An AI Exports Program Without a Map

Created by me using Midjourney

When the Trump administration issued Executive Order 14320 last July, directing the Commerce Department to create an American AI Exports Program (AAEP), its general rationale made sense. U.S. companies have been exporting AI hardware, cloud infrastructure, and AI models for years, but commercial incentives alone won’t win every contest that matters.

In many countries, especially where capital is scarce and state capacity is more limited, Chinese firms operate with advantages created by state-backed financing and sustained diplomatic engagement. The United States, by contrast, has relied primarily on commercial channels with less government support, which works well where markets are robust but can fall behind the Chinese approach where they aren't.¹

A targeted U.S. program that fills those gaps would push into the AI era the bipartisan conviction, shared across earlier administrations, that keeping Chinese vendors out requires putting credible alternatives in their place. The Biden administration's digital solidarity strategy made this premise explicit, drawing lessons from 5G competition with China and committing the U.S. to be the partner of choice for emerging economies deploying digital infrastructure. The first Trump administration's Clean Network framework operated on the same premise, promoting trusted vendors and proposing to back that commitment with financing. The AAEP, in theory, extends that approach to AI exports.²

Whether the Trump administration can execute this idea effectively depends on whether the program can resist two gravitational pulls that the April 1 Federal Register notice implementing the AAEP does little to counter.

First, commercial gravity, which will draw consortia toward easy markets where deals would happen anyway.

Second, political gravity, which will pull the program toward deals that administration officials and well-positioned private sector actors want to advance, regardless of whether those deals would have a strategic impact in the markets that matter most.

The notice does not identify any priority target markets, sets no usable criteria for strategic prioritization, and provides only thin guidance on where government support is most needed.

The repeated invocation of “national interest” — in four different contexts, never defined — is the clearest symptom of this: a program that gives government officials wide discretion, outsources the government's strategic analysis to private-sector applicants, and provides partner countries with few coherent signals about what the United States is trying to accomplish.

Four Uses of “National Interest,” But No Definitions

The notice uses “national interest” as the standard for designation decisions, the threshold for National Champion Enterprise (NCE) exceptions, a required element of consortium proposals, and an expansive standard for target market eligibility. In each of these four uses, the term is invoked but never defined. And the four uses aren't interchangeable.³

When “national interest” governs a designation decision (Section IX), it is the standard by which the Commerce Secretary approves or rejects a proposal — a determination made in consultation with State, Defense, Energy, and the White House Office of Science and Technology Policy (OSTP), with no fixed scoring formula and no single dispositive factor.

When it governs an NCE exception (Section II), it is a geopolitical judgment about whether a specific foreign firm's inclusion advances the “national interest”—a different question that requires different institutional inputs.

When it appears as a required proposal element (Section VI), it shifts the burden of strategic argument to the applicant. Each consortium must explain why its package advances U.S. national interests in its chosen markets, even though the government has published no theory of which markets or objectives matter most.

And when it’s meant to guide target market selection (Section V), it is standing in for a strategic map the government hasn't drawn that would tell consortia where American backing is most needed and most consequential.

Treating all four as a single, undefined phrase leaves each vulnerable to inconsistent, unpredictable, or politically influenced outcomes.⁴

The notice’s own language confirms how wide that opening is. It states that “no single factor is necessarily dispositive” and that Commerce will not rank proposals under a fixed scoring formula. Generally, that flexibility is appropriate for the case-by-case judgments the program will face. But flexibility without criteria produces discretion without guardrails — and discretionary national interest determinations made by political appointees, with no public ranking methodology and no stated criteria, are structurally vulnerable to outcomes driven by political pressure or institutional disagreement rather than strategic merit.

Companies Will Do the Strategic Thinking

Per the notice, consortia must “identify at least one target country or regional bloc for which its package is intended” and explain “why the proposed offering would advance U.S. national interests in those markets.” The government, meanwhile, offers no list of priority countries and little guidance on where U.S. engagement would be most decisive.⁵

The geopolitical analysis — the work of determining where Chinese AI infrastructure is gaining ground and where American engagement would be most consequential — is in effect being outsourced to the private sector. But the problem extends beyond outsourcing to a structural selection bias.

