The Price of Sovereignty

How the Gulf is learning that building an AI power requires more than money

There is an odd scene taking shape in the Gulf. Across Abu Dhabi and Riyadh, some of the world's most ambitious artificial-intelligence infrastructure projects are moving ahead at a pace that would have seemed excessive even a few years ago. Gigawatts of computing capacity are being planned, billions of dollars are being committed and global technology companies are increasingly treating the region not as a promising market on the edge of the AI economy, but as one of its future centres. At the same time, the assumptions underneath that buildout are changing.

The clearest illustration may be sitting in Abu Dhabi. G42 is at the centre of Stargate UAE, the planned 5-gigawatt AI campus being developed with OpenAI, Oracle, NVIDIA, SoftBank and Cisco. Yet the company is also reportedly discussing a structure that could give American investors majority control, or creating a new US entity, in order to preserve access to advanced AI chips beyond the current April 2027 authorisation window. No final decision has been made, and G42 has not confirmed the reported discussions. But the fact that such a structure is being considered tells us something important about the economics of frontier AI.

The Gulf can finance the data centre. It can provide the land, power and infrastructure. It can attract the world's largest technology companies. What it cannot simply purchase with capital is unrestricted access to every component of the technology stack. The most advanced accelerators remain subject to American export rules, and those rules increasingly reach beyond the question of who buys a chip. They can influence where that chip goes, who controls the company using it and how the resulting computing capacity is governed. In other words, one of the most consequential assets in the AI economy is beginning to sit partly outside the balance sheet.

That is a significant change from the way the Gulf's AI ambitions were initially understood.

The basic proposition was compelling. The region possessed something the early AI industry desperately needed: enormous pools of capital, abundant energy, large-scale infrastructure capabilities and governments able to think in decades rather than quarterly earnings. If advanced computing was becoming the foundation of the next economic cycle, there was little reason for the Gulf to remain merely a customer of that economy. It could become one of its builders.

And it is doing precisely that.

The mistake would be to interpret the latest developments as evidence that this strategy is failing. In many ways, the opposite is happening. The strategy is becoming more sophisticated because the technology itself has become more strategic. Once AI infrastructure reaches a certain scale, the question is no longer simply how much money is available. It becomes a question of how capital, technology, regulation and ownership fit together.

G42 is a particularly revealing example because its relationship with Washington has already evolved considerably. Microsoft invested $1.5 billion in the company in 2024, took a board seat and became a major technology partner. G42 also moved away from Chinese hardware suppliers as it sought to reassure US authorities that its infrastructure could be trusted with American technology. Now, according to reports, the company is considering another structural step to secure longer-term access to advanced chips.

The interesting part is not that an Abu Dhabi company might bring in more American capital. International capital has been flowing into Gulf technology for years. The interesting part is that ownership itself is becoming part of the conversation about technological access.

That is a subtle but important development. Export controls traditionally appeared as a problem for procurement teams, lawyers and compliance officers. Frontier AI is turning them into a boardroom question. A company's ownership, jurisdiction and relationships can become strategically relevant because they determine which technologies it can acquire and under what conditions. The cap table, once a relatively private financial document, is beginning to tell part of the geopolitical story.

Saudi Arabia is approaching the same problem from a different direction.

HUMAIN was created by the Public Investment Fund as a full-stack AI company, with ambitions extending from data centres and cloud infrastructure to models and applications. It was never designed as a small software company hoping to find a profitable niche. The intention was to build a substantial domestic AI ecosystem around which other businesses could develop. That ambition requires infrastructure on a scale that makes conventional venture financing look almost decorative.

Now HUMAIN is preparing for an IPO and seeking $2.5 billion from local and international investors for another wave of data-centre development. The initial fund would support around 250 megawatts of capacity, with the project potentially expanding to 1 gigawatt. CEO Tareq Amin has previously discussed a possible dual listing in Saudi Arabia and New York by 2029.

Again, the important development is not simply the amount of money.

It is the change in the composition of the money.

State capital is exceptionally good at starting something that private markets are not yet prepared to finance. It can absorb long time horizons, tolerate uncertainty and build infrastructure before demand has fully materialised. But once an industry begins to mature, bringing institutional and private investors into the structure introduces another discipline: capital has to be allocated against expected returns, customers have to be found and projects have to be sequenced carefully.

That is not a retreat from Saudi Arabia's AI strategy. It is a natural evolution of it. The state can remain the strategic anchor while outside capital helps expand the system. In fact, that combination may ultimately prove more powerful than either approach on its own.

The more interesting question is what kind of technological system that capital is financing.

Last week, HUMAIN unveiled humain-m3, an Arabic-language frontier model developed by China's MiniMax and commissioned by HUMAIN. The model is based on the MiniMax-M3 lineage and was further trained on more than one trillion tokens of Arabic-native content.

That single detail complicates the simplistic version of the Gulf's AI story.

