The Sovereign Meter

On July 1st, MGX, an Abu Dhabi state fund, closed its first vehicle at roughly $49 billion. It now holds equity in all three American frontier labs — OpenAI, Anthropic, and xAI (the last now folded into SpaceX). The next day, OpenAI floated the idea of granting five percent equity to a United States sovereign wealth fund. Six weeks earlier, the Trump Administration's own plan to build such a fund had quietly stalled. The result: foreign sovereigns are buying the AI stack outright while Washington keeps signaling its intent and stalling.

This signals a shift in how AI infrastructure gets funded going forward; a maturation beyond private capital markets and into the infrastructure-finance era. Every prior buildout of this scale from railroads to electrification to telecoms followed the same arc: venture capital funds the invention, corporate balance sheets fund early scaling, and then the capital requirement blows past what return-seeking private money can recycle, and state-adjacent patient capital takes over the base layer.

The forcing mechanism is scale. Each frontier lab now burns on the order of tens of billions of dollars per year to sustain growth; the data-center pipeline alone runs to hundreds of billions annually. Sovereign entry isn't a curiosity, and on the surface it looks benign — patient money arriving exactly when the buildout needs it. This Letter argues the surface is misleading: what arrives as financing hardens into leverage over access.

Setting aside AI's broader externalities, why should anyone care how the infrastructure is financed, so long as the money advances the technology and widens access to it? This Letter unpacks what it means for the emerging Grid-Silicon Order and the associated structural shifts in the global economy and geopolitical risk going forward.

Ultimately, it boils down to control, and for executives staking enterprise strategies on ubiquitous AI, at least for the foreseeable future, the trend creates an open question around who will set the price, terms, and availability of the intelligence their business runs on. For two years the AI story has centered on who owns the best models, then who can secure the most power. Now it is moving again to who owns, and on what terms, the capital base underneath both.

MGX is worth watching as a barometer of two things: Gulf sovereign capital flowing into Western AI, and the U.S.–UAE relationship around chip access and AI-security guarantees. The fund is focused almost entirely on artificial intelligence and the infrastructure beneath it. Against a backdrop of Senate scrutiny of UAE-U.S. dealings, including sworn-testimony demands to administration officials over a UAE investment and questions about affiliate G42's China ties, MGX closed its $49 billion fund, oversubscribed, on July 1st.

On July 2nd, OpenAI proposed handing five percent of its equity, estimated at $40+ billion, to a U.S. fund, preliminary and contingent on Congress. Within twenty-four hours, reports indicated that Anthropic and the administration had held no such talks, and that Anthropic reportedly preferred a tax-funded "digital dividend" to ceding stock.

Meanwhile in the Senate, Senator Bernie Sanders introduced a bill to take a fifty-percent public stake in large AI firms, which drew zero co-sponsors. President Trump's own effort, wrapped up in Executive Order 14196 signed back in February 2025, had by May 2026 run into a wall: the Treasury and Commerce plan to actually stand up a sovereign fund was sent back unapproved, reportedly caught on legal, fiscal, and political friction — and on the recognition that a standing pooled fund of that kind likely needs congressional authorization an executive order cannot supply.

The contrast matters. Across these weeks, foreign equity cleared and closed on schedule, while every U.S. attempt to take an equity stake in a frontier lab — the executive-order fund in May, the Sanders bill in June, OpenAI's own July float — stalled, drew no support, or waits on a Congress that has not acted.

Sovereign ownership of AI is not one financial instrument or structure. It is four different things, and they are not interchangeable, each turning the gears on AI deployment in different ways with different implications for anyone dependent on AI's development (i.e., all of us).

It starts with the underlying shift they all turn on. Intelligence (i.e., inference) is becoming a metered input to the economy with electricity as the closest analogy. A business does not own its power plant; it pays for what it draws, on terms someone else sets. AI is evolving the same way. Whether for enterprises or for individuals, at least for the foreseeable future, it's a capability you meter rather than own. As this arrangement hardens, the contest shifts from being about who builds the smartest model to who sets the meter. In other words, who controls the price, the terms, and the availability of the thing everyone else has to buy?

Each of the four instruments below sets that meter, and each works differently.

Also critical to this puzzle: owning a piece of something is not the same as controlling it. The U.S. government's stake in Intel is explicitly passive with no board seat or voting interest. OpenAI's public-benefit structure gives every outside investor, sovereigns included, economic upside but not governance. India's "sovereign" compute runs on American chips and American cloud. A sovereign check buys a share of the returns.

Where the actual power sits — i.e., who can turn the access on or off — is a separate question, and often a separate instrument.

Looking across all four instruments, a consistent trend is emerging. Foreign equity into AI is accelerating and broadening, not merely testing the water.

