Tesla’s Dual-Foundry Strategy Shakes Semiconductor Industry
In late October 2025, Tesla officially confirmed its latest AI5 self-driving chip will adopt a dual-foundry production strategy, with manufacturing shared between Samsung Electronics and TSMC. Production facilities will be located at Samsung’s new fab in Taylor, Texas, and TSMC’s facility in Arizona. This decision marks Tesla’s more aggressive risk diversification approach in semiconductor supply chain management, while also bringing a major breakthrough for Samsung’s foundry business that has long operated under TSMC’s shadow. This collaboration continues the previous AI6 chip production model, demonstrating Tesla’s long-term commitment to a dual-supplier strategy.
Dual-Foundry Collaboration Details
Production Allocation and Capacity Planning
Samsung Texas Taylor Fab: Samsung’s new wafer fab in Taylor, Texas represents a total investment exceeding $17 billion, expected to enter mass production between late 2025 and early 2026. The facility will handle approximately 40-50% of AI5 chip capacity, with a monthly production target of 5,000-10,000 12-inch wafers.
TSMC Arizona Fab: TSMC’s first Arizona wafer fab began mass production in late 2024, with the second fab expected to come online in 2026. TSMC will handle the remaining 50-60% of AI5 chip capacity, targeting monthly production of 10,000-15,000 12-inch wafers.
Total Capacity Estimate: Combining both foundries’ capacity, Tesla’s AI5 chip is projected to produce 15,000-25,000 wafers monthly, equivalent to supplying AI5 chips for 2-3 million self-driving vehicles annually.
Process Technology and Specifications
2nm-Class Process: The AI5 chip is expected to utilize 2nm-class advanced process technology. Samsung will use its 2nm process with GAA (Gate-All-Around) technology, while TSMC will use its N2 process (also featuring GAA architecture). Both foundries’ process technologies have distinct characteristics but comparable performance.
Chip Specifications Forecast:
- Computing Performance: Expected to reach 500-800 TOPS (trillion operations per second), representing a 2-3x improvement over the AI4 chip (approximately 250 TOPS)
- Process Node: 2nm GAA architecture, with transistor density 20-30% higher than 3nm
- Power Consumption: Target TDP (thermal design power) of approximately 80-100W, slightly higher than AI4’s 72W but with significantly improved performance
- Neural Network Engine: Integrates more dedicated AI acceleration units, optimized for Transformer architecture inference performance
Design Compatibility: Tesla’s design team must ensure the AI5 chip design can be manufactured at both Samsung and TSMC foundries, requiring detailed adjustments at the PDK (Process Design Kit) level to ensure consistency in yield and performance.
Timeline Planning
2025 Q4-2026 Q1:
- Samsung Taylor fab completes production line verification, begins small-batch trial production
- TSMC Arizona second fab comes online
- Both fabs simultaneously produce AI5 chip engineering samples
2026 Q2-Q3:
- Mass production ramp-up phase, yield targets increase from initial 60-70% to 80-85%
- Tesla integrates AI5 chip into new Model Y and Cybertruck full self-driving systems
2026 Q4 and Beyond:
- Full mass production, monthly capacity reaches target scale
- AI5 chip becomes standard across Tesla’s entire vehicle lineup (high-end Full Self-Driving versions)
Strategic Significance Analysis
Supply Chain Risk Diversification
Single Supplier Risk: Over the past decade, TSMC has dominated the high-end chip foundry market with advanced process technology, commanding over 60% market share. However, TSMC’s capacity is highly concentrated in mainland Taiwan, facing potential threats from geopolitical risks (Taiwan Strait tensions) and natural disasters (earthquakes, typhoons).
Dual-Supplier Advantages:
- Capacity Assurance: If one foundry encounters problems (equipment failure, yield decline, political factors), the other can supplement capacity
- Bargaining Power: Tesla gains greater negotiating leverage, obtaining better terms on pricing, delivery schedules, and technical support
- Technology Hedging: Samsung and TSMC each have technical advantages, allowing Tesla to benefit from their competition
Historical Lessons: During the COVID-19 pandemic (2020-2022), the global automotive industry reduced production by millions of vehicles due to chip shortages, with total losses exceeding $200 billion. While Tesla was less affected (due to in-house chip development and rapid order switching capability), it still recognized the importance of supply chain resilience.