The region is not choosing between an American AI stack and a Chinese one. It is assembling something more selective. American companies remain essential to the most advanced computing infrastructure and semiconductor supply chain. Chinese companies can offer capable model architectures and open technologies. Gulf institutions bring capital, local demand, data, energy and the ability to build infrastructure at extraordinary scale.

There is a logic to combining those advantages where possible.

The result is not technological neutrality in the abstract. It is diversification in practice.

Whether that balance remains comfortable as AI becomes more deeply connected to national security is another matter. The United States may distinguish between hardware and software, between an open model and the chips required to train it at enormous scale, between a commercial application and infrastructure capable of supporting frontier systems. Those distinctions may become increasingly important as Gulf AI companies grow.

For the moment, however, the region has little reason to choose a single technological camp if it can build relationships with several.

That pragmatism is one of the strengths of the Gulf's position. It is also one of the reasons the region's AI strategy is becoming harder to understand through the usual geopolitical binaries.

There is a similarly practical reality on the physical side.

The damage to AWS infrastructure in the UAE and Bahrain during the regional conflict was a reminder that the cloud is not actually in the sky. AWS reported that two of its UAE facilities were directly struck by drone attacks, while a nearby strike caused physical impacts at a Bahrain facility. The incidents disrupted power and caused structural damage, with some fire-suppression efforts producing additional water damage. AWS said two of the three Availability Zones in its UAE region were significantly impaired.

For an industry accustomed to discussing AI in terms of models, parameters and benchmarks, the lesson was almost stubbornly physical.

Compute needs buildings.

Buildings need electricity.

Electricity needs infrastructure.

Infrastructure needs protection.

And when enormous quantities of computing power are concentrated in a small number of facilities, resilience becomes part of the economics of the business.

That does not make the Gulf a poor place to build AI infrastructure. It simply means the old calculation becomes more complicated. Redundancy, backup power, network diversity, geographic distribution and physical security all have a price. Insurance and business-continuity planning become more important. A facility that looks inexpensive on a spreadsheet can become considerably more expensive once the system around it is designed to keep operating under stress.

Yet the projects keep moving.

That is perhaps the most important fact of all.

The region's governments are not treating these risks as a reason to step away from AI. They are treating them as engineering and investment problems that have to be incorporated into the strategy. The objective remains much larger than any individual data centre. For Saudi Arabia and the UAE, advanced computing is increasingly part of the broader effort to build economies in which capital, technology, energy and human talent reinforce one another long after hydrocarbons cease to occupy the same economic position they do today.

The question, then, is not whether the Gulf can become an AI power. It almost certainly can.

The more difficult question is what kind of AI power it will become.

There was once a straightforward version of technological sovereignty: own the infrastructure, buy the machines and build the capability yourself. Frontier AI is making that definition less useful. No serious AI economy is completely self-contained. Semiconductor manufacturing, cloud platforms, advanced networking, software frameworks and research talent are spread across multiple countries and companies.

Sovereignty therefore becomes less about owning every component and more about controlling the critical dependencies.

That may mean ensuring access to several chip suppliers rather than one. It may mean developing local models rather than relying entirely on foreign ones. It may mean keeping strategic infrastructure under regional control while bringing in foreign capital where it provides access to technology and markets. It may mean building enough domestic capability that no single external partner can determine the fate of the entire ecosystem.

This is why the ownership discussions around G42 and the financing strategy around HUMAIN deserve more attention than another headline announcing a multibillion-dollar AI commitment.

They reveal how the Gulf is adapting its strategy to the actual structure of the AI economy.

G42 may need deeper American integration to secure access to American technology. HUMAIN is bringing external investors into a project anchored by Saudi sovereign capital while simultaneously using technology developed by a Chinese AI company. The region's infrastructure is becoming more international even as the strategic objective remains firmly regional.

That is not a contradiction to be resolved.

It is the model being built.

For investors, this changes what deserves attention. The headline figure attached to the next AI campus will still be impressive, and there will be plenty of them. But the more revealing details will sit elsewhere: who owns the operating company, who supplies the critical hardware, who finances the next phase, where the models come from, where the data is held, how much capacity is actually contracted and how resilient the infrastructure is when conditions become difficult.

The biggest number on the press release may not be the most important one.

The cap table might be.

The Gulf's AI strategy was never really about buying a collection of expensive machines. It was about converting capital, energy and infrastructure into a position in the next economic system. The closer the region gets to that objective, the more complicated the construction becomes. Access to advanced technology brings regulatory conditions. International capital brings new forms of governance. Greater computing density brings greater physical exposure. Diversification brings more complicated relationships between technology blocs.

None of that diminishes the ambition.

It clarifies it.

The Gulf is not trying to become technologically independent by withdrawing from the global technology system. It is trying to become powerful enough within that system to retain strategic room to manoeuvre.

That may prove to be a more realistic definition of sovereignty in the AI era.

The first phase of the Gulf's AI race was measured in money, chips and gigawatts.

The next phase will be measured in something less visible: who controls the relationships between them.

That is the less glamorous part of the story, but probably the part that will decide how much freedom the Gulf actually has when the next chip restriction, financing round or regional crisis arrives.