This includes MGX's $49 billion close mentioned earlier, which came with equity in all three labs, a $40+ billion consortium data-center acquisition (Aligned Data Centers, with BlackRock's GIP), and a 15% stake in the U.S. TikTok entity. Kuwait's fund joined the BlackRock-anchored AI Infrastructure Partnership (of which MGX is a founding partner) as its first non-founder anchor. Qatar stood up its own vehicle and a $20 billion joint venture with Brookfield. Singapore's GIC and Temasek anchored Anthropic's latest round. Abu Dhabi's Mubadala led all sovereign funds globally in the first half of 2026.

One lab's near-vertical valuation is a direct beneficiary. Anthropic went from a $30 billion round at a $380 billion valuation in February to a $65 billion round near $965 billion just months later with sovereign money in both.

Of the instruments, Gulf equity is being waved through by the U.S. This includes a CFIUS fast-track for known allied investors and (reportedly) a chip-friction provision being stripped out of the defense bill under White House pressure. With U.S. sovereign investment equity not clearing Congress, Washington is reaching for what it can execute without a vote: the passive Intel stake, the chip-revenue share, the model-access agreements. The pattern is not that the U.S. is absent from the capitalization table. It is that the U.S. is present through every instrument except equity.

Ownership keeps decoupling from control. The sovereign money buying in is chasing returns, not governance. No frontier lab has granted a sovereign investor a board seat or a bespoke control right; the structures are built specifically to prevent it. That decoupling is critical to understanding the current landscape, but it is also its fragility.

Saudi Arabia's Public Investment Fund (PIF), its primary sovereign wealth fund, abandoned its $100 billion chip-and-electronics manufacturing ambition (its Alat vehicle) and redirected investment to data centers. This suggests the physical manufacturing layer resists sovereign ownership even where the capital is willing. By several counts — the headline figure is contested — close to half of the planned 2026 U.S. data-center buildout is delayed or cancelled, held up by transformer and switchgear lead times and multi-year interconnection queues. Capital is not the issue especially as the pool of ownable physical assets keeps shrinking. Norway's sovereign fund has used its positions to constrain rather than own, pressing companies rather than controlling them — a reminder that sovereign capital is not one bloc with one intent. These are where the read could be wrong.

Oil provides the clearest historical model for a strategic resource whose ownership evolves through a sequence of instruments.

Control of Middle Eastern crude did not pass from foreign to national hands in one step. It moved through instruments, each one capturing more of the rent than the last. First came the concession: a foreign company paid a flat royalty and took the oil, and the producing state owned the ground but almost none of its value. Then came fifty-fifty profit splits in the early 1950s, which handed producing governments half the take without transferring an asset. Then the participation agreements of the 1970s granted minority state equity phased upward over time. Finally nationalization, the state taking the operating companies outright. At every stage the barrel remained the same. Around it, the investment instrument changed, and the instrument, not the oil, decided who captured the rent.

The AI stack, while early in the same sequence, is following a similar trend. It's become a strategic resource everyone increasingly needs with capital-rich states moving to capture a share of it and the instruments escalating as the resource proves indispensable. Foreign sovereigns appear to be roughly between concession and participation — buying economics, not yet in control.

But three differences suggest the AI case will evolve differently.

First, the resource (turning sand into intelligence) is not a fixed well. A lab is intelligence and compute, not a reservoir under sovereign soil, and its corporate structure can hold economic benefits and governance apart in a way an oil concession never could. Second, two of the four instruments have no analog in the oil precedent. The revenue-share and access-right instruments capture rent with no equity and no visible ownership at all, so the most consequential positions may never appear on a capitalization table. Third, the physical layer resists capture: Saudi's PIF walked away from its fab ambition, and you cannot nationalize a supply chain that can't be stood up. Oil's endgame was the state owning the asset. This resource's asset is partly intangible, partly unbuildable, and the ownership that matters may be a permission rather than a share.

Sovereign capital is arriving exactly when the frontier labs and AI face a "move past the hype" moment. Token burn, widespread enterprise adoption, and value capture challenges taint an otherwise historic adoption curve. Labs are still running in the red as corroborated by many outlets. While those losses are the reality of commercializing disruptive emerging technologies, outsized returns remain very much in play. Still, private money can be skittish and withdraws when returns disappoint. That market reaction is what stops an overbuild.

Patient sovereign capital does not answer to the same financial physics. It keeps deploying straight through the shocks that make private money flinch, including drone strikes on Gulf data centers and short-term market swings.

For the next year, the two-track regime is much more likely to harden rather than converge. Foreign-sovereign equity will increase and MGX-like funds are likely to enter the game, of which, Qatar's Qai paired with its Brookfield venture is the leading candidate.

While possible, it's unlikely the U.S. ratifies frontier-lab equity, because that still requires Congress to approve. Instead, the U.S., especially under the Trump Administration, is likely to increase investment in the instruments it can execute without a vote: revenue shares, access rights, review powers. Inference settles into what it has been becoming, a conditionally priced input whose terms are set by the holders of those instruments, not by shareholders.

