order inflows suggest Samsung could reach foundry profitability by Q3 2026.
Intel Foundry Also Benefits TSMC's capacity constraints are fragmenting what was near-monopoly control of leading-edge production. Intel Foundry is reportedly gaining Apple's M7 production on its 18A-P process in 2027 and has received a Google order for over 3 million TPUs for 2028, further diversifying a supply chain that AI infrastructure demand has pushed to its limits.
Scoring Rationale #
Nikkei-sourced report with wide industry pickup covering multiple confirmed and in-discussion customer shifts to Samsung Foundry for AI-critical chips including Google TPUs, Groq LPUs, and Tesla AI6 inference silicon - directly relevant to AI hardware supply chains. Scored 7.0 (notable infrastructure shift) rather than higher because most engagements remain discussions not confirmed production commitments, and Samsung still trails TSMC on yields.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.