The Signal — in one line

For two years the market priced US technology on the assumption that anything built in America could be sold to everyone.

Three developments now say that assumption may be dead — and a frictionless-global premium may be repriced into a permissioned-access discount across the entire US commercial stack, not just AI.

This is a regime call, not a stock call. In this house, regime selection has always mattered more than any single model, name, or quarter — it is the larger source of both risk and edge. What changed is not a company. It is the regime the entire US technology complex operates inside.

The thesis in three pillars

The mistake is to read last week as an AI story. It is a US-commercial-output story with three reinforcing legs:

1 Supply side — universality of designation. It is no longer just chips and GPUs. Any US product, service, platform, API, workflow, model, software feature, or commercial capability can be moved into the restricted bucket the moment Washington deems it strategic. The category is not “AI.” The category is “anything the United States controls and chooses to control further.”

2 Identity — the firewall is gone. The new restriction reaches not only foreign countries but foreign nationals working inside the United States, for US companies — including a company’s own non-citizen employees. A “domestic transfer” now needs a license. That converts an export-control question into an identity-based access-control regime, and it means even the domestic market and a firm’s own workforce are now permissioned, not open.

3 Demand side — trust is already walking. And here is the part that does not require Washington to ever act again: even if the US never reclassifies another product, the rest of the world no longer trusts that it won’t. That single realization is enough. Governments, enterprises, and buyers are already legislating, budgeting, and migrating away from US dependency — Microsoft, US cloud, US SaaS, US infrastructure. The demonstrated possibility of restriction is itself the catalyst for diversification.

Stack all three filters — lost foreign markets, lost foreign-national access, lost trust — and the conclusion is mechanical:

If the realistically serviceable market is only ~25% of the global market the multiples were built on, the multiples must come down. Not cyclically. Structurally.

Level 1 — Regime · What rules are we playing under?

The old regime for US technology was the most benign in the history of capitalism: build once, deploy globally, sell everywhere, add users at near-zero marginal cost, expand margins as usage scales, and let the whole world treat US infrastructure as neutral plumbing. That regime is the foundation of every premium multiple on every US technology balance sheet.

The hardware layer left that regime years ago — chips, equipment, compute caps, geo-tracking. Investors learned to discount silicon for control risk. The working assumption was that the control stopped at the wafer.

It did not. Control has now jumped the wafer and can attach to any layer of any US commercial product, and the gating variable is no longer geography alone — it is identity and government permission. When access can be suspended by nationality, including inside the US, and when foreign buyers can pre-emptively walk to avoid the risk, US technology has left the open-distribution regime entirely.

A regime change is precisely the event that invalidates a model built for the prior regime. Every premium multiple on US tech is a model fit to the old regime. It is now mis-specified.

Regime read: 🔴 broken — open distribution has given way to permissioned access plus a trust exodus.

Level 2 — Trend · One-off or direction?

The direction is the whole point.

Trend read: 🔴 escalating, two-sided, and structural — not episodic.

Level 3 — Health · How sound is the thing being repriced?

The businesses are mostly excellent. That is not the question. The question is the quality of the multiple, not the quality of the company.

Health read: 🟡 businesses sound; valuation architecture fragile.

Level 4 — Context · The catalysts

The supply + identity proof-of-concept.

Context read: live, named, legislated, and self-reinforcing.

Level 5 — Timing · Is it priced? When does it bite?

It is barely priced. Most of the US technology complex is still carried on a global-TAM assumption — global enterprise rollout, global consumer scale, global developer usage, global cloud distribution. The repricing to a serviceable, permissioned, trusted-jurisdiction TAM has not happened.

The trade is the gap between a market still pricing the event clock and a world now governed by the trust clock.

Timing read: 🟡 early — the world has begun repricing; the multiples have not.

Level 6 — Decision · Posture

This is not advice. It is the positioning logic applied to our own

Posture: demand a discount for the global-scale premium on US commercial output — broadly, not just in AI

• Separate the premium from the business. Own the cash flows; question the portion of the multiple that is a pure frictionless-global assumption. That portion is what now carries the discount, and it sits in far more names than the obvious AI ones.

