On May 12, the Trump administration announced plans to rescind the AI diffusion rule, a sweeping global framework that sought to govern the export of AI chips and model weights worldwide. The rule, published in the last weeks of the previous administration, was abandoned just days before its implementation deadline.
The administration followed this decision with a series of major chip export deals during President Donald Trump’s Gulf tour of Saudi Arabia, Qatar, and the UAE. Much about these deals remains unknown, with the administration yet to disclose — or perhaps even determine — crucial details. But the scale of some of the announced investments, if taken at face value, makes clear that they will dwarf most other AI projects worldwide.
These deals have prompted both criticism and support. For critics, the deals represent a decision to “outsourc[e] to the highest bidder the most critical infrastructure of the 21st century” — a move that will cannibalize domestic U.S. data center construction and offshore U.S. AI infrastructure to authoritarian regimes whose interests only partially align with those of the United States. For supporters, by contrast, the deals have opened “a trillion‑dollar floodgate of capital for AI infrastructure,” tied Abu Dhabi and Riyadh to the U.S. AI stack, and pulled key swing states out of the “arms of China.”
It is too early to pass final judgment one way or the other, as much will depend on implementation. But beyond the details of the deals themselves, key questions remain. How will the rest of the world, particularly China and Europe, react to and learn from the Gulf’s strategy? How will these deals reshape the politics and business of AI back in the United States? And what will global U.S. AI policy look like beyond the Gulf?
The Future of the Deals
According to media reports, the United States will allow the United Arab Emirates to buy the equivalent of 500,000 Nvidia AI chips each year. The Emirati AI champion G42, in partnership with several U.S. companies, will build a 5 gigawatt (GW) AI campus in Abu Dhabi that could ultimately house up to 2.5 million B200s, Nvidia’s most advanced AI chip — more than every other major AI infrastructure project announced so far, including OpenAI’s U.S.-based Stargate project. (By comparison, Elon Musk’s Tennessee-based Colossus, for now the world’s largest AI cluster, currently has 200,000 high-end chips on site.) Saudi Arabia, meanwhile, will gain access to around 1 GW of AI computing power, with an initial order of 18,000 Nvidia chips, as part of a broader $600 billion deal with the United States.
While the scale of ambition is clear from the bare bones press releases, many crucial details remain to be announced — or, perhaps, decided. Reporting suggests that G42 will receive one fifth of the UAE’s 500,000 chip export cap each year, with the remainder going to U.S. companies operating data centers in the 5 GW cluster. It is unclear, however, exactly how those chips will be divvied up, what U.S. companies must do to secure their slice, or what role G42 and other local companies will play in managing or accessing the chips they do not own. It is also unclear whether the project will truly involve a single 5 GW chip cluster, as the initial announcements suggest, or whether U.S. officials or companies will push to break it into smaller pieces. Companies may worry that a single campus increases the risk of commercial espionage, or they may find that concentrating so much power in one place is trickier than they expect. The U.S. government, meanwhile, may fear a single national project coming to aggregate all the chips on site. This fear will be especially salient if the deal allows G42 to own the entire cluster while giving U.S. companies the exclusive right to rent their portion.
More broadly, because these deals will play out over several years, and involve huge sums of money and high strategic stakes, a key question is how leverage will shift over time. Each side will have options if the facts on the ground change. If the United States comes to believe that Gulf states have violated deal terms, for example, it could retract or limit export licenses. Likewise, if the Gulf states conclude that the United States isn’t keeping up its end of the bargain, or if AI becomes so strategically important that they believe they must exercise more control over it, they could freeze payments, re-engage with Chinese competitors, or even try to seize control of the chips. The headline figures of both chip deliveries and Gulf investments in the United States may also change if the economics of the AI industry do not play out as companies hope, or if the region’s countries face budget shortfalls thanks to declining oil prices or regional instability.
It is too soon to say if the deals are sufficiently well-balanced to stick. It is not yet clear, for example, what governance rights the U.S. government will retain over how the chips are used, what security and compliance information it will receive, what physical and cyber security conditions it will attach to the chips, who will be able to access the data centers remotely, or what penalties the United States will be able to impose for violations. Some commentators have proposed models such as “data embassies,” under which data stored in these data centers would be subject to U.S. jurisdiction, but it is unknown whether U.S. officials have adopted any of these approaches, or whether local partners using the data centers would agree to them.
