“Why are we putting data centers and research hubs in Dubai?” U.S. Rep. Ro Khanna in a video posted on X in May. “We should have those high-paying new technology jobs in the United States. What happened to ‘America First’?” U.S. President Donald Trump’s czar for artificial intelligence, David Sacks, fired back: “The deal has a matching investment provision, so UAE will fund the build-out of AI infrastructure in the U.S. at least as large and powerful as that in UAE. And in UAE, the vast majority of the compute will be owned & operated by American cloud companies, to serve the region as well as the Global South.”
Theirs was not a mere rhetorical spat. It centered on the outcome of Trump’s May trip to the Gulf: the U.S.-UAE AI Acceleration Partnership, anchored by an AI data center in Abu Dhabi. Stargate UAE, an infrastructure platform and the inaugural project in the OpenAI for Countries initiative, will be part of the data center, which will be one of the world’s largest. Developed in close coordination with the U.S. government and personally championed by Trump, the partnership entails a dual investment: a 1-gigawatt (GW) cluster in Abu Dhabi (with the first 200 megawatts expected to go live in 2026) and an Emirati commitment to expand U.S. Stargate infrastructure. Saudi Arabia is adopting a similar approach through HUMAIN, the Public Investment Fund’s flagship AI venture. Backed by deals with, among others, Nvidia and Qualcomm, HUMAIN is also designed as a Saudi platform with regional and global ambitions, targeting model training, inference, and AI services across emerging markets—all firmly embedded in the U.S. tech stack. Together, the UAE and Saudi Arabia, and the broader Gulf region, are positioning themselves as potential backends of AI for emerging markets across Asia and Africa, laying the groundwork for a U.S.-aligned model of AI partnerships that could, over time, outpace China in the global AI race.
As shown by Khanna and Sacks’ exchange, though, the deal—coupled with the rescission of the Biden‑era AI diffusion rule—has sparked sharp debate, even before its official rollout. The stakes include the question of which superpower will own the infrastructure for the most important technology of our time.
Critics in what might be called the “AI nonproliferation camp,” including Khanna, argue that Washington is effectively the United States’ strategic computing power to the Middle East. They argue that compute—the capacity to process data, powering everything from basic calculations to advanced AI systems—is the 21st-century equivalent of enriched uranium, a critical lever of national power that should be distributed cautiously, if at all. The United States controls roughly 75 percent of the world’s frontier‑class compute, powered largely by Nvidia GPUs and Arm‑based designs. Many in Washington believe that China, constrained by export controls, cannot soon assemble, let alone deploy, a competitive AI stack at scale. Those in the “nonproliferation” camp argue that winning the race to artificial general intelligence requires maintaining this advantage. They support a tiered global system, in which cutting-edge chips are granted only to trusted allies, while the rest of the world—including energy-rich Gulf states—is left on the outside.
Sacks and Sriram Krishnan, a senior White House advisor on AI, lead what might be called the “global diffusion of the U.S. AI stack” camp. They argue that the chokepoint in AI infrastructure is no longer chips but electricity. With U.S. data center demand set to consume up to 9 percent of the national grid by 2030, they note that the United States cannot build high‑voltage lines or substations fast enough to keep pace. Gulf partners, by contrast, can finance GW‑scale campuses as low as $0.03 per kilowatt-hour and bring them online in months. In Sacks’s view, tapping this surplus power is the only credible way to scale compute quickly enough to keep the United States in front of the global AI race.
This camp also contends that the technological gap between the United States and China is far narrower than AI nonproliferation advocates suggest. Huawei’s Ascend chips, paired with the DeepSeek model suite, provide Beijing with an end-to-end domestic stack that is already seeding pilot projects abroad. Huawei is approaching near parity with U.S. chips and China’s broader push for technological self-reliance, especially in sectors like automotive chips—is accelerating, with major EV makers like BYD and Geely preparing vehicles using 100% domestic chips by 2026 and mature-node chip capacity in China projected to reach nearly 40% of global supply by 2027, up from 31% in 2023. Even if these targets aren’t fully met, the direction of travel is clear, and the shift already carries global implications. This should prompt a serious rethinking of the current state of U.S.-China tech competition. And because this competition isn’t happening in a vacuum, other nations are already facing a choice between two stacks: the U.S. stack or the Chinese stack. In fact, that choice is unfolding now.
