Washington is turning a chip tariff into an industrial-policy contract. Read the allowance, not the rate.

The reported version of the story is a number: a 100% tariff, coming for laptops and the servers that run artificial intelligence. It is a good number. It is easy to repeat, it moves memory stocks, and it tells you almost nothing about what the policy would actually require of anyone. As of this week the proposal is a set of deliberations reported by several outlets, not a signed proclamation, and the useful questions are the ones the headline skips: who would be bound, what would the operative clause require them to do, and by when would it bite.

Read past the rate and a different instrument comes into view. What the Commerce Department is reported to favour is not a simple border tax on chips. It is a system that pairs a punitive headline rate with a duty-free import allowance — a volume of chips a company may bring in without paying the tariff — and pegs the size of that allowance to how much semiconductor manufacturing that same company commits to build inside the United States. That is not a tariff in the ordinary sense. It is a licensing regime wearing a tariff's clothes, and the load-bearing text is the formula for the allowance, not the percentage on the sign.

What the January order actually did

Start with the law that already exists, because the new proposal is an extension of it, not a fresh idea. In January the administration used Section 232 of the Trade Expansion Act of 1962 — the national-security authority that lets the president adjust imports of a good deemed critical to defence — to impose a 25% tariff on a narrow set of advanced logic semiconductors, effective mid-month. Section 232 matters here for a procedural reason people gloss over: it runs through presidential proclamation on the back of a Commerce finding, which means the scope, the rate and the carve-outs can be redrawn by the same mechanism that created them, without a new act of Congress.

The January order was narrow on purpose, and it came with exemptions that did most of the practical work: carve-outs that spared chips destined for data centres, for research and development, for startups, and for finished consumer devices. If you found the January tariff survivable, those exemptions are the reason. They are also the reason the current reporting should be read as more than a rate change. Officials are said to be weighing whether those carve-outs carry over at all. A tariff you were exempt from and a tariff you are no longer exempt from are, in practice, two different laws, even if the number on the page never moves.

The part that binds: an allowance, not a rate

Here is the mechanism as reported, stated plainly. A company would be allowed to import a certain volume of semiconductors free of the tariff. The size of that duty-free volume would be tied, proportionally, to the company's committed investment in United States chip-fabrication capacity. Build more domestic output, or finance it, and you earn the right to import more without paying. Build nothing, and your allowance shrinks toward zero, at which point the headline rate stops being a threat and becomes your cost of doing business.

Lawyers will recognise the shape of this, because it is not a tariff schedule. It is a quota with a price attached, and quotas behave differently from tariffs. A tariff sets a cost and lets the market decide how much to import at that cost. A quota-plus-allowance sets a quantity and makes the government the party that decides who gets how much of it. The discretion moves from the importer to the administrator. Whoever writes and administers the allowance formula — how investment is counted, over what horizon, with what verification, and how a dollar of committed capacity converts into a chip of duty-free import — is the person actually setting policy. The rate is the advertisement. The formula is the contract.

A tariff sets a price and lets the market decide the quantity. This sets the quantity and lets the government decide who gets it. That is not a smaller intervention. It is a larger one.

This distinction is not pedantry; it changes who has leverage over whom. Under an ordinary tariff, a firm's response is a spreadsheet exercise: absorb the cost, pass it on, or source elsewhere. Under an investment-linked allowance, the firm's response is a negotiation with the state about how much it will build, where, and how fast — with its ability to import the components it needs held as the counterparty's collateral. The tariff stops being a tax and becomes the enforcement arm of an industrial-policy bargain. That is a coherent thing for a government to want. It is simply a much more interventionist instrument than the word 'tariff' admits, and it should be described as what it is.

Who would be in scope

The expansion that generated the headlines is the move from the chip to the thing the chip is inside. The January measure reached advanced logic semiconductors. The proposal under discussion is reported to reach finished goods that contain them: data-centre servers, laptops, and games consoles among them. This is the part that turns a specialist trade action into a story about consumer prices and AI build-out budgets at once, because the same instrument would touch a graphics card in a hyperscaler's rack and a machine on a retail shelf.

