The Quiet Transfer: What Chip Export Controls Are Really About Now

The news, if you read it quickly, looks like a regulatory shuffle. The US Commerce Department has withdrawn a planned rule that would have tightened restrictions on AI chip exports. A broader framework is coming. In the gap, a deadline on 13 April lapsed for integrated circuit designers not approved under the Bureau of Industry and Security framework, who lost their "authorised" status, a deadline most procurement teams, by all accounts, missed.¹

This is the sort of story that gets filed under "trade policy" and read by almost nobody. I want to argue it belongs somewhere else. What is actually happening, under the cover of technical rulemaking, is that the United States is building an apparatus in which access to the most consequential industrial input of this decade, advanced AI compute, becomes contingent on private firms' willingness to align their overseas investment decisions with American strategic preferences. The chip is the lever. The firm is the fulcrum. And what's being moved is the boundary between public authority and private capital in a way that deserves more attention than it's getting.

This is the sort of story that gets filed under "trade policy" and read by almost nobody.

I want to be careful here. This is not a piece about whether chip export controls are good or bad in the abstract. There are serious arguments for constraining the flow of frontier compute to geopolitical adversaries, and I take them seriously. It is a piece about what we are actually building when we build "conditions-based controls," and who ends up holding power inside that structure.

The shape of the shift

The framework the US is reportedly developing would link export permissions to overseas investment commitments and strategic alignment with the United States.² Read that sentence again. It is doing more work than it appears to.

The older model of export control, going back decades, is categorical. Certain goods, to certain destinations, full stop. You either have the licence or you don't. The grounds for the licence are spelled out in regulation, they apply equally to all applicants, and, at least in principle, the decision is bounded by law rather than by negotiation.

Conditions-based controls are a different creature. They are, in effect, a discretionary permissioning regime in which the conditions can be shaped case by case. The firm wants to sell chips into a jurisdiction. The government wants the firm to build a fab in Arizona, or not build one in Malaysia, or not hire certain researchers, or sign onto certain commitments about who can and can't use the downstream model. The export licence becomes the currency of the negotiation. The firm pays in capital allocation decisions. The government pays in access.

This is not new in kind. The United States has used export controls as industrial-policy instruments before, the Foreign Direct Product Rule applied to Huawei was an early template, and the October 2022 and October 2023 chip controls extended it.³ What's new is the degree of explicitness, and the degree to which the conditions now extend past destination of sale to the full strategic posture of the exporting firm.

Why the framing matters

The dominant way of writing about this, and I've read a lot of it this month, treats it as a question of effectiveness. Are the controls working? Is China's indigenous chip industry catching up anyway? Are the loopholes too big? These are real questions. They're not the most interesting ones.

The more interesting question is what kind of relationship between state and firm we are building, and whether it has the legitimacy it needs to hold.

Consider the asymmetry. A small number of US firms, Nvidia most prominently, but also AMD, Intel, and the design and equipment firms that support them, hold what has turned out to be near-monopolistic control over the physical substrate of advanced AI. Nvidia's data centre revenue alone passed $115 billion in its fiscal 2025, and the company's market capitalisation has, at various points in the last eighteen months, exceeded the combined GDP of several G20 economies.⁴ When the US government negotiates conditions with a handful of firms like this, it is not negotiating with "industry" in any pluralistic sense. It is negotiating with a concentrated oligopoly whose interests will shape the outcome at least as much as any abstract public purpose.

And here is where I find it hard to be sanguine. A conditions-based regime run through a handful of gatekeeper firms is, structurally, a regime in which those firms become partial instruments of US foreign policy, and, simultaneously, in which US foreign policy becomes partially shaped by the commercial interests of those firms. These are not contradictions. They are the same phenomenon viewed from two angles.