The deals have reportedly split the administration: China hawks are skeptical of the scale, speed, and apparently lightweight nature of the deals, while officials closer to the technology industry seem more likely to believe that China is close behind and ready to offer alternatives should Washington restrict exports. It is possible that, having lost the first round, the hawks and those instinctively opposed to industrial offshoring hope to slow-walk license approvals, add further security and monitoring conditions, limit the future role of foreign companies, or even re-open entire deals should conditions change.
How the Rest of the World Reacts
These deals mark the emergence of a new powerhouse in the AI race. As an unnamed administration official indicated, the G42 deal could mean that “the most powerful A.I. training facility in 2029 would be in the United Arab Emirates, rather than the United States.” Countries around the world will now have to decide how to react. The UAE and Saudi Arabia have been singularly focused on the potential of AI to transform their oil-dependent economies, but other states have AI ambitions, too, and now need to adapt to the reality of Gulf influence.
China
The Trump administration has partly justified its dealmaking by claiming that if the United States does not sell AI chips to the Middle East, China will. An open question, therefore, is whether Chinese companies, led by Huawei, will try to strike similar deals with other countries.
In the short term, at least, that will likely be difficult. China does not appear able to export chips in anything like the quantities the United States has promised to the Gulf. In May, Malaysia announced plans for the first deployment of Huawei AI chips outside China, but the deal involved just 3,000 chips, a fraction of the number the United States has promised to the UAE (and each of these chips is less advanced than Nvidia’s latest). Public reporting suggests the Chinese company SMIC produced about 200,000 Huawei Ascend 910B AI chips in 2024, and this year will produce about 300,000 910Bs and 100,000 910C chips, the successor to the 910B. Even if those projections are underestimates — because SMIC reallocates production from smartphone processors to AI chips, for example, or the company has more semiconductor fabrication plants than public reporting suggests — production will still likely fall short of China’s domestic demand, let alone the levels needed for major exports.
There are few signs that China will be able to supply a full-scale alternative to U.S. AI hardware in the near future. U.S. controls on semiconductor manufacturing equipment, if regularly updated and enforced, should restrict SMIC’s volume production at the leading edge. But Beijing may try to bring countries onside through smaller deals like the planned effort in Malaysia. Chinese policymakers may hope that by encouraging the use of Chinese models like DeepSeek on even small clusters of Huawei chips, they will begin to lock users in other countries into the Chinese tech stack. Significantly larger plans, moreover, would be important signs that China is overcoming U.S. restrictions.
China will also have to decide how to approach its broader relationships with Saudi Arabia and the UAE. Both countries have deep economic ties with China and will not want to alienate one of their largest trading partners. In fact, many of the funds the Gulf states have promised to invest in U.S. projects come from sales of Gulf oil to China. How have Saudi and Emirati diplomats sold the deal to their Chinese counterparts? Will Beijing attempt to extract concessions or commitments in other areas to help them preserve their swing state status? For now, China has been notably restrained in its public reaction; in the past, it has struck a very different tone when other countries have broken with Chinese tech companies.
Middle Powers
Major African, Asian, and Latin American countries will also have to decide how to respond. By and large, these countries do not have the same quasi-ideological belief in the future of AI, but they nevertheless want a piece of the benefits. India, in particular, has ambitions of becoming a global player in AI, although for now it lags far behind the frontier.
For some countries, such as those in Southeast Asia, the Gulf strategy may look attractive: declare yourself a geopolitical swing state by making overtures to both the United States and China, free up capital and energy for AI data centers, and offer U.S. companies an escape from regulatory roadblocks at home. Some American companies could distinguish themselves by avoiding deals with authoritarian states and working with democratic allies and U.S. partners in the Global South like Brazil, India, or Mexico. Because those countries lack the sovereign wealth funds and authoritarian political systems that allow the Gulf states to rapidly prioritize AI development, they will have to find other ways to develop attractive offers to U.S. industry and policymakers. And there are risks to playing the swing state strategy: while it may attract more generous concessions from U.S. policymakers eager to shore up influence worldwide, it may also lead to increased pressure from Washington to pick a side.
Europe
For Europe, in particular, the rise of the Gulf poses a major challenge. European leaders see their continent as a third economic and technological power between the United States and China. Yet the development of frontier-size data centers in the UAE throws a harsh light on Europe’s failure to build AI infrastructure at anything like the same scale. One of the largest announced European projects is a planned collaboration between the UAE tech investment arm MGX and the French AI developer Mistral to build a 1.4 GW data center in France. The United Kingdom is considering establishing “AI growth zones” that would streamline permitting and access to power. But none of these projects come close to the planned 5 GW in the UAE. And amid rising transatlantic tensions and shifting alliance dynamics, Europe should not necessarily assume that it will always enjoy unfettered access to U.S. AI chips.