US Domestic Manufacturing Strategy
CHIPS Act Subsidies: The CHIPS and Science Act provides $52 billion in subsidies to encourage semiconductor companies to build facilities on US soil. Both Samsung and TSMC’s US fabs received billions in subsidies, reducing construction costs.
Political Alignment: Tesla CEO Elon Musk maintains close relationships with the US government, and choosing to produce AI5 chips domestically aligns with political expectations, helping secure policy support (such as relaxed self-driving regulations and tax incentives).
National Security Considerations: The US government views self-driving technology as a strategic industry, requiring critical chips to be produced domestically to avoid supply chain control by foreign powers. Producing AI5 chips entirely in the US satisfies this requirement.
Cost Challenges: US domestic manufacturing costs are approximately 20-30% higher than Asia (primarily due to labor, utilities, and operating costs), but through CHIPS Act subsidies and economies of scale, the cost gap can be reduced to within 10% long-term.
Competitive Pressure and Technological Progress
Samsung vs TSMC: Tesla’s simultaneous use of both foundries creates direct competitive pressure. Both will compete intensively on yield, delivery schedules, costs, and technical support, with Tesla as the primary beneficiary.
2nm Process Race:
- TSMC N2: Expected to enter mass production in H2 2025, with yield targets of 85-90% and density around 140-150 MTr/mm²
- Samsung 2nm GAA: Expected to enter mass production in late 2025, but yield has historically been a weakness, potentially only 70-75% initially, with density around 120-130 MTr/mm²
Yield Difference Impact: If Samsung’s yield is significantly lower than TSMC’s, Tesla may adjust capacity allocation (70% TSMC, 30% Samsung) until Samsung improves. This creates enormous pressure on Samsung to aggressively improve yields.
Strategic Breakthrough for Samsung
Long-term Challenges in Foundry Business
Market Share Gap: In the 2024 global wafer foundry market, TSMC held approximately 62% market share, while Samsung held only 13%, a significant disparity. Samsung’s market share in advanced processes (7nm and below) is even lower, at approximately 5-8%.
Customer Trust Issues: Samsung has repeatedly faced customer concerns about yield and stability in advanced processes. Qualcomm’s Snapdragon 8 Gen 1 (4nm) experienced yield issues with Samsung’s foundry, resulting in subsequent orders shifting to TSMC. Major customers like NVIDIA and AMD have also long preferred TSMC.
Technology Lag Perception: While Samsung leads globally in memory (DRAM, NAND Flash), its logic chip foundry business has struggled with a perception of being “technologically inferior to TSMC,” making it difficult to attract top-tier customers.
Significance of Tesla Order
Flagship Customer Endorsement: Tesla is one of the world’s highest-valued automotive companies (2025 market cap approximately $800 billion), and its order represents powerful endorsement for Samsung’s foundry business, proving “Samsung can produce world-class AI chips.”
Technology Validation: The AI5 chip uses 2nm GAA process, serving as a litmus test for Samsung’s advanced process technology. Successful mass production will break the “Samsung yield unreliable” stereotype and attract more customers.
Capacity Utilization: Samsung’s Taylor fab represents a $17 billion investment, making capacity utilization critical. Tesla’s order may account for 30-50% of the fab’s initial capacity, ensuring return on investment.
Chain Reaction: If AI5 chip mass production proceeds smoothly, other automakers (such as Ford, GM, BMW) may follow suit, commissioning Samsung to produce self-driving chips, creating a virtuous cycle.
Continuation of AI6 Chip Collaboration
AI6 Chip Precedent: In 2024, Tesla signed a $16.5 billion contract with Samsung for foundry production of the next-generation AI6 chip (expected to enter mass production in 2027). Consecutive AI5 and AI6 orders demonstrate Tesla’s long-term confidence in Samsung.