• Re-underwrite on serviceable, permissioned, trusted-jurisdiction TAM. Any thesis that reads “global adoption, rising ARPU, margin expansion, new upgrade cycle” must be re-checked against a world that can restrict access and one that is choosing to leave. Where the growth story leans on access that can be restricted or trust that can be withdrawn, the multiple is too high.

• The most exposed names are not the most commoditized — they are the ones whose entire premium rests on unrestricted, globally-trusted distribution of US-controlled technology. That includes the embedded-dependency incumbents (US hyperscale cloud, US productivity and enterprise software) precisely because they are the targets of the sovereignty migration.

• Watch for the mirror-image beneficiary. A trust exodus is a reallocation, not pure destruction. Non-US and sovereignty-aligned providers, ex-US infrastructure, and “trusted-jurisdiction” alternatives sit on the other side of the same flow. The barbell’s ex-US leg is doing real work here.

• Hold the hardware/software distinction we already drew, and extend it. The silicon was discounted for control risk long ago; that is why semis and software have been diverging rather than trading as one “tech” block. This special extends the same logic across the whole US commercial stack: the discount the chips already wear is now being fitted to software, cloud, and services.

Decision read: trim the US global-scale premium, keep the cycle, fund the trusted-jurisdiction other side.

Level 7 — Risk · What invalidates this signal?

Conviction without the counter-case is just narrative. The thesis is wrong, or early, if:

The counter-case

• The 25% is too aggressive. A 75% cut to reachable TAM does not mechanically equal a 75% cut to the multiple — surviving permissioned markets may carry higher margins, much value sits in the installed base rather than future TAM, and switching costs are real. The direction (multiples down) is robust; the magnitude is the live debate.

• The migration is slower than the headlines. US hyperscalers still hold roughly 70% of the European cloud market, and untangling embedded workloads is a multi-year effort. Plans for autumn 2026 are plans, not completed migrations; full deployment may not arrive before the end of the decade.

• The trust damage is not US-specific. The CLOUD Act reaches any company with US operations — including some European providers — so “sovereign” alternatives may be less clean than advertised, blunting the exodus.

• Capability commoditization defangs the supply-side control. If restricted capabilities are widely available elsewhere, designations become unenforceable and distribution stays effectively global.

• The premium was never truly global. For names already concentrated in the US and trusted allies, there is less global TAM to discount.

Risk read: the event is reversible and the migration is slow; the variable to watch is the trust clock, not the news cycle.

📊 The Instrument — National-Security Discount Exposure Matrix

Insulation score (0–100) · lower = more exposed

The deepest exposure is not in the most commoditized layers — it is the ones whose premium assumes unrestricted, globally-trusted distribution. Closelooknet Exposure Matrix.

Reading the matrix: the deepest exposure is not in the most commoditized layers — it is in the ones whose entire premium assumes unrestricted, globally-trusted distribution of US-controlled technology. Hyperscale cloud and embedded enterprise software score badly not because of any product flaw but because they are the literal targets of the sovereignty migration. The silicon scores higher only because the market already did this work in 2022–2024. The point of the special is that the rest of the stack has not.

How it maps to the names you watch

• US mega-cap software / cloud / NDX: carries the largest unpriced gap between global-TAM pricing and permissioned-and-contested reality. Most room for the discount to land.

• Semis / SMH: already wear a control discount; the marginal new national-security surprise is smaller here — relative insulation, not an all-clear.

• Ex-US / sovereignty-aligned exposure: the other side of the same flow. A trust exodus reallocates spend; it does not vaporize it.

The divergence thesis generalizes. “Hardware vs software” was cut one. “US frictionless-global premium vs permissioned-and-contested reality” is cut two — same engine, applied across the whole stack.

If the reachable market is a quarter of what the multiples assumed, the multiples must decline. The event will be argued in the news cycle. The discount will be priced in the multiple. Watch the second one.