You Can’t Measure a Negative

Pax Silica has a defect no new member can fix: nobody will ever know whether it worked.

This week the European Union signs onto Pax Silica, the State Department’s coalition for securing the AI supply chain, bringing the membership past fifteen and committing the bloc to at least $40 billion in American chip purchases on its way in. The debate in Brussels ran on sovereignty: France called the initiative an attempt to colonize Europe, Germany called it unity, and the Commission split the difference by signing a non-binding declaration and announcing a technological independence package on the same day. All of this concerns whether the coalition will hold. The harder problem sits upstream of cohesion, and nobody in Brussels or Washington has named it. Pax Silica is a strategy whose success cannot be observed.

Start by granting the idea its full strength, because the idea is serious. Washington has reclassified AI from an industrial sector to the organizing principle of national power. Jacob Helberg, the Under Secretary running the initiative, says it plainly: power is now measured by the ability to design, manufacture, and scale the technology stack. Once you accept that premise, the entire input chain becomes a security perimeter, and the perimeter runs through fifteen other countries’ mines, fabs, and ports. Pax Silica follows from the premise the way containment followed from Kennan. As ideas go, this one is coherent, stated openly, and held at the top of the American government.

In my book The Innovation Machine: How China Creates and Adopts Technology Through Governance, I describe how China converts ideas into outcomes through what I call the I5 chain: Ideas flow into Institutions, which deploy Instruments, which produce Infrastructure, which generates Impact — and the impact loops back to fund and correct the next round. The chain is only as good as its joints, and the variable that decides everything is fidelity: how much of the idea survives transmission from each layer to the next. China’s system runs at high fidelity by design. Run Pax Silica through the same five layers and you can watch the idea degrade at every joint.

Begin with the institution layer, and look at what the idea was given to live in. The Pax Silica Declaration, signed in Washington on December 12, is explicitly non-binding, a fact the State Department confirmed in writing to reassure Paris. Its operating machinery is a Steering Group that convenes once a year, supervising working groups on AI safety, semiconductor resilience, and digital trade standards. Institutions earn their keep by performing a handful of functions: binding members to commitments, verifying compliance, sharing burdens when costs land unevenly, settling disputes, and pooling information. Score the declaration honestly and it performs one of the five. It pools information. There is no mechanism to verify that any member implements aligned export controls, no penalty if one doesn’t, no formula for compensating the country that absorbs Chinese retaliation, and no arbiter for the day Washington restricts a sale a partner believed it was promised. Members joined quickly because exit costs nothing, which means the membership count measures enthusiasm for signing, and signing measures nothing.

The instruments tell the same story from a different angle. The American toolkit is the most developed in the world: entity lists, the Foreign Direct Product Rule, investment screening in both directions, tariffs, subsidies. Nearly all of it is built for denial, and denial instruments share a property their owners prefer not to discuss. They depreciate with use. Every export control is simultaneously a denial and a procurement order for the domestic Chinese alternative, delivered with political cover attached. SMIC’s 7-nanometer process, Huawei’s Ascend line, and the domestic memory makers all scaled inside the protected market the controls created. The Foreign Direct Product Rule, the extraterritorial workhorse that lets Washington reach any chip made anywhere with American tools, decays the same way: each application gives ASML, Tokyo Electron, and their Chinese competitors another reason to design American content out of their machines. Pax Silica exists because these instruments stopped working unilaterally. Multilateralizing them makes them slower without making them tighter. One licensing decision in Beijing moves faster than fifteen capitals harmonizing anything, and asymmetric cycle time is itself an instrument property. It favors the side that doesn’t need a summit.

Below the instruments sits the layer where the intangible becomes tangible, and where the coalition’s real deficit shows. Declarations, rules, and budgets are made of words, and words are cheap to revise. A fab is five years of construction and twenty of payback, and it gets built only by someone who can hold a decision for a decade. China assigned that job to the state, which holds decisions across five-year plans as a matter of institutional design. America assigned it to private capital hedging against a state that reverses itself every election, and increasingly to a state that cannot hold its own public-works decisions either; ask California about high-speed rail. Now count what the coalition’s toolkit actually constructs. The denial instruments build nothing by definition. The construction side so far consists of a chip purchase order and one zinc smelter in Tennessee, the first built in America since the 1970s, announced three days after the summit as proof of concept. A coalition can multiply signatures. It cannot multiply the ability to hold a decision.

Which brings the chain to its final layer, impact, where strategies either produce results or excuses, and where Pax Silica’s structural problem lives. Infrastructure is capacity; impact is capacity used; and impact is what closes the loop, because results flow back and tell a system which of its tools worked. China wired that feedback in deliberately. Plans get revised against measured outcomes, officials get evaluated against targets, and the revenue from infrastructure that runs funds the round that follows. The American export-control regime has never once been recalibrated against its outcomes, because no institution in Washington is accountable for them. The Bureau of Industry and Security answers for enforcement, never for effect. The signal arrives each quarter and no desk is assigned to receive it.

But the deeper defect is not that Washington fails to read the results. It is that a denial strategy produces no results to read. The success of Pax Silica, on its own terms, is an absence: Chinese capability that never materialized. An absence generates no number. You can estimate a counterfactual, but you cannot observe one, and no government on earth holds anyone accountable against an estimate. Construction is self-auditing; the fab yields or it doesn’t, the price of a token falls or it doesn’t. Denial never finds out. The coalition’s annual Steering Group can convene for a decade and at no meeting will anyone be able to say whether the strategy is succeeding, because the metric of success is invisible in principle. You can’t measure a negative.

Beijing’s fabs are countable. The fabs that supposedly don’t exist because of export controls are a modeling assumption. One side of this competition is audited by reality every quarter. The other side grades its own homework.

---