Technology Roadmap Alignment: Samsung and Tesla are establishing a long-term partnership, allowing both parties to jointly plan technology roadmaps for the next 3-5 years, including development and validation of 2nm, 1.4nm, and 1nm processes.
Impact on TSMC
Market Share Pressure
Non-Exclusive Order: In the past, TSMC often secured exclusive customer orders (such as Apple’s A-series and M-series chips), but Tesla’s AI5 uses a dual-supplier model, with TSMC receiving only 50-60% of capacity, compressing its market share.
Intensified Competition: Samsung’s re-entry into the high-end market through Tesla’s order may lead to capturing more TSMC customers in the future. TSMC needs to more aggressively improve technology, reduce prices, and enhance services to maintain its lead.
US Capacity Validation
Arizona Fab Operations: TSMC’s US fab represents its first large-scale overseas advanced process production, facing challenges in operational efficiency, yield control, and talent development. Tesla’s AI5 order is an important opportunity to test the US fab’s capabilities.
Cost Structure Adjustment: US fab production costs are higher than Taiwan fabs, requiring TSMC to reduce costs through automation and process optimization, or profit margins will be compressed.
Global Semiconductor Supply Chain Reorganization
Geopolitical Drivers
US-China Tech War: The US has implemented semiconductor export controls on China, restricting exports of advanced process equipment and technology. China is actively supporting its domestic semiconductor industry (such as SMIC), but technology still lags TSMC and Samsung by 2-3 generations.
Supply Chain Localization: Governments worldwide are encouraging semiconductor manufacturing localization to reduce dependence on single regions. The US, EU, Japan, and India all provide substantial subsidies to attract semiconductor manufacturers.
Taiwan Strait Risk Awareness: Business awareness of Taiwan Strait conflict risks has increased, with more companies (such as Apple, Qualcomm, AMD) requesting foundries to add capacity outside Taiwan.
Industry Landscape Changes
Three-Way Competition: The future wafer foundry market may form a three-way competition between TSMC, Samsung, and Intel. Intel, through its IDM 2.0 strategy, is opening foundry services to external customers, with Intel 18A process (approximately 1.8nm) expected to enter mass production in 2026.
Pressure on Smaller Players: Second-tier foundries like GlobalFoundries, UMC, and SMIC struggle to keep pace with advanced processes, focusing instead on mature processes (28nm and above) and specialized applications (automotive, industrial).
Equipment Suppliers Benefit: Equipment suppliers like ASML (extreme ultraviolet lithography), Applied Materials (deposition, etching), and Tokyo Electron (coating, developing) benefit from the global wafer fab expansion wave, with full order books.
Impact on Taiwan’s Industry
TSMC Order Dispersion
Revenue Share Decline: While Tesla is not TSMC’s largest customer (Apple, NVIDIA, and AMD have higher shares), the AI5 order being shared with Samsung means TSMC’s dominant position in the automotive chip market is weakening.
Automotive Market Competition: Automotive chips represent one of the fastest-growing semiconductor segments, with projected CAGR of 12% from 2025-2030. TSMC needs to aggressively pursue other automaker orders (such as Mercedes-Benz, Volkswagen self-driving chips) to compensate.
Supply Chain Opportunities
Design Services: Taiwan IC design service companies (such as GUC, Faraday) may participate in AI5 chip design verification and IP integration work, earning design service fees.
Packaging and Testing: AI5 chips may use advanced packaging technologies like CoWoS (chip-on-wafer-on-substrate) or InFO (integrated fan-out), potentially bringing packaging orders to ASE Technology and SPIL.
Material Supply: Taiwan material suppliers (such as Chang Chun Petrochemical’s photoresist, Episil’s quartz components, Chemtron’s chemicals) supply both Samsung and TSMC foundries, benefiting from double orders.
Talent Flow Pressure
Overseas Recruitment: Samsung and TSMC’s US fabs require large numbers of process engineers and equipment engineers, potentially poaching from Taiwan with high salaries. Taiwan’s semiconductor industry faces talent outflow pressure.
Rising Salary Costs: To retain talent, TSMC and other Taiwan semiconductor companies need to increase salaries and benefits, raising operating costs.
