The American AI Exports Program faces its first major test this summer, with initial bids due at the end of June. On its current trajectory, it risks mostly claiming credit for deals U.S. firms were going to close anyway, while doing little to attract the foreign governments it is meant to win over.
The program is the administration’s flagship vehicle for promoting American AI abroad. It aims to use federal financing, expedited export licensing, and diplomatic support to build market share across the AI stack before Chinese competitors lock it up. The strategic logic is sound. China is moving fast to position its AI products as the default across Southeast Asia, Latin America, and other contested markets. While chip constraints prevent it from offering full-stack solutions as advanced as America’s, it will pair the chips it can export with broader tech infrastructure and state backing that American firms struggle to match. If its export push succeeds, a version of the 5G story could repeat, with Chinese infrastructure embedded in critical digital supply chains throughout the developing world.
But the U.S. program’s Call for Proposals (the Call), released last month, suggests its design will not meet the moment. The Program seeks to direct priority exports to markets of geopolitical significance to the United States, but the Call asks American companies to assemble full-stack export packages and identify target markets independently, with scant guidance on which markets matter or how proposals will be judged. It puts the onus instead on applicants to identify markets and explain how U.S. national interests can be advanced in them. It also gives governments wary of U.S. lock-in limited voice in what the program offers them, raising the risk that they might not participate either. On its current trajectory, the program risks serving neither U.S. industry nor foreign partners, including those currently being wooed by Chinese competitors.
There is still time to fix it. American AI is competitive on its merits in most of these markets; the export program’s job is to reduce barriers to strategically important deals, not to assemble offerings no one asked for. Doing so will require guiding industry to the right partners—important clients in key geopolitical swing states—while giving it the flexibility to shape technical offerings, rather than outsourcing key strategic decisions to companies and relying on opaque processes to evaluate them.
Pre-Set Consortia Need Clearer Strategic Direction
The U.S. Department of Commerce announced in the Call that the program would run on two tracks. The first involves pre-set consortia: groups of companies that assemble in advance to offer foreign buyers a full stack of American AI services covering hardware, data infrastructure, models, applications, and security. The AI stack offers a “menu” of options, and partners are allowed to procure specific layers without signing onto the full stack package. The second track involves on-demand consortia, which will form in response to specific export opportunities and cover the particular layers of the stack that a given deal requires. The administration is launching with pre-set consortia, with bids due June 30; it has deferred on-demand projects to a later date.
In theory, pre-set consortia could be well-suited to accomplish the goals of the program. By delivering full-stack U.S. AI solutions, they lock customers into the American tech stack and lower the risk of subsequent switching to Chinese vendors. They also provide an American alternative directly positioned to compete with Chinese full-suite digital products and suitable for less experienced customers who would struggle to assemble these solutions on their own.
In practice, however, pre-set consortia may struggle to accelerate U.S. AI exports to contested markets. They rely on industry to self-assemble into consortia, outsource strategy to whoever happens to organize a bid, and disadvantage smaller companies whose specialized products may be exactly what specific markets need. The Department of Commerce should either give the track real strategic direction or deprioritize it in favor of on-demand work.
The program is arriving at a fraught political moment. The push for sovereign AI is gaining momentum, and many prospective partners want American technology without looking dependent on Washington. The program risks making that balance harder to strike. Federally branded full-stack packages tie participating companies to the administration and hand sovereign AI advocates a ready-made foil. Even projects without federal branding have met with significant backlash. France’s decision to host its government Health Data Hub on Microsoft Azure, for example, became a years-long political controversy over potential U.S. government access to data and ended with France switching to a domestic vendor. The more that Washington’s offer looks like a way to deepen dependence rather than a tool that partners can direct to their own economic purposes, the harder it is for many foreign governments to say yes.
This is not a reason to abandon the program. But it does mean that the program needs to be narrowly targeted, guided by a clear theory of where the government genuinely adds value that it can articulate to U.S. companies to guide their proposals. The Call’s breadth and vagueness suggest the administration may have no such theory.
First, it provides companies little guidance on which markets the program seeks to focus on. It names no target countries or criteria for who buyers should sell to, despite the program’s aim of steering exports toward strategic markets. If the administration has strategic preferences, stating them would help companies decide whether to participate and how to craft proposals. Commerce’s silence suggests that it either has no view yet, does not want to publicly defend its view, or intends to defer to whichever markets industry brings forward—a shaky foundation for a program of economic statecraft.
The selection criteria are similarly underspecified. The Call says consortia will be evaluated according to a “national interest determination” and lists four factors, including a consortium’s “potential to advance the policy goals” of the executive order. But it does not define national interest, and each use of the phrase in the Call implies a slightly different meaning. It is also remarkably brief in its explanation of the four factors, outlining them in fewer than 70 words. Discretion has its place, but it should accompany a strategic framework, not substitute for one.
Commerce should publish a list of priority markets and provide more guidance on evaluation criteria for pre-set consortia. Regarding priority markets, the program should focus on large, strategically significant countries at risk of being locked into the Chinese stack, not close allies where U.S. firms already operate and where these risks are not at play. (We and others have suggested candidate lists elsewhere.) And while Commerce need not publish a full rubric, it should elaborate on the goals of the program, how to fulfill them, and what evidence will be considered at greater length.
