New Superpowers and Alliances

In May 2025, a president of the United States stood in Abu Dhabi and helped announce a building project that had no obvious American precedent. Not a factory, not a military base, not an embassy — a computer. A very large computer. Sprawled across roughly 26 square kilometers of Emirati desert, it was designed to eventually draw five gigawatts of electricity, enough to power a mid-sized country, all of it dedicated to training and running artificial intelligence. The land was Emirati. The capital was largely Emirati. The chips would be American, the models would be American, and the whole thing would be operated by an Emirati firm, G42, under terms negotiated between two governments and a handful of the most valuable companies on Earth.

Try to file that arrangement under a single category and it slips away. Is it a domestic Emirati project or an American one? Public or private? Commercial or strategic? It is all of them at once, and that refusal to fit the old boxes is the whole point. Something new is being built in the desert, and not just a data center.

The instinct, when we think about who controls AI, is to look at a two-horse race: the United States and China, Silicon Valley and Shenzhen, two giants slugging it out while everyone else watches. That picture is not wrong, but it is dangerously incomplete. Underneath the superpower rivalry, a more interesting story is unfolding — one in which a cluster of middle powers has discovered that you do not need to win the whole race to shape its outcome. You just need to own a stretch of track that everyone else has to run on.

The middle-power moment

Frontier AI is not one thing. It is a stack — a layered dependency chain running from raw materials up to finished applications. At the bottom sit energy and land. Above them, the chips: the advanced semiconductors that do the actual computation. Above the chips, the data centers that house them and the capital that pays for them. Then the training data, the models, the applications, and finally the governance frameworks that decide what any of it is allowed to do.

No country controls the whole stack. The United States dominates the top — frontier models, chip design, the leading labs — but it does not have enough energy, land, or capital to build all the infrastructure its own ambitions require. China excels at industrial-scale deployment but has been throttled at the chip layer by export controls. And this is the crack that middle powers have pried open. Because the stack is a chain, controlling even one critical link grants leverage over everything downstream. You do not have to be good at everything. You have to be indispensable at one thing.

The UAE understood this earlier and more clearly than anyone. Its bet was on the least glamorous layers imaginable: power, land, and money. AI in the 2020s turns out to be brutally physical. Training a frontier model means filling warehouses with hundreds of thousands of chips that run hot and drink electricity, and the binding constraint is increasingly not talent or algorithms but gigawatts. Goldman Sachs began calling this the "gigawatt ceiling" — the point at which AI ambition collides with the physical limits of the electrical grid. The UAE has cheap energy, empty land, a sovereign wealth apparatus willing to write ten-figure checks, and a government that can approve a 26-square-kilometer megaproject in the time it takes a Western utility to file an environmental review.

So Abu Dhabi built. Through its investment vehicle MGX and the AI firm G42, it committed to Stargate UAE, a one-gigawatt supercomputing complex whose first cluster was targeted to come online in 2026, developed alongside OpenAI, Oracle, Nvidia, Cisco, and SoftBank. Beyond that sits the five-gigawatt campus announced during the 2025 Gulf visit, the largest AI infrastructure project outside the United States. The strategy has a name worth remembering: not dominance, but indispensability. The UAE is not trying to out-research Silicon Valley or out-manufacture Shenzhen. It is trying to become the place where everyone else's AI physically runs.

Saudi Arabia is running a near-identical playbook with its own accent. In May 2025 the Public Investment Fund launched Humain, a state-owned AI company that within months had signed agreements to buy tens of thousands of Nvidia's most advanced chips, plus deals with AMD, Qualcomm, and Amazon Web Services. Like the UAE, Saudi Arabia is converting oil wealth into compute — trading capital and energy for technology access, in a bargain that neither side needs to pretend is a friendship.

India's leverage comes from a different quarter of the stack. It cannot match Gulf capital, but it has something the Gulf does not: people and data at civilizational scale. India's IndiaAI Mission has assembled a national pool on the order of 34,000 GPUs and subsidizes access to it, while the country's vast tech workforce — millions strong — supplies engineers to labs on every continent. Its population generates training data in dozens of languages at a volume few nations can rival, an increasingly structural advantage as models grow hungrier for diverse, real-world text. India's play is to be the talent-and-data layer, and to do so without picking a permanent side.

