Silicon Valley’s Bad Bet on the Gulf – Foreign Affairs

Since its founding in 1922, Foreign Affairs has been the leading forum for serious discussion of American foreign policy and global affairs. The magazine has featured contributions from many leading international affairs experts.
AARON BARTNICK is a Nonresident Fellow at the Columbia Center on Global Energy Policy. From 2024 to 2025, he served as Assistant Director for Technology Security and Governance at the White House Office of Science and Technology Policy.
Why the AI Build-Out Was Doomed From the Start
Aaron Bartnick
When President Donald Trump returned from a trip to the Gulf in May 2025, he trumpeted $2.2 trillion in bilateral deals the United States had signed with Qatar, Saudi Arabia, and the United Arab Emirates. In addition to defense and economic partnerships, a significant share of those deals addressed artificial intelligence—and, specifically, building American AI infrastructure in the Gulf. The proposition for American technology companies was compelling: abundant cheap energy, access to capital from sovereign wealth funds, lower regulatory barriers for data center construction, and U.S. government approval for previously restricted chip sales. For Washington, the deals offered an opportunity to speed up advancements in American AI capabilities and incentivize the Gulf to partner less with China on AI.
But recent events have revealed a major risk: the infrastructure that such critical technology relies on is under attack, caught in the crossfire of a war it was not prepared for. On March 1, Iran hit two Amazon Web Services data centers in the UAE with a barrage of Shahed drones; an AWS facility in Bahrain was also damaged, likely with debris from a nearby strike. The Islamic Revolutionary Guard Corps claimed the facilities supported U.S. military AI operations, which could make them legitimate targets under the laws of armed conflict. (Although the U.S. military uses AWS, there is no publicly available evidence that these particular facilities supported U.S. military operations, as the IRGC claimed.) The effective closure of the Strait of Hormuz and disruption in the Red Sea also put at risk the submarine cables that carry the vast majority of data traffic between Asia, Europe, and the Gulf.
These campaigns are inflicting serious short-term costs on the users of those networks, both civilian and military. But the bigger problem is that investment in the Gulf is not limited to new bilateral technology partnerships between the United States and a handful of Gulf monarchies. It has also created a web of digital dependencies that now extends to dozens of countries that had no say in the arrangement and no awareness of the risk they were assuming.
The U.S. bet on the Gulf for AI infrastructure was a mistake. By encouraging U.S. companies to build vulnerable and immobile critical infrastructure in the Gulf, Washington gambled on an unstable security environment—often a result, at least in part, of U.S. intervention, including the latest war with Iran. Once digital infrastructure investments are made, the dependencies that develop are hard to reverse: multibillion-dollar physical infrastructure projects cannot simply be turned off, nor can their capital be easily redistributed, particularly when they serve as the foundation for what has become one of the world’s most valuable technologies. The costs that companies, consumers, and U.S. national security interests are now facing are likely to persist, or even rise further. Some projects have already reached this point of no return, and companies will likely have to eat those costs. For investments that are early in the building and planning process, however, tech companies should reconsider their pivot to the Gulf and work with Washington to start bringing projects back to where it is safest and, as energy and other costs rise, increasingly cost competitive—the United States.
The Gulf hosts roughly 1.5 percent of the world’s computing capacity—about double its share of global population. This may seem small, but its capacity is growing by some nine percent per year. The ambitious deals that resulted from Trump’s 2025 Gulf tour are a significant part of this rapid growth. AMD, an American chip company, announced a $10 billion partnership with Humain, the Saudi sovereign wealth fund’s AI enterprise. The U.S. chipmaker Nvidia committed to shipping 18,000 cutting-edge chips to power Saudi Arabia’s AI infrastructure. One consortium of American companies began planning an AI campus with the Emirati AI firm G42 under the umbrella of the U.S.-UAE AI Acceleration Partnership that was signed on the trip. Submarine cables in the Red Sea carry an estimated 90 percent of data traffic between Europe and Asia and about 17 percent of all global Internet traffic.
When Amazon’s data centers in the UAE went offline, the damage spread across the Gulf’s digital economy, knocking out major banks, payment apps, ride-hailing and delivery services, enterprise software providers, and some operations of at least one large American data company, Snowflake. But the effects of the attack were not limited to the Middle East. Many affected companies did not operate in the region at all but used cloud providers who routed their workloads through Gulf-based servers. And some may not have even been aware that their data went through the Gulf until those servers failed.
