TSMC executive calls AI a ‘three-year-old superman’ as chipmaker takes cautious internal stance

The world’s most important chipmaker is pumping the brakes on its own internal AI adoption, even as it manufactures the chips that make AI possible for everyone else.

At a career forum held at National Taiwan University, TSMC Co-COO Y.J. Mii compared artificial intelligence to a three-year-old superman: extraordinarily powerful, and almost entirely without the judgment to use that power wisely.

TSMC is the foundry that produces chips for Nvidia, Apple, and virtually every other company betting its future on AI.

The judgment problem

Mii’s concern centers on the gap between capability and wisdom. AI can write code. It can accelerate electronic design automation, the complex software-driven process of laying out semiconductor circuits. Those are real, measurable wins.

Where AI falls short, in Mii’s view, is in the high-stakes decisions that define next-generation chip development. Semiconductor process engineering at the leading edge involves tradeoffs that are subtle, costly, and sometimes irreversible.

The data leak risk is particularly pointed for a company like TSMC. Its manufacturing process knowledge, the specific parameters and recipes that let it produce chips no one else can reliably make at scale, is its core competitive asset. Feeding sensitive R&D data into AI tools that route information through external systems is not a theoretical risk.

Mii’s framing aligns with a broader moment in the industry. Leaders at OpenAI and Anthropic have both called for measured approaches to AI development, citing concerns about systems behaving in ways their designers did not anticipate, including cases where AI models have initiated unexpected interactions with external platforms.

What TSMC is actually doing

TSMC is in the middle of a significant manufacturing expansion, driven precisely by the surging demand for AI-capable chips. The client list includes Nvidia’s GPUs, Apple’s custom silicon, and a roster of hyperscalers building their own accelerators.

The company’s caution is internal and specific. AI tools in areas like coding assistance and EDA workflows are apparently fair game. What Mii is guarding against is the premature deployment of AI in the research and process development work that defines TSMC’s technological edge.

What this signals for the broader market

For investors and analysts watching the semiconductor space, Mii’s comments carry weight precisely because TSMC sits at the center of the AI supply chain. The rapid manufacturing expansion TSMC is pursuing suggests the demand side of the AI story remains robust, but the cautious internal posture suggests that productivity gains AI boosters promise for R&D may take longer to materialize in environments where the cost of error is high and proprietary data is involved.

The three-year-old superman metaphor is a useful one to hold onto. The question TSMC’s leadership seems to be asking is not whether AI will eventually be ready for high-stakes semiconductor work, but whether it is ready right now. Based on Mii’s remarks, the answer is a considered no, and the company is acting accordingly.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.