TSMC Capacity Crunch: Why Google Is in Talks With Samsung for AI Chip Production

TLDR

• MediaTek is shifting from chip design to full system-level AI hardware integration, targeting Google’s TPU and Elon Musk-affiliated companies’ AI rack projects.
• Analyst Ming-Chi Kuo says the move carries little impact on earnings over the next two years but positions MediaTek for long-term growth.
• Google is also in talks with Samsung to help make parts of its next AI chip, codenamed Icefish, due to tight capacity at TSMC.
• The Samsung talks reflect how strong demand for advanced AI chip manufacturing has become, with even top customers needing backup suppliers.
• MediaTek’s system-level push targets a 40–50% gross margin through an asset-light model, outsourcing manufacturing while leading design and validation.

MediaTek has expanded its artificial intelligence strategy beyond chip design, targeting full system-level hardware integration. The Taiwan-based chipmaker has set its sights on two early opportunities: the printed circuit board assembly for Google’s TPU and rack-level work for Elon Musk-affiliated companies building their own AI chips.

Alphabet Inc., GOOGL

This comes from Ming-Chi Kuo at TF International Securities, who says the move reflects a long-term repositioning rather than a quick earnings play.

MediaTek’s Two Targets: Google and Musk-Affiliated Companies

The two opportunities are not the same. Google already has a well-established hardware assembly ecosystem, so MediaTek’s chances of winning the top-tier rack integration work there are described as slim.

Its more realistic entry point with Google is at the board-level, starting with the TPU v10 chip known as Icefish.

The Musk-affiliated opportunity is different. Those companies are still building out their own AI silicon at scale, and the rack assembly supply chain around that hardware is not yet formed.

MediaTek plans to target a gross margin of 40–50% in this segment by leading design and validation while outsourcing manufacturing, keeping costs lean.

Google Turns to Samsung as TSMC Capacity Tightens

At the same time, Google is reportedly in talks with Samsung to manufacture a memory input-output die for the same Icefish chip. TSMC would still handle the main computing engine using its advanced 1.4-nanometer process.

Wedbush analysts said the Samsung talks likely reflect limited capacity at TSMC rather than a move away from the company. In short, demand for advanced AI chip manufacturing is so high that even major customers like Google may need to spread production across multiple foundries.

Using Samsung would not be without challenges. Splitting chip production across manufacturers adds complexity and could affect chip yields and costs.

For Google, the goal is securing enough supply to meet future AI workload demands. For Samsung, it represents a chance to win more advanced foundry contracts.

Kuo’s bigger concern is that MediaTek’s existing ASIC chip design business could slow within two to three years as the industry shifts to a new semiconductor model. That risk is part of why he views the system-level push as a necessary move, even if it adds little revenue near term.

The clearest milestone to watch is whether MediaTek wins any qualification work on the TPU v10 Icefish chip. On the Musk side, any concrete timeline remains to be seen.

The post TSMC Capacity Crunch: Why Google Is in Talks With Samsung for AI Chip Production appeared first on CoinCentral.