So the AI buildout story is more likely to continue marching forward in spite of physical infrastructure constraints. Once patient sovereign capital takes over the base layer, the impact of private capital's potential withdrawal is increasingly muted.

Aroko plots the Grid-Silicon Order, which describes the collision between the Energy Transition and the AI Infrastructure buildout. The rise of sovereign investment suggests the Grid-Silicon Order remains firmly entrenched in Sovereign Stacks, reflecting a fragmented geopolitical architecture and AI infrastructure growth expansion. The market remains nominally open, but the capital base of the intelligence economy is increasingly held by individual countries, each bloc financing and owning its own corridor of the stack.

The rise of sovereign investment suggests the Grid-Silicon Order remains firmly entrenched in Sovereign Stacks. The market remains nominally open, but the capital base of the intelligence economy is increasingly held by states, each bloc financing and owning its own corridor of the stack.

The counter-sovereign wave — Korea, Japan, the U.K., India each building rather than buying in — is Sovereign Stacks assembling in real time. The drift is not toward one government owning the labs. It is toward a stack whose financing, and therefore whose terms of access, fragment along sovereign lines.

For executives and boards, the planning variable going forward shifts. The input your business will run on is metered, and the ones who set the meters are increasingly sovereign countries rather than markets. See the H20 chip: banned from China, then relicensed only once a 15% revenue cut was attached; access turned on, off, and repriced by policy, not by the market. The exposure risk going forward is less who your model vendor is; it is the conditionality sitting behind them: whose export license, whose review right, whose sovereign anchor keeps the capacity funded? Price is the visible risk, but terms of access are the real one.

For allocators, the ownable surface is bifurcating. Equity captures economics but not governance, and two of the four instruments — revenue-share and access rights — never touch a capitalization table, which means public-market exposure systematically understates where sovereign leverage actually sits. Track conditionality, not ownership. And weigh the return-insensitivity directly: capital that does not need to clear a hurdle rate can sustain an overbuild long past the point where private money would have forced a correction, which changes both the shape and the timing of any eventual repricing. In other words, bubble or no bubble, the AI train marches on.

For operators and policymakers, the most consequential instrument is already domestic and already quiet. Washington's real leverage over the labs is not the equity stake it keeps failing to take; it is the access-and-review rights it has already executed. Conditionality is the instrument, and it cuts both ways: capital that a government admits, a government can later expel, and the same is true of the access it grants.

The frame keeps moving. First it was who owns the best model, then who can secure the most power. Now it is who sets the terms of access to intelligence itself. Increasingly sovereign states dictate the terms, not shareholders. The scramble for the stack is not really a scramble to own it. It is a contest over who gets to set the meter.

The forecast is just a forecast, not a certainty. If it's proven wrong before mid-2027 it will be because of any of the following:

Any one of these developments would impact the forecast. The board-seat tripwire is the single cleanest tell, because the day ownership stops decoupling from control, this analysis no longer holds true.

An Abu Dhabi fund, MGX, now holds equity in all three American frontier labs, while Washington's attempts to take one has stalled. A proposed executive-order sovereign wealth fund has yet to materialize, a bill proposed by Senator Sanders drew zero co-sponsors, and OpenAI's five-percent float to the U.S. government waits on a Congress that has yet to act. That asymmetry defines this moment.

Meanwhile, intelligence (i.e., inference) is turning into a metered input priced like electricity, drawn and paid for on terms someone else sets. As that arrangement hardens, the competitive question moves from who builds the best model or secures the most power to who sets the price, terms, and availability of what everyone else must buy.

Sovereign capital sets that meter through four instruments, and only equity shows up on a capitalization table. Revenue shares and access rights do not, so public markets understate where the leverage sits. Ownership is decoupling from control; the money buys economics, not governance.

Oil provides a precedent for how this could eventually play out, from concession to nationalization, and the instrument, not the barrel, decided who captured the rent. Through mid-2027 the two tracks harden rather than converge. The cleanest tell is a sovereign board seat at a frontier lab: the day ownership stops decoupling from control, the read inverts.

About Aroko: Aroko provides strategic advisory and capital allocation intelligence at the intersection of energy transition, technology infrastructure, and geopolitical risk. Our analytical process combines proprietary evidence infrastructure with human-directed thesis formation. Every keystone claim is verified against primary sources, and all editorial judgment and capital allocation framing is conducted by Aroko’s team. The Letter is published biweekly for institutional allocators.

Disclaimer: This article is for informational purposes only and does not constitute financial, investment, legal, or tax advice. The opinions expressed regarding macro trends and infrastructure investments are solely those of the authors. Past performance does not guarantee future results. Readers should consult with a qualified financial professional before making any investment decisions.