A design that asks companies to self-nominate target markets will likely surface proposals from places where firms already have pipelines and high confidence of closing. Those are precisely the markets where government support is least necessary. The markets where public backing is outcome-determinative — where the deal doesn't happen without diplomatic weight and financing — are the ones that few if any companies would choose first.

This outsourcing is underscored by a difference between the EO and the notice. The executive order directed proposals to identify “specific target countries or regional blocs.” The notice widened that to packages eligible for “any foreign market where designation would advance the national interest.” The administration started with country language in the EO and ended with market language in the notice, softening the program's geographic focus that would have given it strategic shape.

It may be that the program is sequencing deliberately — building a supply-side catalog of export-ready packages before committing to demand-side country priorities. The only available consortium path today is the pre-set one with standing teams offering a full stack on an ongoing basis, with no named buyer required. On-demand consortia, built around specific opportunities in specific countries, are deferred to a future release. Perhaps Commerce is assembling inventory first and will sharpen its country focus later.

The problem with this explanation is that it inverts the logic of good strategy. If the primary goal isn’t revenue but a strategic advantage in building AI infrastructure to outpace China, you build a pipeline around key contested markets where USG involvement and support would move the needle. Following corporate priorities in this context instead hands the strategic map to whoever submits first or submits loudest.

In this setup, consortia will rationally propose packages for markets where they already have a customer pipeline, even if specific potential customers aren’t named in proposals. The result will be an AAEP shaped by existing commercial interests rather than strategic necessity, systematically underweighting the contested markets where government intervention matters most and overweighting the allied and partner markets where commercial deals were already happening without a federal program and should continue to happen without one.

This isn’t a hypothetical concern. The first deal labeled as an AAEP project is Reflection AI’s partnership with Shinsegae Group (a Korean retail conglomerate) to build a 250MW data center in South Korea — a treaty ally with strong institutions and an existing U.S.-Korea technology cooperation framework that was already driving deals before the AAEP existed.⁶ Other deals tell the same story: AWS in South Korea, Google in India and Germany, Microsoft in Portugal, NVIDIA across Southeast Asia, OpenAI in Norway — all markets where commercial interest is already enough to drive significant AI and cloud transactions.

If the program’s portfolio continues to look like that list, the Trump administration will have built a government program to support deals that didn’t need supporting while the actual contested markets — in Africa, Latin America, and elsewhere — remain largely unaddressed.

The Missing Target Markets List

One approach to identifying priority AAEP markets starts with infrastructure geography. Around two-thirds of existing data centers are located in the United States, China, or Europe. Very few have been built in Africa, Latin America, or other parts of the Global South. That scarcity is the opportunity, and early engagement could materially shape the ecosystem. The window is open now and won’t stay open indefinitely.

China has already recognized this. Barred from the United States and increasingly restricted in Europe, Huawei has made emerging markets a priority. Its integrated stack — chips, cloud, models, and applications — is designed to be replicated across Central Asia, Latin America, Africa, and Asia. The foundation for this push was laid in telecom when Huawei built approximately 70 percent of Africa’s 4G networks and is now leading 5G expansion across more than 30 African markets.

One explanation for the notice’s silence on target markets might be institutional caution in the wake of the AI Diffusion Rule, which generated significant blowback from allies who felt treated as potential adversaries. But that caution, if it exists, reflects a category error. The Diffusion Rule was a supply-side restriction whose country designations carried restrictive consequences for every U.S. AI export, which is why it was diplomatically damaging. The AAEP is the opposite because naming a country as a priority target means telling it that the United States wants to compete for its business and is prepared to put diplomatic weight and financing behind that commitment.

As long as it is true (and it is generally understood) that the AAEP is a booster program for a few strategic deals that leaves untouched the vast majority of American AI export deals, explicit inclusion of a country in the AAEP won’t pose a problem.