Technical Challenges and Risks
Yield Control
2nm Process Difficulty: The 2nm GAA process represents one of humanity’s most advanced mass-production technologies, with extremely difficult yield control. Minor process variations, impurity contamination, or equipment parameter deviations can cause chip failures.
Dual-Fab Consistency: Tesla needs to ensure AI5 chips produced at Samsung and TSMC have similar performance, power consumption, and yields, or separate software/firmware tuning will be required, increasing complexity.
Ramp-up Schedule Risk: If either foundry’s yield ramp-up is slower than expected, it could delay AI5 chip mass production schedule, affecting Tesla’s new vehicle launch plans.
Cost Pressure
US Manufacturing Premium: US domestic manufacturing costs are high; even with CHIPS Act subsidies, long-term costs will still be 10-15% higher than Asian fabs. Tesla must evaluate whether to absorb the extra cost or compress margins.
Economies of Scale: AI5 chip annual demand of approximately 2-3 million units (assuming one per vehicle) is relatively small compared to smartphone chips (1.5 billion units annually), making it difficult to fully leverage economies of scale, resulting in higher costs.
Design Complexity
Multi-Supplier Management: Tesla must simultaneously manage Samsung and TSMC, coordinating design changes, process adjustments, and capacity allocation, significantly increasing management complexity.
Quality Consistency: Although the design is identical, minor process differences between the two foundries may result in slightly different chip characteristics, requiring Tesla to establish strict quality control and testing procedures.
Self-Driving Technology Competition
AI5 Chip Performance Significance
Computing Performance Leap: The AI5 chip is expected to reach 500-800 TOPS, far exceeding current competitors. NVIDIA DRIVE Orin offers approximately 254 TOPS, Qualcomm Snapdragon Ride approximately 700 TOPS, and Intel Mobileye EyeQ6 approximately 300 TOPS.
Full Self-Driving Capability: Greater computing power supports more complex neural network models, processing more sensor data (cameras, radar, ultrasound), improving self-driving decision accuracy and safety, advancing toward L4/L5 fully autonomous driving.
Real-Time Processing: AI5 can process 4K image streams from 12 cameras in real-time, performing semantic segmentation, object detection, and trajectory prediction with latency below 50 milliseconds, meeting high-speed driving safety requirements.
Competitor Dynamics
Waymo: Google’s Waymo uses in-house TPUs (Tensor Processing Units) for self-driving computation, already providing robotaxi services in San Francisco, Phoenix and other cities, with leading technology but high costs.
Cruise (General Motors): Cruise uses multiple NVIDIA Orin chips, with computing power around 1,000 TOPS, but suspended operations in 2024 due to accidents, adopting a more conservative strategy when restarting in 2025.
Chinese Manufacturers: Huawei, Horizon Robotics, and Baidu offer self-driving chips with computing power around 200-400 TOPS, still behind Tesla technologically but progressing rapidly.
Business and Financial Impact
Tesla Cost Structure
Chip Cost Ratio: The AI5 chip is estimated to cost $300-500 per unit (2nm advanced process), with one chip per vehicle, representing approximately 0.5-1% of total vehicle cost (assuming $50,000 vehicle price). While not a major cost component, it impacts profit margins.
In-House vs. External Procurement: Tesla’s in-house chip development avoids paying expensive NVIDIA GPU fees (a single H100 costs approximately $30,000). If procured externally, each vehicle would require 2-4 GPUs, costing thousands of dollars, making it completely impractical.
Mass Production Scale Benefits: As production increases to over 3 million units annually, chip unit price can decrease to $200-300, further improving cost structure.
Foundry Revenue Contribution
Samsung: Assuming Samsung handles 40% of AI5 chip capacity, producing 1.2 million units annually at $400 each, annual revenue would be approximately $480 million. Combined with future AI6 chip orders, Tesla may become one of Samsung’s top ten foundry customers.
TSMC: TSMC handles 60% of capacity, producing 1.8 million units, with annual revenue around $720 million. For TSMC, this represents a small percentage (2024 TSMC revenue approximately $75 billion), but has strategic significance (entry point to automotive market).
