Anthropic spent much of 2024 telling anyone who asked that it would not take money from the Gulf. The reasoning, laid out internally and later reported, was about national security and moral positioning: an AI lab building potentially transformative systems, the argument went, should not be capitalised by state actors with their own strategic agendas. By mid-2026, that position had reversed. MGX, the Abu Dhabi vehicle chaired by the UAE’s national security adviser, Sheikh Tahnoon bin Zayed, is now a repeat investor in Anthropic, which was valued at $965 billion in its most recent round. The company was, by then, spending upward of $1.25 billion a month on compute alone, in what one investor called “a very delicate dance between growing too fast and going bankrupt.” Principle met arithmetic, and arithmetic won.
That reversal is worth dwelling on, because it is not really a story about one company changing its mind. It is a preview of a decision every capital-intensive technology business, and increasingly every large enterprise raising growth capital, is going to have to make over the next few years: whether to take sovereign money, on what terms, and with what governance attached. Executives who treat this as a financing question, no different from choosing between a growth-equity fund and a strategic investor, are underpricing what has actually changed underneath them.
The capital gap venture never filled
Start with why sovereign wealth funds are showing up at the term sheet stage at all. Roughly 90 percent of US venture capital flows into software, a business that scales with astonishingly little capital relative to revenue. AI infrastructure does not work that way. Training and serving frontier models, building fabs, laying data-centre capacity and securing the power to run it are capital-intensive in a way that looks more like heavy industry than software, and traditional VC funds, built around ten-year fund lives and modest check sizes relative to what a gigawatt data-centre campus costs, simply were not constructed for this. Sovereign funds were. Patient, permanent capital with multi-decade horizons and a mandate that already includes energy and infrastructure is a far better structural match for the AI buildout than a Sand Hill Road partnership raising its fourth fund.
That is why MGX closed a $49 billion AI-dedicated fund in mid-2026, why Saudi Arabia’s Public Investment Fund spun up Humain as a standalone AI vehicle, why Qatar Investment Authority created Qai, and why the UK government, not usually associated with sovereign industrial policy, launched a £500 million Sovereign AI Unit in 2025. Even the United States, historically the last government you would expect to run an equity portfolio, has a sovereign vehicle now holding a 10 percent stake in Intel and financing niche semiconductor plays such as XLight’s free-electron-laser lithography effort. The capital gap left by a venture industry over-indexed on software is being filled by nation-states, and it is happening at a pace and scale that has quietly made sovereign funds one of the largest single sources of capital available to the companies building the AI stack.
Why this capital is not neutral
Here is the part boards tend to skate past. A pension fund wants a return. A sovereign wealth fund wants a return plus something else: a national competitiveness objective, a hedge against technological dependency, a foothold in a strategic sector, sometimes a diplomatic relationship it wants to deepen. That second objective does not disappear once the wire lands. It shapes what the investor wants in follow-on rounds, what information rights it negotiates, what board access it expects, and how it behaves if geopolitics between its home government and the country where the company operates deteriorates. None of that shows up on a cap table as anything other than a percentage.
US regulators have started treating that gap between form and substance seriously. The Committee on Foreign Investment in the United States has effectively redefined what counts as “control” for review purposes: a foreign investor no longer needs a board vote to trigger a mandatory filing. Board observer rights, access to technical information, or involvement in strategic decision-making can be enough on their own. That is a meaningfully lower bar than most founders and even most board counsel assume, and it means a sovereign fund holding what looks, on paper, like a passive minority stake can be legally adjacent to control in the eyes of the agency that decides whether the deal closes at all. The Amazon-OpenAI relationship, complicated in part by Middle Eastern government participation in OpenAI’s cap table despite both companies being US-domiciled, is the live illustration of how tangled this has already become, with CFIUS, the FTC and Commerce Department export-control staff all working the same transaction from different angles.
The practical consequence is that sovereign capital is not just a bigger, patient version of a growth-equity cheque. It arrives with a governance shadow that most cap tables, and most board decks, do not currently have a line item for. A due-diligence process that checks a sovereign investor’s source of funds and sanctions exposure, then treats the rest as a normal term sheet negotiation, is checking for the wrong risk. The real question is what happens three years and two geopolitical crises from now, when the investor’s home government’s foreign policy and the company’s customer base or export-control obligations stop pointing in the same direction.
The quiet substitution of state capital for VC discipline
There is a second-order effect here that gets less attention than the headline cheque sizes. Venture capital has always performed a disciplining function beyond writing cheques: it prices risk, forces founders to justify burn, and withholds capital from ideas that do not scale efficiently. Sovereign capital, by design, is willing to fund things that do not need to generate venture-scale returns on venture timelines, because the underlying objective is strategic capacity, not IRR. That is genuinely useful for financing the unglamorous, capital-heavy layers of the AI stack, power, fabs, and compute, that pure financial capital was always going to underfund. But it also means the allocator increasingly deciding which frontier AI capabilities get built, and how fast, is a set of finance ministries and sovereign investment committees rather than a market of venture partnerships competing for returns. Industrial policy has re-entered the AI sector through the cap table rather than through a subsidy programme, and most executives are still analysing it with a financing lens rather than a policy one.
Where India sits, and does not sit, in this
India is a useful counterpoint precisely because it is not yet a player on the allocating side of this trade. The National Investment and Infrastructure Fund, India’s closest analogue to a sovereign vehicle, recently secured a fresh $2 billion in commitments, but it remains anchored in infrastructure and real assets rather than frontier AI equity, and the IndiaAI Mission’s emphasis has been on subsidising domestic compute access rather than taking global stakes in AI labs. That is a meaningful asymmetry: India is overwhelmingly a recipient of sovereign capital rather than a source of it, and Gulf funds, ADIA, PIF and QIA among them, already hold substantial positions across Indian renewables, telecom and infrastructure. The governance questions raised above are not hypothetical for Indian promoters; they are already live in board rooms in Mumbai and Gurugram, just one sector removed from AI itself. As Indian AI and deep-tech startups start raising the kind of capital that only a sovereign balance sheet can supply, the same information-rights and control questions that are reshaping Anthropic’s cap table will arrive on Indian term sheets, without India currently having an equivalent seat on the other side of the table.
What boards should actually do about it
None of this is an argument against taking sovereign capital; for the businesses building the physical layer of AI, it may be close to unavoidable. It is an argument for treating a sovereign term sheet as a geopolitical instrument with a coupon attached, not a cheque with a flag on it. That means capping information and observer rights before they accumulate into something a regulator treats as control, building sunset or step-down clauses tied to ownership thresholds, war-gaming what happens if the investor’s home government and a key market or customer fall out, and giving the board a standing framework for evaluating state-linked capital rather than relitigating the question deal by deal. The founders who get this right will have simply priced a cost that was always there. The ones who do not will discover, the way Anthropic nearly did in reverse, that the terms they thought were about money were actually about who gets a seat at the table when it matters most.