The Sovereign AI Provision That Doesn’t Mention Sovereignty

The NCE provision allows foreign enterprises to serve as the “highest value” provider in Layer 1 (hardware and infrastructure) or Layer 3 (AI models and systems) within designated packages, provided they are from the package’s target country and the USG determines that their inclusion advances the national interest.

This is, in effect, a codification (with limits explained in footnote 7) of the G42 and HUMAIN deal structures — arrangements in which a Gulf partner’s national AI champion serves as the local anchor while American technology and security requirements shape the stack’s architecture. As a mechanism for deepening relationships with strategically important partners with significant AI ambitions and complex technology relationships, this is the right idea, and the administration should be encouraged to build an AI exports program that involves the active participation of allies and close partners.⁷

But the notice never uses the words “sovereign AI,” “AI sovereignty,” or any variant of either. This is a significant omission given that the Trump administration has deployed sovereignty language prominently in other settings. Through OSTP, it has explicitly stated its support for helping partners build sovereign AI capabilities with American technology — and even included the idea in the press release announcing the notice. But that language does not appear in the notice, the document that partner governments will read when deciding whether to engage. The program that is supposed to operationalize the administration’s sovereign AI effort doesn’t acknowledge the demand it is meant to answer.

The reality is that vagueness about NCEs and silence on sovereign AI will raise significant questions about the administration's commitment to international AI partnerships and the sovereign AI agenda that OSTP was actively promoting at the New Delhi AI summit just weeks before this notice was published.

Three Things the AAEP Should Do

None of this is an argument against the AAEP, but it is one for building it correctly. A program truly focused on competing with China’s state-backed AI infrastructure push — and on being a credible partner to countries that want American technology on predictable terms — would look different from the notice in three specific ways.

First, it would publish a priority market framework grounded in publicly available indicators of Chinese digital infrastructure penetration such as Chinese development financing, telecom deployment, data center presence, and AI model uptake. The organizing principle would be where USG support has a strong likelihood of changing the outcome. The framework should distinguish markets where private commercial demand is already sufficient from those where government backing is what makes an American offer viable or competitive. That distinction would give consortia a clearer signal about where U.S. support is strategically most valuable and give Commerce a principled basis for allocating scarce diplomatic attention and financing tools. Here’s what that might look like:

A priority market framework for the AAEP⁸

As currently designed, the AAEP risks drifting toward commercially self-propelling markets and the easiest contested markets, rather than toward early-stage and entrenched markets where government backing would have a greater impact.

Second, the program should stop using “national interest” as a single empty label for four different determinations. Market selection, package designation, NCE approval, and consortium composition are different questions and should be governed by different findings. Market selection should ask whether a country or region is strategically worth targeting. Designation should ask whether a specific package materially advances that objective. NCE approval should ask whether a local partner promotes sovereign adoption without introducing unacceptable security risk. Consortium review should ask whether the package contains sufficient U.S. or trusted-partner content across the stack. Commerce can retain final authority, but it should say which finding it is making and on what basis. A defined framework would (and should) not eliminate discretion, but it would channel discretion toward the program’s stated goals rather than leaving it open to whatever political dynamics happen to be operating at the moment of decision.

Third, the program should require public reporting on designation decisions stated in terms of the program’s criteria — not the confidential commercial details, but the national interest rationale articulated at a level of generality that allows Congress, partner governments, and the public to assess whether the program is doing what it says it is doing. Accountability is not the enemy of flexibility. It is the mechanism by which flexibility remains legitimate.

1. Conflict of Interest Disclosure: The views I share in this essay are my own. Although I hold professional affiliations and advisory roles with several organizations, I don’t receive compensation or instructions from them (or any other entities) regarding the views I express in essays, panels, or other public and semi-public forums unless otherwise stated in a specific piece or setting. In some cases, I might receive an honorarium from the publisher.