The workers, customers, and citizens affected by this arrangement, both inside the US and outside it, have, as far as I can tell, no meaningful voice in how the conditions are set. A Malaysian fabrication worker whose employment depends on whether a US firm is permitted to locate capacity in her country is not consulted. A Taiwanese engineer whose career prospects shift with each iteration of the framework is not consulted. A British or German AI researcher whose access to frontier compute is mediated through these bilateral negotiations is not consulted. And inside the United States, the public debate about what strategic alignment should mean, about what we are actually trying to achieve, at what cost, is almost entirely absent, because the negotiation happens in a register that doesn't make it to the front page.

The missed deadline

I want to come back to that small detail in the news, because it illustrates the texture of this regime better than any abstract argument. On 13 April, IC designers not approved under the BIS framework lost "authorised" status. Most procurement teams missed the deadline.

Think about what that means in practice. Somewhere in a mid-sized European electronics firm, or a South Korean industrial buyer, or a Brazilian cloud operator, a procurement manager is now discovering that orders placed last month may be in regulatory limbo. The manager did not know the deadline was coming, or did not understand what compliance required, or assumed, reasonably, that the rule being withdrawn meant there was more time. There wasn't.

This is what conditions-based regimes feel like from below. The rules are complex, the deadlines are shifting, the enforcement is discretionary, and the burden of understanding falls on thousands of firms with neither the legal sophistication nor the political access to navigate it. The firms that can afford dedicated export-controls counsel, generally, the largest and best-connected, do fine. The rest absorb the cost in the form of missed deadlines, seized shipments, compliance-driven delays, and a general chilling of cross-border commerce that never gets measured because it shows up as deals that simply weren't done.

There is a literature on this, mostly from trade economists, that documents how compliance-heavy export regimes disadvantage smaller firms disproportionately.⁵ What's less well-studied, because it's harder to measure, is the political effect: the way that regimes like this concentrate influence in the firms capable of navigating them, which tend to be the firms already closest to government. The flywheel is unsubtle.

The "strategic alignment" problem

The phrase "strategic alignment with the US" is doing a lot of work in the reporting on the forthcoming framework, and it deserves to be pulled apart.

In the most benign reading, strategic alignment means something like: firms receiving US export permissions agree not to undermine US national security through their downstream conduct, no re-export to sanctioned entities, no assistance to designated military end-users, reasonable KYC on their customers. That is coherent and, within limits, defensible.

But the phrase is capacious. It can also mean: firms agree to locate capacity in politically favoured jurisdictions. Firms agree not to pursue certain partnerships. Firms agree to participate in information-sharing arrangements with US agencies. Firms agree, in effect, to treat their global footprint as something to be negotiated with Washington rather than optimised on commercial terms.

Each of those is a more significant ask than the previous one, and each takes us further from export control as traditionally understood and closer to something that looks like managed industrial policy, but managed without the democratic accountability that industrial policy, at its best, has historically required.

I don't think this is a conspiracy. I think it's what happens when a regulatory toolkit designed for one purpose (stopping specific goods reaching specific destinations) gets repurposed for another (shaping the global allocation of a strategic industry) without the legal and institutional scaffolding being rebuilt to match. The result is a regime with the enforcement power of sanctions and the reach of industrial policy but the deliberative process of neither.

What the benefits case looks like, taken seriously

I said at the top that I take the arguments for compute controls seriously, and I want to honour that. The case is not trivial.

Advanced AI systems, trained on the largest available clusters, have military and intelligence applications that a responsible government has legitimate interests in shaping. The 2022 and 2023 controls appear, by most assessments, to have meaningfully slowed Chinese progress on the highest end of compute, not indefinitely, but measurably, and that slowdown has bought time for other policy responses to develop.⁶ A world in which compute flowed entirely frictionlessly to every state actor willing to pay would be a worse world for most of the things I care about, including the welfare of people inside the states in question.

So this is not a piece arguing that export controls are illegitimate. It's a piece arguing that the form they are now taking, discretionary, conditions-based, negotiated firm by firm, is doing something different from what export controls have traditionally done, and that the difference matters.