Scope, in trade law, is a question of the importer of record — the entity that brings the good across the border and is liable for the duty — and of how the good is classified when it arrives. Tariff a bare processor and you bind the chip's importer. Tariff the server the processor sits in, and you bind the systems builder, the cloud operator, or the retailer that imports the finished box, and you invite a long argument about classification: how much of a machine has to be assembled abroad, and how much silicon it must contain, before it counts as the tariffed article rather than something adjacent to it. Those definitional fights are not a footnote. They are where months of the policy's real meaning will be decided, well after the rate is announced and the press has moved on.

The cost, wherever the duty formally lands, moves along the chain. A tariff paid by a server importer is a cost the cloud operator sees in its capital budget, which the AI company renting the capacity sees in its bill, which — eventually, partially — the customer sees. That is not an argument for or against the policy. It is a reminder that 'a tariff on servers' and 'a tariff on chips' are not interchangeable phrases: they bind different parties, at different points, with different room to pass the cost along.

When it would bite

This is the discipline the coverage most often skips: the difference between a proposal, a rule in force, and a rule enforced. What is reported now is deliberation — tiered rates and import caps that would vary by trading partner, a phased rollout, exemptions that may or may not survive. Those familiar with the talks caution that the plan could change substantially before anything is signed. A Section 232 action becomes law when the president issues the proclamation and it takes effect on the stated date; it becomes real for a given company when that company's exemption lapses or its allowance is set. Until then it is a strong signal and a live risk, which is worth pricing, but it is not yet an obligation, and it should not be reported as one.

The load-bearing date, when it comes, will not be the announcement. It will be the effective date of the proclamation, the expiry of the January carve-outs, and the first allowance year — the period against which a company's committed investment is measured to size its duty-free imports. Those are the deadlines that will actually govern behaviour. The number that leads the story rarely is.

The 100% is the headline. The formula is the law.

About that 100%: it is real as a stated intention and conditional as a policy. Administration officials have said repeatedly that chipmakers in South Korea and Taiwan which decline to invest in the United States could face tariffs at that level. Read in the context of the allowance mechanism, the 100% is not really a rate anyone expects to collect at scale. It is the price of refusing the bargain — the number that makes the duty-free allowance worth negotiating for. A punitive rate that almost no compliant company pays, paired with an exemption almost every company will restructure itself to earn, is a familiar design. The rate exists to make the exemption valuable. Reporting the rate without the allowance is reporting the stick without the deal it is meant to enforce.

The tiering is the other tell. Rates and caps that vary by trading partner, and by the size of that partner's leading manufacturers, are the marks of an instrument aimed at specific companies in specific countries — a handful of firms in Taiwan and South Korea whose fabrication capacity the policy is trying to relocate — rather than a neutral levy on a category of goods. That is a legitimate aim, pursued through an unusually precise tool. It is also, definitionally, discretionary: a levy calibrated to named firms is a levy someone administers case by case, and discretion administered case by case is where trade policy and industrial policy stop being distinguishable.

The Washington Effect

I spend most of my time tracing a pattern that runs the other way across the Atlantic — the one where a rule written in Brussels becomes the global default because no company builds a separate product for one market, and so everyone lives by the strictest terms on offer. What is being drafted in Washington is a version of the same leverage, aimed at a different point in the stack. Europe conditions access to its consumers on compliance with its text. The United States is reported to be preparing to condition access to its market on investment in its factories. The mechanism is not a fine for what you do to users; it is an allowance for what you build on American soil.

That is a meaningful shift in how a great power uses a market as an instrument, and it will have the same extraterritorial reach the European version does. A Taiwanese firm's decision about where to build its next fab, a Korean memory maker's capital plan, a systems assembler's choice of which country to finish its servers in — all of it starts bending around a United States allowance formula, whether or not the company sells much into the United States directly, because the firms it sells to are bound. Set the terms on which the world's most important market may be supplied, and you set terms the whole supply chain reorganises to meet. Brussels writes the rules and everyone lives by them; Washington is learning to write the invoice and make everyone build toward it.

None of this is settled, and the honest summary is the unglamorous one. There is no proclamation yet. The rate may never be 100%, the exemptions may partly survive, and the phased rollout may blunt the shock the memory-stock moves are pricing. But the design being reported is clear enough to describe accurately, and describing it accurately means not stopping at the rate. The story is not that Washington may tax laptops. The story is that Washington is turning a tariff into a licence to import, and pricing that licence in domestic factories. Read the allowance. That is the part that binds.