A Digitimes analysis published this week argues TSMC's grip on leading-edge production gives CEO C.C. Wei unusual sway — from pricing on the 3 nm and 2 nm nodes to the supply terms he can offer Elon Musk's emerging Terafab project.

The piece, framed as a leader-strategy brief rather than a product report, runs through the capacity levers TSMC has accumulated under Wei's tenure since he took the CEO role in 2018. The headline finding: tight advanced-node utilization plus a customer roster built around hyperscalers and AI accelerators leaves the foundry with pricing power rivals cannot quickly match.

What is Terafab?

Terafab is Musk's stated intent to build a vertically integrated fab covering logic, memory and packaging under one roof. Musk has not published a process roadmap, a capex figure or a production date. The Digitimes piece treats Terafab as the demand question of the moment — whether it draws volume away from TSMC, sits alongside it, or pulls TSMC in as a partner.

What does Wei actually control?

Three knobs, according to the framing:

• Allocation on N3 and N2 wafers through 2026
• CoWoS (chip-on-wafer-on-substrate, the 2.5-D advanced packaging method) capacity, the bottleneck behind every AI accelerator shipped today
• Pricing on long-term supply agreements, including the 2025–2027 cohort

The analysis pairs that with a second-order point: TSMC's customer base is now concentrated among a small number of AI-era buyers. Concentration cuts both ways — it raises revenue risk, but it also raises switching costs for the buyers.

What does this change for Musk?

If Terafab is real, Wei holds the only proven 2 nm-class line in commercial volume today. Samsung Foundry's 2 nm GAA (gate-all-around, a transistor architecture that wraps the gate around the channel) process is pre-revenue; Intel Foundry Services has yet to ship external 18A (1.8 nm-class) wafers at scale. Musk can build a fab, but the equipment supply chain, PDK (process design kit, the rules file chip designers need to tape out a chip) libraries and yield learning curve for sub-3 nm logic still run through Hsinchu.

That, the brief argues, narrows Terafab to a few practical shapes: a partnership that buys TSMC wafers while Musk funds capacity, a packaging-only facility that depends on CoWoS, or a delayed program that lets TSMC set the timeline.

What it doesn't settle

The analysis does not name contract values, customer identities or yield figures. It also does not address whether Musk's recent comments about Terafab were operational or rhetorical. "The capacity picture is concrete. The Terafab picture is still a paragraph," one industry watcher quoted in adjacent coverage this week said — a sentiment the Digitimes piece endorses without restating.

The practical takeaway for foundries, OSAT (outsourced semiconductor assembly and test) shops and equipment vendors: Wei's 2025–2027 allocation calls are the single most-watched supply variable in advanced packaging, and Terafab is the wildcard that could redraw the table or vanish into it.