What's actually at stake

Let me try to name it cleanly. When the state's instrument for pursuing public purposes becomes the conditional licensing of a handful of private firms, three things happen that don't happen under more categorical regimes.

First, the boundary between public policy and private interest blurs in ways that are hard to unblur. The firms being regulated become partners in the regulation. Their preferences become policy inputs, not just compliance concerns. Over time, this tends to produce regulation that reflects firm interests better than it reflects public interests, not through corruption but through the simple dynamics of who's in the room.

Second, the affected publics, workers in affected sectors, consumers in affected markets, citizens of affected allied and partner countries, have less purchase on the outcome than they would under a more transparent regime. There is nothing to lobby on, nothing to vote on, nothing that shows up as a discrete policy choice. There is just the ongoing murmur of bilateral negotiation between Commerce and a handful of firms, occasionally punctuated by a rule that's announced and then withdrawn and then replaced with something broader that takes another year to come into focus.

Third, the whole arrangement becomes brittle in a specific way. It depends on the continued willingness of the gatekeeper firms to play along, and on the continued capacity of the state to credibly threaten them. If either weakens, if a firm finds it can prosper outside the regime, or if a new administration decides the conditions should be entirely different, the system has no inertia. Categorical rules outlast the people who wrote them. Negotiated arrangements do not.

What to watch

I'd watch three things over the coming months.

The first is the shape of the broader framework when it emerges. Specifically, how explicit the conditions are, whether they're codified in rule or left to case-by-case negotiation, and whether there's any mechanism for affected parties outside the negotiating dyad to weigh in.

The second is the behaviour of the gatekeeper firms. Do they push back publicly on terms they don't like, or do they absorb the conditions quietly and pass the costs downstream? The answer will tell us a lot about the real balance of power inside the arrangement.

The third is what happens to the firms and countries at the edges of the regime, the mid-sized buyers who missed the 13 April deadline, the smaller designers caught out by shifting authorisations, the allied states discovering that their own industrial policy is being shaped by Washington's conditions on US suppliers. These are the people who pay for the friction, and whether they have any recourse will tell us whether the regime we're building is one that takes their interests seriously or one that treats them as collateral.

None of this means export controls should be abandoned. It means we should be honest that what we are now building is not export control in the older sense. It is something larger, and messier, and more consequential, and it deserves the kind of public argument that the forthcoming "broader framework" will almost certainly not receive.

I find it hard to believe that a regime this significant, shaping the allocation of the defining industrial input of our decade, should be built mostly out of view. And I think we should say so while the scaffolding is still going up.

---

Footnotes

Footnotes

1. Withdrawal of the planned rule and the 13 April lapse of BIS-approved status for non-authorised IC designers as reported in the source research file; the procurement-teams-missed-it detail reflects the widespread reporting that firms outside major counsel relationships had limited notice of the cutover. ↩

2. The "conditions-based controls" framing, linking export permissions to overseas investment commitments and strategic alignment, is drawn from the Commerce Department's own signalling about the forthcoming framework, as referenced in the source research file. ↩

3. The Foreign Direct Product Rule's expansion to cover Huawei in 2020, and its later extension in the October 2022 and October 2023 BIS controls on advanced computing chips and semiconductor manufacturing equipment, are the proximate precedents. See BIS rule 87 FR 62186 (October 2022) and subsequent amendments. ↩

4. Nvidia fiscal 2025 data centre segment revenue, as reported in the company's Form 10-K. ↩

5. See in particular Crozet and Hinz on the firm-level compliance costs of complex trade regimes, and Evenett's work at the Global Trade Alert documenting the disproportionate burden of discretionary regimes on smaller exporters. ↩

6. Assessments of the 2022–2023 controls' effectiveness include work from CSIS (Allen), CNAS, and the Rhodium Group, which broadly converge on the view that the controls slowed frontier Chinese compute capacity meaningfully while accelerating indigenous substitution efforts at lower nodes. ↩

¶