Malaysia’s flip-flopping over a Huawei-backed national AI cloud is a case in point. As Sacks said, “As I’ve been warning, the full Chinese stack is here. We rescinded the Biden Diffusion Rule just in time. The American AI stack needs to be unleashed to compete.” Once Ascend hardware and DeepSeek models spread across emerging markets, Sacks thinks they could erode the current U.S. compute edge and accelerate China’s momentum in the race for AI dominance. Even if Malaysia ultimately opts out of deploying the Chinese stack, the stack is already here and will spread. Based on China’s expanding digital footprint, a growing number of countries could potentially adopt its AI stack, including Bangladesh, Pakistan, Thailand, South Africa, Algeria, and Egypt. The Chinese stack has no shortage of potential customers. It’s only a matter of time.
For this reason, the U.S.-UAE and the US-Saudi partnerships shouldn’t be viewed as offshoring compute capacity—this is capacity that didn’t exist in the first place. The United States’ current compute infrastructure is already strained by severe domestic bottlenecks in energy and capital, which makes the UAE and Saudi Arabia necessary partners for expanding U.S. compute capacity. According to the U.S. Department of Energy (DOE), data center power demand in the United States is projected to rise sharply, from about 4.4% of total U.S. electricity consumption in 2023 to between 6.7% and 12% by 2028. Meeting this surge—requiring 50 to 60 GW of additional infrastructure for generative AI alone—is constrained by seven-year timelines for new power generation and grid connections, compounded by challenges in securing capital and land permits, as well as growing community resistance. In this context, the U.S.-UAE partnership and the U.S.-Saudi deals—and the UAE and Saudi Arabia’s commitment to financing the buildout of U.S. compute infrastructure—represent the most realistic path for Washington to scale up compute capacity in partnership with willing and capable nations.
If the question of compute capacity is ultimately one of energy—and if the U.S.-UAE partnership helps resolve that constraint—some critics still argue that such cooperation should be limited to nations that share the United States’ democratic values. This argument rests on the assumption that compute is too sensitive to share. In reality, however, compute is not a classified asset; it is a commodity. Like oil in the 20th century, the value lies not in secrecy but in capacity, scale, and reliable access. You don’t need to reveal the source code of an iPhone to sell it; the value lies in mass production, performance, and distribution. Compute works the same way. What matters is who can build and scale the infrastructure, not who hides it behind a firewall.
Through its partnership with the United States, the UAE, Saudi Arabia, and the broader Gulf could position itself as a backend compute service provider for countries, particularly across emerging markets, that lack the capacity to train and run AI systems but are eager to leverage them for national objectives. In many ways, that is the essence of sovereign AI.
If the United States turns away from willing partners like the UAE and Saudi Arabia—nations that have demonstrated a clear preference for the U.S. stack by working with U.S. companies and investing in U.S. AI infrastructure—out of suspicion or through the narrow lens of the democracy-versus-autocracy framework, Washington risks creating a vacuum that China will be quick to fill. Unlike in traditional areas of geopolitical competition, where Beijing has yet to prove that it can replace the United States as a global security provider, technology is a different story. In this area, China has outpaced Western firms, building a global network that offers its stack with no strings attached. Chinese 5G hardware, renewable technologies, electric vehicles, and other mass-produced technologies have already won the global race. The choice, then, is not between trusting autocracies or preserving democratic values; it is between leading the global diffusion of U.S. AI infrastructure or standing idle as the Chinese stack becomes the default.