Israel's influence is narrow and deep. A country of under ten million hosts thousands of startups, a large share of them AI-driven, coordinated through a dedicated AI directorate in the Prime Minister's Office. But its real leverage sits in a specific, sensitive niche: defense and cybersecurity AI. Israeli firms supply AI-enabled surveillance, targeting, and cyber tools to militaries and intelligence agencies across the democratic world, giving a small state outsized purchase on some of the most consequential — and most ethically fraught — applications of the technology.

Then there are the specialists whose single node is simply irreplaceable. Taiwan fabricates the world's most advanced chips through TSMC, a concentration of capability with no near-term substitute and, not coincidentally, the most dangerous chokepoint in the global economy. South Korea and Japan dominate the memory chips, manufacturing equipment, and industrial robotics that the rest of the stack depends on. Singapore has turned neutrality itself into an asset, positioning as the trusted venue where firms and governments from rival camps can sign agreements they could not sign anywhere else.

None of these players will overtake the United States or China in overall capability. They do not need to. Each has found a link in the chain it can grip, and a grip on one link is felt all the way along.

Why nobody can go it alone

Behind this scramble for individual nodes lies a deeper shift in the logic of AI geopolitics: self-sufficiency is dead, and interdependence has replaced it. This is not an ideological preference. It is arithmetic.

The numbers involved in frontier AI have grown too large for any single actor to absorb. A leading-edge chip fabrication plant costs on the order of hundreds of billions of dollars and takes years to build. A hyperscale data center campus runs to tens of billions. The energy infrastructure to feed them — power plants, transmission lines, in some proposals dedicated nuclear reactors — represents decades of construction. Stack these together and the capital requirements of staying at the frontier exceed what most governments can raise and what even the largest companies can finance alone.

So the actors pool. The United States has the models and the chip designs but needs energy and capital; the Gulf has energy and capital but needs the models and chips. Taiwan makes the silicon but needs the security guarantees only Washington can offer. Each partner brings one thing the others cannot easily manufacture, and the combination produces something none could build in isolation. Specialization plus partnership beats autarky — not because cooperation is virtuous, but because the alternative is unaffordable.

graph LR
  A[US: models, chip design] --> E[Frontier AI system]
  B[Gulf: energy, capital] --> E
  C[Taiwan: chip fabrication] --> E
  D[India: talent, data] --> E
  E --> F[Capability no single actor could build alone]

This is what makes the UAE's indispensability strategy viable rather than delusional. Abu Dhabi is not offering something superpowers merely happen to lack today; it is offering something they cannot cheaply replicate. The United States could, in theory, build five gigawatts of data centers domestically — but its grid is strained, its permitting slow, its energy more expensive, and its politics allergic to the land and water such projects consume. The UAE offers speed, cheap power, patient sovereign money, and a studied geopolitical neutrality that lets a firm host American models without being drawn into every American quarrel. That bundle is genuinely hard to substitute, which is precisely why technology companies and governments line up to use it.

The alliances take shape

If interdependence is the logic, alliances are the structure it builds. And in 2025 and 2026 they proliferated at remarkable speed, each embodying a different theory of how AI power should be organized.

The bilateral deals came first and most bluntly transactional. In November 2025 the United States and Saudi Arabia signed a Strategic Artificial Intelligence Partnership spanning infrastructure, research, and deployment (U.S. Department of State, 2025). The trade was clear: Washington gained influence in the Gulf and access to Saudi capital and energy; Riyadh gained advanced chips, technology transfer, and alignment with the leading AI power. Neither side had to share values to share interests. The same pattern — infrastructure and capital flowing one way, technology and strategic alignment the other — recurs across nearly every Gulf-U.S. AI agreement.

Middle powers also began building their own networks that route around both superpowers. India's AI Impact Summit, hosted in New Delhi in February 2026, was the clearest signal (ORF, 2026). Rather than a communiqué of good intentions, it produced substantive bilateral deepening: India and Israel extended cooperation across AI, climate resilience, and digital governance; India and the UAE widened an already dense relationship spanning defense, space, energy, and technology. The message was that complementary middle powers, pooling their respective nodes, can construct alternatives to dependence on Washington or Beijing.

Security alliances quietly became AI alliances. The Quad — the United States, Japan, Australia, and India — and AUKUS — Australia, the United Kingdom, and the United States — both folded AI into their core agendas, coordinating on secure chip supply chains, shared compute, and cyber-threat intelligence. AUKUS's second pillar in particular targets AI-enabled military capability: autonomous undersea systems, electronic warfare, hypersonics, quantum sensing. What began as conventional defense pacts have become vehicles for aligning who gets access to which chips and which models.