Future Outlook
AI6 and More Advanced Processes
AI6 Chip Planning: AI6 is expected to enter mass production in 2027, using 1.4nm or more advanced processes, with computing power target of 1,000-1,500 TOPS, supporting fully autonomous driving (L5) and in-vehicle AI assistant functions.
Process Evolution Roadmap:
- 2025-2026: 2nm GAA (AI5)
- 2027-2028: 1.4nm or A14 (AI6)
- 2029-2030: 1nm (AI7), potentially using CFET (Complementary FET) or GAA stacking technology
Physical Limit Challenges: Processes below 1nm approach the physical limits of silicon-based semiconductors, with quantum effects, leakage current, and process variation becoming more severe. The industry may need to transition to new materials (such as carbon nanotubes, graphene) or new architectures (such as 3D stacking, Chiplets).
Capacity Expansion Plans
Samsung: If AI5 chip mass production proceeds smoothly, Samsung may add second-phase capacity at the Texas Taylor fab or establish dedicated automotive chip production lines at the Pyeongtaek fab in Korea.
TSMC: TSMC’s third Arizona fab (2nm) is expected to come online in 2027-2028, with some capacity potentially allocated to Tesla’s AI6 chip.
Other Automakers Following Suit
Ford, GM: US traditional automakers may emulate Tesla’s dual-foundry strategy if AI5 succeeds, commissioning Samsung and TSMC to produce self-driving chips, reducing dependence on single suppliers.
European Automakers: Mercedes-Benz, BMW, Volkswagen and other European automakers currently partner with NVIDIA, Qualcomm, and Mobileye, but may transition to in-house chip + dual-foundry models in the future.
Chinese Automakers: BYD, NIO, XPeng and other Chinese automakers are restricted by US export controls from using advanced processes. They may commission SMIC to produce self-driving chips using 7nm or more mature processes, with lower performance but reduced costs.
Industry Insights
Supply Chain Resilience Importance
Single Supplier Risk: The COVID-19 pandemic and geopolitical tensions revealed single supplier risks. Companies need to establish diverse supply chains; even with slightly increased costs, it’s worthwhile to ensure capacity.
Strategic Resource Localization: Strategic resources like semiconductors, batteries, and critical raw materials see governments encouraging local production to reduce dependence on specific countries.
Technology Competition Normalization
TSMC Not Irreplaceable: While TSMC leads technologically, Samsung and Intel continue pursuing, with the technology gap narrowed to 1-2 years. Customers have more choices, requiring TSMC to continuously innovate to maintain leadership.
Yield is Key: Advanced process competition ultimately depends on yield. Whoever can more quickly raise yields to 85-90% will win orders.
In-House Chip Trend
Tech Giants Going In-House: Apple (M-series, A-series), Google (Tensor, TPU), Amazon (Graviton, Trainium), Meta (MTIA), and Microsoft (Maia) are all developing in-house chips, reducing dependence on Intel and NVIDIA.
Automotive Industry Following: Tesla leads the automotive industry’s in-house chip trend; in the future, more automakers will invest in in-house development, commissioning foundries for production, forming a “design-manufacturing separation” model (similar to the smartphone industry).
Conclusion
Tesla’s AI5 chip adopting a dual-foundry production strategy with Samsung and TSMC reflects the current semiconductor industry’s supply chain reorganization mega-trend: heightened geopolitical risk awareness, increased localized production demands, and diversified suppliers reducing risks. For Samsung, this represents a major breakthrough in returning to the high-end foundry market; if AI5 mass production succeeds, it will attract more customers and narrow the gap with TSMC. For TSMC, while market share is compressed, it maintains technology and yield advantages, requiring more aggressive innovation to maintain leadership. For the global semiconductor industry, Samsung and TSMC’s competition in 2nm processes will accelerate technological progress, driving continued AI computing performance improvements. In coming years, as AI5 and AI6 chips enter mass production, Tesla’s full self-driving technology will advance significantly, deepening the convergence of automotive and semiconductor industries, ushering in a new era of intelligent transportation.