The framework for the Gulf AI build-out was not designed for such hazards. Much of it was guided by the Pax Silica initiative, a U.S.-led effort to secure AI supply chains. Since the initiative’s creation in December, Qatar, the UAE, and 12 other countries in Europe and Asia have signed on to the agreement. But its proposed mechanisms—including joint ventures, coinvestment, and pooled buying power to secure more favorable deals—were designed for managing peacetime technology competition with China, not for defending the physical supply chain from missiles and drones. They addressed important concerns related to competition, including measures to prevent countries from diverting advanced chips to U.S. adversaries; to vet foreign ownership structures, such as G42’s former ties to Huawei; and to establish reporting and monitoring requirements for U.S. hardware deployed overseas. But beyond a vague platitude advocating efforts to “further strengthen economic and national security cooperation,” it had little to say about what happens when the infrastructure itself becomes a military target.
The challenge of a commercial data center becoming a lawful military objective was foreseeable. Legal scholars have debated the possibility for years. Submarine cables have also been targets in the Baltic Sea, around Taiwan, and even in the Red Sea, where fighting around Yemen in 2024 damaged three cables. What was new in the March AWS attack was only the method: missile and drone strikes on data centers rather than anchors dragging across the ocean floor.
The United States has spent decades building frameworks to protect domestic critical infrastructure from exactly this kind of threat, including by designating certain infrastructure sectors as critical and requiring physical security standards for the electric grid. But for some reason—perhaps lulled into a false sense of regional security, or out of a desire to not offend its negotiating partners—it did not apply that same thinking to the infrastructure it was encouraging American companies to build abroad.
For many companies, the math that made the Gulf compelling in May 2025 no longer holds. There is no realistic way to protect a commercial data center against cruise missiles or drone swarms. Retrofitting a data center as if it were a military installation would be both prohibitively expensive and ultimately insufficient. As years of war between Russia and Ukraine have proved, no air defense can intercept 100 percent of incoming strikes, and to deliver its commercial value, AI infrastructure must run almost constantly.
The most pressing issue for firms is legal liability. The means by which many companies have justified their stoppage of operations, and thus limited their liability, are notices of force majeure, which effectively excuse a company from fulfilling its expected operations because of an event beyond its control. But courts and legal commentators have signaled that such defenses might not stand in the Middle East, which has a well-documented history of political instability and decades of conflict. Companies that entered contracts during or after a period of obvious tension may find that insurers or courts will treat military disruption as a foreseeable risk. The downstream liability chain is long: every enterprise customer whose operations were interrupted by the AWS outages is now assessing its own contractual position, and where there is no specific language to the contrary, cloud providers may find themselves required to absorb losses and compensate clients for disruptions in service.
Those AI companies with projects under construction are thus reassessing whether such projects are still viable. Costs for new data centers in the Gulf are rising significantly because they need to add more backup systems, enhance physical security measures, and pay higher insurance premiums. When the 12 members of the International Group of P&I Clubs—a group of mutual insurance associations that provide marine liability cover for 87 percent of the world’s oceangoing tonnage—announced the cancellation of their maritime war risk cover for the Gulf, it signaled to the broader insurance market that the region’s risk profile had fundamentally changed. For some Gulf projects, the revised all-in cost now exceeds what it would have cost to build equivalent capacity in places where energy tends to be more expensive, including northern Europe, India, or the southern or western United States.

Those still in the planning phase still might choose to continue building AI infrastructure in the Gulf. The incentive packages from Gulf governments to build remain attractive on paper. But many are likely to choose to cut their losses and go elsewhere. Any honest appraisal now requires pricing in war risk insurance, the rates for which have roughly quintupled since the war began, as well as security costs, force majeure liability exposure, the reputational consequences of operating in an active conflict zone, and the possibility that hostilities could resume at any moment.
Companies that are already operational in the Gulf have no choice but to spread their risk. Amazon, Microsoft, Google, and Oracle—the four firms that dominate the market—each run at least two data center clusters in the region, hosting wide-ranging operations such as regional banking systems and mobile apps for consumers. They will all need to migrate particularly sensitive workloads to facilities in other regions and augment insurance coverage for the work that remains. They must also increase transparency about their Gulf operations. Many of their clients do not know that their workloads route through Gulf data centers, which risks anger or even lawsuits in the event of future outages. Companies will be most protected if they publicly separate military and commercial data and projects to the greatest extent possible and locate truly sensitive operations outside the Middle East.