At the governance layer, the democratic states worked to write shared rules. The G7's Hiroshima AI Process produced a code of conduct for advanced AI grounded in transparency, accountability, and the protection of individual rights — explicitly conceived as a counterweight to China's state-integrated model. The European Union pursued a parallel, broader track, engaging partners across every continent while leaning on the Brussels Effect: because firms selling into the EU's vast market must comply with its rules, EU standards tend to become de facto global ones, extending regulatory influence even where Europe cannot compete in raw compute.

And then there is the hybrid form that captures the era best: coalitions that fuse governments and corporations into a single actor. The Partnership for Global Inclusivity on AI assembled the U.S. State Department alongside Amazon, Anthropic, Google, IBM, Meta, Microsoft, Nvidia, and OpenAI (U.S. Department of State). Its stated aim is to widen AI access in developing countries; its structural effect is to project American influence through corporate channels, ensuring U.S. firms help set the terms on which AI arrives elsewhere. This is what Goldman Sachs meant in forecasting that 2026 would be defined by "mega alliances" — partnerships of unprecedented scope in which sovereign governments, technology giants, energy providers, and investment funds operate as one, blurring every line between public and private, domestic and international, commercial and strategic (Goldman Sachs, 2026). The Abu Dhabi campus is the archetype. It is not a deal between countries or between companies. It is both, simultaneously, and the distinction has stopped meaning much.

The costs of a fragmenting map

All of this pooling produces real capability. It also carries a bill, and the bill is fragmentation.

As alliances harden around rival visions, the global AI ecosystem is splitting into blocs that increasingly cannot interoperate. The American-led model favors market-driven, lightly regulated AI aligned with liberal-democratic norms. China's model fuses AI with state control and industrial policy. Europe's centers on rights-based regulation. The Gulf's prizes infrastructure neutrality over any ideology at all. These are not easily reconciled, and every hardening seam adds friction — technical incompatibilities, conflicting data-sovereignty laws, regulatory regimes that bar models trained under a rival's rules. Push the trajectory forward and you get a world where American and Chinese systems run in mutual isolation, data cannot cross borders, and infrastructure is duplicated wherever trust runs out.

The costs are concrete. Duplicated fabs and data centers waste staggering sums. Fragmented markets shrink the returns that fund the next round of research. And where incompatible systems meet in contested domains — financial networks, communications, the military uses of AI — the absence of shared standards raises the odds that a technical misunderstanding becomes a strategic one.

Whether this counts as catastrophe or merely friction is genuinely contested, and honesty requires holding both readings. The optimistic case is that competition among governance models is a feature, not a bug: it lets institutions experiment, prevents any single regulatory monoculture from freezing the field, and gives smaller states real choices about whom to align with rather than a single global regime imposed from above. The pessimistic case is a digital Cold War that throttles progress through wasteful duplication and multiplies the flashpoints where rival systems collide. The truth is that we do not yet know which dominates. Some fragmentation clearly spurs useful variety; beyond some threshold it becomes pure deadweight loss. Nobody can currently say where that threshold sits.

Fragmentation also excludes. As capability concentrates in a handful of mega alliances, access to those networks becomes the price of admission to the AI future — and the price is steep. It requires capital most states do not have, or geopolitical positioning most cannot achieve, or a critical node most will never control. Much of Africa, Latin America, and South and Southeast Asia risks being left as consumers of AI built elsewhere, under terms set elsewhere, with no seat where the standards are written. The Partnership for Global Inclusivity gestures at this problem, but a coalition led by the firms that profit from access is an uneasy guarantor of equitable access. The obligation that alliance-leading nations have toward the excluded — whether meaningful global participation can be preserved through open compute grants, shared standards bodies, or genuinely neutral hubs — remains one of the least resolved questions of the whole enterprise.

How solid is any of this?

A final, necessary skepticism. It is easy to narrate the middle-power moment as a settled new order. It is not settled, and two doubts deserve to sit at the front of the mind.

The first concerns the Gulf specifically. Does building data centers actually buy geopolitical influence, or does it just buy the business of hosting other people's AI? A landlord is not a superpower. If the UAE's leverage rests on renting compute to American firms running American models under American export licenses, then Washington retains the ability to revoke the chips, restrict the models, or attach conditions — and the UAE's "indispensability" turns out to be a convenience that can be withdrawn. The counterargument is that once enough of the world's AI physically depends on Gulf infrastructure, switching away becomes prohibitively costly, and dependence flows both directions. Both stories are plausible. We are early enough that the evidence does not yet decisively favor either, and anyone claiming certainty about the Gulf's strategic weight is running ahead of what we can actually know.