On-Demand Consortia: Establish Government and Industry-Led Tracks
The administration has released fewer details about on-demand consortia, the teams to be formed by industry in response to a specific export opportunity. This track could work, but only if Commerce structures it so that the government shapes strategy while industry leads execution.
The administration should split the work of identifying buyers between government and industry. Industry-led opportunities give Commerce a light-touch way to facilitate strategically important deals by illuminating areas where U.S. government engagement is most needed to assist firms in entering new markets. These scenarios are most likely to happen when a firm has a successful product that it has sold in certain places but not others, whether because of regulatory barriers or lack of familiarity with the local market. Here, the U.S. government can offer introductions and other support to help these firms break into strategically significant markets where the U.S. does not yet lead.
But the government will still need to decide which industry-led opportunities to support. The right test is additionality: does federal support actually change what happens? Without answering that question, the program risks subsidizing deals in friendly markets that would have closed anyway. There are already signs of this. The first project marketed as part of the program was an AI data center in South Korea, a close ally with a sophisticated tech sector and existing American vendor relationships, not a swing state where U.S. firms need a push.
The current version of the program outsources this additionality test to industry and fails to address it directly. The Call asks companies to state the specific types of government support they need and leaves Commerce to consider whether those tools would add value. It appears that Commerce intends to use the specificity of a company’s proposal as a loose proxy for how much they need the assistance—an indirect way of getting at the necessity of federal support. Instead, Commerce should ask companies to identify the specific obstacle that is blocking a prospective deal, explain why the tools of the Exports Program would assist them, and scrutinize such claims.
When assessing which proposals to back, Commerce does not need to publish a checklist or a formula, but it does need a more disciplined framework for judging which deals are strategically valuable, feasible, and most in need of government support. On strategic value, the key question is: does the deal target a serious client, in a geopolitical “swing state,” for a use case sticky enough that integration costs and switching frictions make it a foundation for further American AI presence? To verify feasibility, the administration should weigh whether the deal can actually succeed by examining credible buyer-intent signals and a business plan (which the Call, to its credit, already requests for on-demand consortia), as well as the previous track record of the American vendor. Conditioning support on basic national security guardrails, including customer verification, cybersecurity requirements, and anti-chip smuggling measures, can also substantially reduce risk.
In rare cases, the U.S. government should take the lead in initiating deals. It should do so sparingly, because this option carries familiar risks: slow deployment, weak technical fit, and the danger of misreading partner demand or steering industry towards projects with poor commercial prospects. In many countries, visible U.S. government involvement may in fact be a liability.
But certain strategically important opportunities may face constraints only the U.S. government can resolve. Some governments may want assurance that an AI infrastructure deal fits within a broader strategic relationship with Washington. Major foreign banks, telecoms, or utility companies may want security, financing, or regulatory assurances that no commercial service-level agreement can provide.
In these cases, the government’s role should be to validate partner demand and remove diplomatic or regulatory barriers (where both possible and strategic to wider U.S. interests), not to centrally plan opportunities or dictate terms to companies. U.S. government involvement will likely be most helpful where data localization or intellectual property concerns are stumbling blocks to otherwise promising deals.
Here, the Department of Commerce and the U.S. Trade and Development Agency (USTDA) can play a matchmaking role. Commerce and USTDA should not choose between bids that satisfy program requirements, but should translate the foreign partner’s demands into a concrete procurement memo that firms can evaluate. The foreign partner would then select the vendor(s) they wish to do business with from a set of responding proposals compiled by Commerce and USTDA. Clarifying the opportunity can attract interest from firms that have relevant capabilities but might not otherwise have relationships in that country or know how to navigate their local procurement processes. Competition between American firms can signal to partners that the U.S. is serious about meeting partners’ needs and that they will not be locked into a single vendor, making participation more attractive. The Export-Import Bank, the U.S. International Development Finance Corporation, and USTDA can then provide financing and risk-mitigation support, including loan guarantees and political risk insurance, to whichever eligible bid the partner chooses.
***
Ultimately, the Exports Program needs buy-in from two groups: American industry and foreign governments. On its current trajectory, it risks serving neither. It outsources key big-picture strategic decisions to industry while giving companies little guidance on target markets, evaluation criteria, or partner demand. And while the government might be engaging with foreign partners behind closed doors or through bilateral Memoranda of Understanding, many firms would be none the wiser: there are no public indications of what use cases or pain points they should address.
The program was always going to involve hard tradeoffs. To maximize impact, it should support deals that would not otherwise happen without government support—but those are the deals most likely to be white elephants with no sustainable business case. To win contested markets, it needs to compete where China is a credible alternative—but those are precisely the markets where governments most want to hedge between Washington and Beijing. A serious program would confront those tradeoffs in its design. The current one ducks them by handing the core strategic questions to whichever consortia turn up at the door.
The Trump administration can still fix this. It needs to pick the markets that matter, take partners’ sovereignty demands seriously, and show up with a real offer that speaks to local economic needs before China does. In other words, it needs to use the program for economic statecraft—not for green-lighting deals industry would pursue anyway.