How the Deals Reshape the U.S. AI Industry
At home, meanwhile, the Gulf deals may have an unintended consequence: they might ease political pressure to fix the problems that make it hard to build AI infrastructure in the United States. If U.S. companies can run their operations in the Gulf, they will spend less time pushing for permitting and energy reform at home — reducing the odds that the administration can build a world-beating U.S. AI ecosystem. The administration has several options to reduce this risk, including requiring major cloud providers to keep a large proportion of their AI chips in the United States or conditioning future tranches of the Gulf deals on equal progress building at home. Absent such measures, the default will likely be a race to the bottom.
These deals also mark the emergence of the AI labs as powerful diplomatic actors. The world’s most valuable companies have persuaded the U.S. government that its interests and theirs coincide in promoting the sale of valuable technology abroad and encouraging massive investment by foreign sovereign wealth funds. U.S. strategy in the Gulf and elsewhere will reflect, in part, the view these companies have of their future interests.
Although champions of the deal have declared that “All American infrastructure suppliers are set to benefit,” the reality is more complicated. Despite their existing partnerships with Gulf companies, the traditional hyperscalers — Amazon Web Services, Google, and Microsoft — do not appear to be the biggest players in the initial buildout in the UAE. They will need to decide when and how they want to invest in a section of the new chip cluster. (It is not yet clear how or under what conditions U.S. companies will be able to launch projects within the framework.) Elon Musk reportedly opposed the UAE deal because, in his view, it favored OpenAI over other AI companies, including his own firm, xAI. Another of OpenAI’s rivals, Anthropic, publicly supported the Biden administration’s diffusion rule and has called for keeping chips in democratic countries. U.S. companies will surely adapt to the new landscape — xAI is reportedly a likely candidate for future allocations in the UAE cluster — but the deals are likely to help some more than others.
More broadly, the change in administration has produced a shift in industry strategy. Under the Biden administration, traditional tech giants like Microsoft with extensive experience in compliance took the lead in partnering with Gulf companies. The Microsoft-G42 deal, although it did not formally involve the U.S. government as a party, took place with the public blessing of senior U.S. officials. The negotiations emphasized security commitments and verifiable progress, reflecting the administration’s preference for cautious progress toward bringing the UAE into the U.S. technology ecosystem.
Under the Trump administration, another model of engagement — one that emphasizes headline-grabbing projects — has enjoyed more success. The first phase of the U.S.-UAE partnership, for example, is being led by “Stargate UAE,” an ambitious collaboration between G42, Nvidia, OpenAI, Oracle, Cisco, and SoftBank to build a world-leading chip cluster. Still, these companies should not get too complacent: the rapid nature of the deals makes it unclear whether they are robust enough to survive a change in partisan control of Congress and the White House.
The Future of U.S. Strategy Elsewhere
Finally, the administration will need to decide whether its approach in the Gulf is a one-off, or whether it plans to make similar offers to other countries. In order to replicate the leverage the United States enjoyed in the Gulf deals thanks to the preexisting U.S. ban on chip exports to the Middle East, policymakers may choose to impose license requirements elsewhere in their replacement for the AI diffusion framework. Even if the administration does not impose formal restrictions, U.S. companies may be reluctant to engage in major deals with non-allies without at least the unofficial blessing of the U.S. government.
Yet whether the deals are replicable elsewhere remains unclear. Policymakers may want to bring in U.S. partners feeling left out in the cold by the bonanza in the Gulf, encourage the development of rival centers so that the UAE and Saudi Arabia are not the only major AI powers outside the United States and China, and promote the use of the U.S. AI stack around the world. Yet officials may also conclude that no other country can offer a comparable partnership, if they believe that the Gulf is unique in its regulatory freedom, ability to deploy capital and energy, and governmental focus on AI development.
The administration is currently working on its overall strategy for global AI export policy. Bilateral dealmaking in which chips are used as leverage in broader negotiations will likely play a bigger role than in past administrations. But several key questions remain, including whether the baseline will be one of default openness or restriction until a deal is reached, the status of more traditional U.S. allies, and the extent of any exceptions to license requirements.
The stakes are high. Because the United States remains, for now, the only game in town when it comes to large-scale AI hardware, the answers to those questions will have major implications for economics and geopolitics worldwide. The Gulf deals represent the administration’s first answer to the question of how widely the United States should share its AI technology. They will not be its last.