The second doubt concerns durability. Mega alliances are complex, multi-party arrangements built on sustained trust, aligned interests, and continuous coordination. That networked structure is resilient in one sense — if one partner falters, others can partly compensate, in a way no bilateral dependency can. But it is fragile in another: complexity means many simultaneous relationships, each of which can crack under a conflict, a sanction, a leadership change, or an economic shock. A single election, in Washington or anywhere else, can rewrite the terms of a decade-long partnership overnight. We simply have not watched these arrangements weather a real crisis yet. Whether they are durable strategic relationships or merely commercially convenient ones that would shatter under genuine geopolitical stress is, at this moment, unknown — and it is the kind of thing you only learn after the stress arrives.

The hedge that the UAE, Singapore, and India are all running — cultivating productive ties with rival blocs at once, refusing to choose — is the smartest available bet precisely because so much remains uncertain. Its viability depends on the very openness that fragmentation threatens. As long as the blocs stay porous and overlapping membership is tolerated, the hedgers thrive as the essential connective tissue between camps. Once loyalty starts being demanded over pragmatism, the middle ground narrows, and the countries betting on it will eventually be forced to pick. How long that window stays open is the question that will define the next decade of AI geopolitics — and the answer is not yet written. That, in the end, is the honest posture: the digital Cold War is a warning drawn from current trends, not a forecast locked in. The map is still being drawn, in Abu Dhabi's server halls and New Delhi's summit rooms and Brussels' regulatory chambers, and the hand holding the pen has not yet decided whether to draw walls or doors.

Summary

1. Middle powers are gaining real influence by owning single links in the AI stack, not by matching the superpowers overall. The UAE and Saudi Arabia control compute infrastructure through energy and sovereign capital; India controls talent and data at scale; Israel dominates defense and cybersecurity AI; Taiwan fabricates the advanced chips; South Korea and Japan supply memory and robotics; Singapore sells neutrality; the EU exports regulation via the Brussels Effect. Because the stack is a dependency chain, a grip on one critical node grants leverage over the whole.

2. Self-sufficiency has given way to strategic interdependence because the capital, energy, and resource requirements of frontier AI now exceed what any single actor can manage. This is arithmetic, not ideology, and it is what makes the UAE's "indispensability over dominance" strategy viable: it offers speed, cheap power, and patient capital that superpowers cannot easily replicate.

3. Alliances are proliferating in every form — transactional bilateral deals (U.S.–Saudi), middle-power networks (India's AI Impact Summit), securitized pacts (Quad, AUKUS), governance frameworks (G7, EU), and government-corporate hybrids. Goldman Sachs's "mega alliances" — fusing states, tech giants, energy firms, and sovereign funds — blur every line between public and private, commercial and strategic, with the Abu Dhabi campus as the archetype.

4. The central risk is fragmentation: rival blocs with incompatible standards, duplicated infrastructure, and diverging governance impose real costs and structurally exclude countries lacking capital or position. Whether governance competition is a healthy feature or a wasteful bug is genuinely unresolved.

5. Two honest uncertainties remain. It is not yet clear whether Gulf infrastructure investment buys durable geopolitical influence or merely commercial hosting capacity, nor whether mega alliances are durable relationships or conveniences that would fracture under real stress — because they have not yet been tested by one. The hedging strategy of the UAE, Singapore, and India works only as long as the blocs stay open. The digital Cold War is a warning, not a certainty; the map is still being drawn.

Sources

• Joint Statement – The Strategic Artificial Intelligence Partnership | U.S. Department of State
• AI Impact Summit 2026: India's Value Proposition for Trusted Global Partnerships | ORF
• What to Expect From AI in 2026: Personal Agents, Mega Alliances, and the Gigawatt Ceiling | Goldman Sachs
• EU international engagement on Artificial Intelligence | European Commission
• How the UAE Became an Emerging Superpower in the Global AI Economy | Times of Israel
• UAE and Saudi Arabia among world's top three AI superpowers | The National
• From Crude to Compute: Building the GCC AI Stack | Middle East Institute
• India, Israel to deepen AI cooperation at global summit | The Hans India
• UAE, India expand partnership in defense, tech, trade | The Jerusalem Post
• AUKUS and Allied AI: Building Trilateral Defense Capabilities
• Partnership for Global Inclusivity on AI | U.S. Department of State

Last updated: 2026-08-10