It’s possible that none of the Gulf-based digital infrastructure is used specifically for military operations. But as long as the U.S. military continues operating in the Middle East, U.S. infrastructure will be a target—especially as commercial AI companies take on more defense-related functions. This is a problem not only for the United States but also for its partners. The prospect of becoming a global AI hub promised new jobs, high returns for Gulf sovereign wealth funds, and technological prestige. It also made Gulf states higher-value targets for any adversary of the United States. As long as American AI cloud service providers operate facilities in their territory that may serve U.S. government functions, the distinction between civilian and military infrastructure will be blurred in the eyes of adversaries. Any future contracts in the region will need provisions that explicitly allocate liability in the event of state-sponsored military action—elements that prewar template agreements do not sufficiently address. Companies are already diversifying their geographic concentration, shifting projects to northern Europe, Southeast Asia, and India.
The U.S. government’s active encouragement of the Gulf AI build-out through bilateral investment deals, chip export approvals, and the Pax Silica framework was a major miscalculation, especially in light of its subsequent actions in Iran. The set of interdependencies that resulted from its actions has the potential to harm both U.S. economic and national security interests. Even if Washington were serious about retrenching its military from the Middle East—which it has not yet demonstrated—the region’s persistent instability makes it ill-suited to host the technology most critical to economic growth and great-power competition. Technology partnership frameworks need to account not only for supply chain integrity and economic upside but also for the physical security of that critical infrastructure and the potentially catastrophic costs of damage to it.
The solution is not to retreat from the Pax Silica agreement, nor to let it quietly wither. The chip deals and Pax Silica framework were meant to pull the Gulf into the American technology orbit and away from China; backing out could have the reverse effect by damaging important relations with Gulf countries and emboldening Beijing. The U.S. intervention in Iran has presented China with an opening: companies such as Huawei are already pitching Gulf clients on the perils of “single-region dependency.”
Instead, the United States should focus on reinforcing and building on AI cooperation in more advantageous areas such as those geared toward pooling U.S. and Gulf states’ buying power for critical mineral inputs and for rebuilding manufacturing capacity on allied soil, all while unwinding the commitments on vulnerable critical infrastructure. Workloads can be migrated, future investment can be redirected, and projects still in the planning phase can be reconsidered entirely.
The window for making those decisions is closing. The bet that the current war is an aberration rather than a preview of the region’s future looks less wise by the day. The incentive packages for investing in the Gulf that were designed for peacetime are becoming irrelevant. The longer the United States and its allies wait to protect their critical infrastructure by moving it elsewhere and to redefine their AI partnerships, the more expensive, complicated, and ultimately dangerous it will get to do so.
Subscribe to Foreign Affairs to get unlimited access.
Already a subscriber? Sign In
How to Counter Beijing’s Unauthorized “Distillation”
Jared Dunnmon, Avanika Narayan, and Jon Saad-Falcon
How AI Agents Will Threaten Global Security
Brianna Rosen and Jam Kraprayoon
What It Will Take to Gain the Advantage Over China
Jake Sullivan
Cooperation Is Necessary—and Possible
Christina Knight and Scott Singer
And Why the U.S. Military Isn’t Ready for It
David Petraeus and Isaac C. Flanagan
How Weak Cyberdefenses Threaten U.S. Tech Dominance
Fred Heiding and Chris Inglis
America Depends on It—and Can Restore It
Philip H. Gordon
And Why the Open Seas May Soon Be Tolled
Isaac Kardon
What American Military Force Can and Cannot Do
Ricardo Zuniga
How a Remade Islamic Republic Will Reshape the Middle East
Narges Bajoghli and Vali Nasr
Russia’s Loss Is NATO’s Gain
Gonul Tol
Trump’s Overreach Has Finally Forged Continental Unity
Matthias Matthijs and Nathalie Tocci
Beijing’s Blind Spots Hinder Real Reform
Francesca Ghiretti
Our editors’ top picks, delivered free to your inbox every Friday.
* Note that when you provide your email address, the Foreign Affairs Privacy Policy and Terms of Use will apply to your newsletter subscription.
Published by The Council on Foreign Relations, Inc.
© 2026. All Rights Reserved.
From the publishers of  Foreign Affairs
We and third parties may collect data using cookies and similar technologies for site functionality and advertising. Choose “Accept All” to consent to optional cookies. Manage your preferences in our cookie consent form. Privacy Policy.
Our editors’ top picks from the week, delivered on Friday.
* Note that when you provide your email address, the Foreign Affairs Privacy Policy and Terms of Use will apply, and you will receive occasional marketing emails.

source