2. Prior to the AAEP, the U.S. government was already financing and de-risking overseas digital infrastructure through development finance, export credit, and grant-funded project preparation. The International Development Finance Corporation (DFC) committed up to $25 million to Convergence Partners’ digital infrastructure fund (2021), invested $50 million in African Infrastructure Investment Fund 4 (2023) with a mandate including digital infrastructure, and made a $50 million equity investment in Cassava Technologies (2024), a pan-African operator of fiber networks and data centers. The Export–Import Bank (EXIM) approved a $66.1 million guarantee for a national data center in Côte d’Ivoire (2025). The U.S. Trade and Development Agency (USTDA) has funded feasibility studies and technical assistance for both data centers and subsea cables, including support for data center and cloud architecture planning in the Maldives and Brazil. Together with other financing examples from the past decade, these activities demonstrate that AAEP builds on an established USG toolkit for financing and enabling global digital infrastructure rather than introducing a totally new approach.

3. The notice does list four factors in Section IX for Commerce to consider: compliance with program requirements, advancing EO 14320's goals, national security posture, and cybersecurity standards alignment. But a list of considerations is not a definition, and the factors neither bind the Secretary's discretion nor specify what outcomes would satisfy them (their relative weight varies by proposal and market, and no single factor is dispositive).

4. I took a quick look at the literature examining flexible and expansive nature and use of “national interest.” Nye described it as a “slippery concept” and argued that democratic politics determines how broadly or narrowly the U.S. defines it. Miroslav Nincic framed the subject in terms of “possible interpretations.” James Miskel warned that, in U.S. practice, the concept is often defined so broadly that it loses practical limits. Other scholars examine how governments construct national interest and shift the relative salience of its components over time (see Sabine Mokry, Peter Trubowitz, and Scott Burchill).

5. The one criterion the notice articulates — that proposals targeting markets where U.S. engagement can “materially influence” choices between American and Chinese technology will fare better — describes many markets where the program would operate and differentiates none of them. The notice uses “countries of concern” rather than naming China directly. In practice, “countries of concern” as defined in the 2026 National Defense Authorization Act encompasses China, Russia, Iran, North Korea, Cuba, and Venezuela, though in the context of the AAEP the main focus is on China.

6. The Reflection AI-Shinsegae deal is a useful case study in how the AAEP’s scope is being defined in practice, even before the program is formally operational. The Korean press described it as an AAEP project; American outlets were more cautious. For example, the Wall Street Journal characterized the deal as “a blueprint for what the Commerce Department hopes to support through an AI export program that is in the works.” And the USG itself did not explicitly call it an AAEP deal, though it has basically called other projects — namely, AI deals with Saudi Arabia and the UAE — models for the AAEP. Taken together, these early signals suggest the AAEP’s scope extends well beyond markets where commercial gravity is insufficient to attract U.S. companies without government support. There is a defensible case for some of that breadth: the Gulf deals involved geopolitical complexity and national security considerations that may have warranted closer USG involvement than a comparable transaction with a long-standing democratic partner would require. But a program without clear scope limits is vulnerable to mission creep that carries two distinct risks: (1) interference with commercial transactions that would have proceeded on their own terms and (2) the concentration of government favor in ways that reward political access over strategic merit.

7. The home-country-only rule for NCE status merits a second look. A firm designated as an NCE for its home country cannot hold that status for packages targeting other countries — so G42 could be an NCE for UAE packages but not for Egypt or India. This prevents a single foreign firm from using NCE status as a vehicle for regional expansion under American diplomatic cover, which provides useful scope clarity, but it would also keep arrangements like Microsoft-G42’s AI diffusion plans for Africa, the Middle East, and Central Asia from qualifying as AAEP consortia. The provision also invites comparison with the Data Center VEU mechanism in the Diffusion Rule. Both designate specific foreign entities as trusted recipients of privileged access to U.S. AI technology, and both require government vetting of entities that would raise security concerns.

8. The country examples in each category are illustrative and the list is deliberately incomplete. Several countries that would likely qualify for, say, Category 3 are omitted because current bilateral dynamics would limit the effectiveness of a USG-backed engagement strategy, however sound the underlying strategic case. An AAEP that relies on diplomatic weight and financing commitments requires a baseline of bilateral trust that doesn’t currently exist in every country where Chinese AI efforts are gaining ground.