Everyone Is Announcing Capacity. Nobody Is Announcing The Number That Decides It.

A wafer start is not a chip that you can sell, and that gap is where this week's story lives. TSMC is pushing three-nanometre toward a hundred and eighty thousand wafers a month, months ahead of the plan everyone published. Five two-nanometre fabs are ramping, and advanced packaging is booked more than a year out. Between a wafer start and a sellable die sits the yield ramp, and the published work puts that at twelve to twenty-four months. Each cycle of learning inside it is measured in weeks, not in press releases. Capital buys the building and it buys the tools, but it does not buy cycles of learning. Look at what TSMC actually published about two nanometre: that its defect density is lower than three, five and seven at the same stage. That is a relative claim, and the percentages you have read — sixty-five, seventy, eighty — are third-party numbers that disagree. Tip of the Week, and it costs you about five minutes. Take any plan somebody sent you this week — a deck, a schedule, a project mail — and open Microsoft 365 Copilot Chat in work mode so everything stays inside your tenant. Ask it to pull out every date commitment and label each one as either the date work starts or the date output becomes usable. Then ask which of your commitments have no usable date at all, because that last list is where your surprises are hiding. It is the same line a factory draws between starting a job and calling the output qualified, and the plan that carries only a start date is the plan that will surprise you. Capacity is the announcement, and yield is the answer. See you next week.