Samsung vs TSMC at 2nm: the data

For the first time in a decade, the gap between TSMC and Samsung at the leading edge is close enough to measure rather than estimate.

Both companies are producing 2nm chips. I spent a weekend pulling every data point I could find from published papers, earnings call transcripts, and industry analyses. Here’s what the numbers say.

Yield

TSMC: approximately 85% yield on 2nm production wafers. That’s exceptional for a new node. It means 85 out of every 100 dies on a wafer are functional. The revenue per wafer is high. The waste is low.

Samsung: approximately 68%. Respectable. Improving monthly. But 17 percentage points behind TSMC. On a wafer with 500 dies, that’s 85 fewer functional chips. At scale, that’s billions of dollars of difference.

Yield is where TSMC’s institutional advantage shows most clearly. Decades of process engineering, contamination control, and yield optimization that can’t be replicated quickly.

Power efficiency

This is where it gets interesting. Samsung’s GAA (Gate-All-Around) nanosheet implementation shows better power efficiency in specific workloads, particularly AI inference at low clock speeds. The power savings are 5-10% in these scenarios.

TSMC’s implementation is more balanced. Better at high-frequency workloads (phones, PCs). Comparable at medium frequencies. Slightly behind Samsung at the low-power end.

If you’re building a phone chip, TSMC is still the choice. If you’re building an AI inference chip that runs at lower frequencies but massive parallelism, Samsung’s 2nm has a real argument.

Transistor density

TSMC: approximately 300 million transistors per square millimeter. Samsung: approximately 280 million.

Close. Closer than at 3nm, where TSMC led by a larger margin. Samsung’s nanosheet architecture is scaling well.

What this means

The race matters because it determines who builds the chips that run the AI models, the phones, the cars, the robots, the data centers. Every percentage point of yield. Every watt of power efficiency. Every million transistors per square millimeter.

Two years ago, the answer to “who makes the best chips?” was unambiguously TSMC. Today, the answer is “TSMC, but Samsung is closer than they’ve been in a decade, and the gap depends on which metric you prioritize.”

That competition is good. For prices. For innovation. For the global supply chain that can’t afford to have a single point of failure at 2nm.

ASML supplies both companies with the same lithography machines. SemiAnalysis and AnandTech have deeper technical breakdowns if you want the full picture.

I’m watching the yields. Yield is where races are won or lost at the leading edge. And right now, TSMC is winning. But Samsung is closing.

Related thinking: