Export Controls and the Global Chip Map

Coverage of chip export controls is a stream of country names and dates, which is the part that goes stale fastest and matters least. The durable thing is the machinery: what the instrument is, what it keys on, and why it binds companies that are not in the country issuing it.

What an export control actually is

It is a licensing requirement, not a ban. A regulation says that a defined item, going to a defined place, or to a defined recipient, or for a defined use, may not cross a border without a licence — and then states the policy for granting one, which ranges from routine approval through case-by-case review to a presumption of denial. The practical effect of “presumption of denial” is a ban with an appeals process, but the legal form matters: it means the boundary can be moved administratively, without new legislation, in either direction.

Two features that surprise people. First, a transfer does not have to cross a border: releasing controlled technology to a foreign national inside your own country can count as an export to their country — the so-called deemed export, which turns hiring and research collaboration into a compliance question. Second, the controls follow the item: a re-export from a third country is still regulated by the original jurisdiction.

The four keys

A transaction needs a licence if any key trips. That is why compliance is not a matter of checking a product against a list.

Enforcement is the part that receives the least attention and does the most work. Because the item is physical and fungible, the practical problem is diversion: a legitimate sale to a permitted buyer in a permitted country, followed by an onward shipment that nobody declared. Countering that shifts the burden onto sellers and freight forwarders — end-use statements, red-flag screening, post-shipment verification, record keeping — and turns a rule about borders into a due diligence regime for private firms, enforced by penalties for failing to look rather than for intending harm. Anyone assessing whether controls “work” should be looking at that machinery, not at the published thresholds.

Why one country’s rules reach the whole chain

The mechanism worth understanding is the foreign direct product rule. Ordinarily a country regulates exports from its own territory. A foreign direct product rule extends jurisdiction to items made outside that country, if they are the direct product of that country’s controlled technology or software, or if they are made by a plant that is itself the direct product of it.

In a supply chain where a small number of tools, design programs and intellectual property blocks are effectively unavoidable, that condition is satisfied almost everywhere. A chip designed in one country, fabricated in a second and packaged in a third can fall under the rules of a fourth that touched none of those steps. This is the single most consequential design choice in the regime, and it is why the practical question “can this company buy that chip” is rarely answered by looking at where either of them is.

It also explains why allied alignment gets so much diplomatic attention. Unilateral controls over an item with non-controlled substitutes leak; the leverage comes from chokepoints where substitutes do not exist, and maintaining a chokepoint requires the other jurisdictions that host it to go along.

Performance thresholds and the design-around loop

Controlling accelerators by technical parameter creates a boundary in a continuous design space. A vendor can build a part that sits below the line on the controlled metric while remaining useful — historically by trading one dimension against another, since a threshold combining, say, arithmetic throughput and interconnect bandwidth can be approached from either side. The regulator then adjusts the metric. This loop is not a failure of drafting; it is inherent to defining a control by measurable properties of a thing designers control.

Two structural gaps recur in commentary. Renting is not exporting: if the hardware stays put and access is remote, the item never crosses a border, so cloud access sits outside a classic goods control and has to be reached, if at all, by a separate instrument. And control by headline performance says nothing about aggregate capability — a large number of below-threshold parts wired together is a policy question the per-item metric does not address, which is why some proposals reach for quantity as well as performance.

A third gap is temporal. Controls apply to transactions, not to installed base, so hardware already in place is untouched and an announced tightening creates an incentive to buy heavily before it takes effect. An assessment that looks only at flows after the date will therefore overstate the effect, and one that looks only at the announcement will miss the stock that moved first.

What controls plausibly do, and what is contested

The arguments, stated as their proponents state them. In favour: the controls impose real cost and delay on a specific capability, and delay has strategic value even if it is not permanent; chokepoints in lithography and high-bandwidth memory are genuinely hard to replicate quickly; and enforcement against diversion is improving as attention shifts from the item to the logistics.

Against: controls create a durable incentive to build a substitute supply chain, converting a temporary dependence into a permanent competitor; they push demand into grey channels that are harder to observe than legitimate sales; they impose compliance costs and lost revenue on domestic firms, reducing the R&D that sustains the lead; and where controls are unilateral, they redirect purchasing rather than preventing it.

Both sets of effects are real and they operate on different time horizons — friction now, substitution later — which is why the net assessment depends almost entirely on the horizon you pick. That is a judgement, not a finding, and people who study this professionally disagree.

This page describes structure only. Lists, thresholds and country groups change frequently and classification is a technical legal exercise; check the current regulations and take advice. This is not legal advice.