Why Google, Tesla and AMD are turning to Samsung for AI chips

Key Concepts

• Foundry: A semiconductor fabrication plant that manufactures chips designed by other companies (e.g., TSMC, Samsung, Intel).
• Multi-Foundry Model: A strategy where a chip designer uses multiple manufacturing partners to diversify supply chains and reduce risk.
• Yield: The percentage of functional chips produced on a wafer; a critical metric for manufacturing efficiency and profitability.
• Advanced Chip Packaging: Specialized techniques to connect and house multiple chips, essential for high-performance AI hardware.
• Tape-out: The final stage of the chip design process before it is sent for manufacturing.
• Process Node: A specific generation of semiconductor manufacturing technology (e.g., 3nm), where smaller numbers generally indicate higher performance and efficiency.

---

1. Main Topics and Market Dynamics

The global semiconductor industry is experiencing a massive "AI infrastructure bailout," leading to severe supply chain constraints. As demand for AI-capable processors outpaces supply, major tech companies are moving away from relying solely on TSMC.

• Market Leaders: TSMC currently dominates with ~70% market share, followed by Samsung (~10%) and SMIC (~5-6%).
• The Shift: Companies like Google, AMD, Tesla, and BYD are actively seeking Samsung as a secondary or tertiary foundry partner to secure capacity for next-generation AI chips, autonomous driving processors, and CPUs.
• Capital Intensity: The cost of entry is astronomical. TSMC alone projected $56 billion in capital expenditure for the current year, following $40 billion the previous year.

2. Real-World Applications and Clients

• Google: Considering Samsung for next-generation ARM-based CPUs and Tensor Processing Units (TPUs) for AI workloads.
• Tesla: Targeting Samsung for "AI 5" and "AI 6" chips to power future vehicles and robotics.
• BYD: Seeking advanced chip manufacturing for autonomous driving systems.
• Rapidus (Japan): An emerging player currently in the pilot stage, already securing clients like Fujitsu and Tenstorrent.

3. Strategic Framework: The Multi-Foundry Model

The transition to a multi-foundry model is driven by three primary factors:

1. Supply Chain Resilience: Avoiding the "all eggs in one basket" risk, particularly regarding geopolitical tensions surrounding Taiwan.
2. Capacity Constraints: TSMC’s facilities are operating at full capacity, forcing designers to look elsewhere to meet production timelines.
3. Geopolitical Diversification: US tech firms are prioritizing manufacturing options in multiple regions (e.g., Arizona, Japan, South Korea) to mitigate regional instability.

Challenges of the Model:

• Design Complexity: Even if two foundries offer the same process node (e.g., 3nm), their design rules, materials, and manufacturing technologies differ. Designers must create separate sets of masks and modify designs for each foundry, significantly increasing costs.

4. Key Arguments and Perspectives

• The "Virtuous Cycle" of Yields: While Samsung currently lags behind TSMC in yield, the influx of new, high-volume customers provides the necessary engineering feedback loop to improve manufacturing performance. As Annie Chen notes, "The more customers and products a foundry has... the faster it can identify issues."
• The "Neutrality" Advantage: TSMC’s primary competitive advantage is its pure-play foundry model—it does not compete with its customers. In contrast, Samsung and Intel design their own chips, which can create conflicts of interest with their foundry clients.
• Market Outlook: The industry is not a zero-sum game. Because the AI boom is expanding the total addressable market so rapidly, all major foundries (TSMC, Samsung, Intel) are positioned to win, provided they can execute on technology and yield.

5. Notable Quotes

• "The entire supply chain is running at full capacity and demand continues to outpace supply, that's when customers become especially eager to develop second and even third sources." — Annie Chen
• "Running leading-edge chip factory is extraordinarily difficult, involving thousands of process steps and the lead time up to 5-6 months for one chip." — Annie Chen

6. Synthesis and Conclusion

The semiconductor industry is undergoing a structural shift from a TSMC-centric model to a diversified multi-foundry landscape. While this shift is primarily a response to the AI-driven capacity crunch and geopolitical risk, it introduces significant cost and complexity for chip designers. Samsung is well-positioned to capture this demand, provided it can improve its yields through increased customer collaboration. Ultimately, the "winner" will be the company that best balances technical execution, production capacity, and customer trust in an increasingly expensive and high-stakes manufacturing environment.