On June 12, 2026, as SpaceX listed on Nasdaq at a $1.75 trillion valuation, Rest of World drew attention to a pattern buried in the prospectus: Gulf sovereign capital is now woven through America's three flagship AI systems. Saudi Arabia's Public Investment Fund (PIF) is in talks to invest $5 billion in SpaceX, which now owns xAI. Months earlier, the PIF-owned company Humain had put $3 billion into xAI's Series E. What makes these deals different from ordinary cross-border investment is the thing attached to them: each is paired with advanced AI data-center capacity built on Saudi soil.
The deal is the infrastructure
The Humain–xAI arrangement is explicit. Announced at the November 2025 U.S.–Saudi Investment Forum, the two committed to jointly develop more than 500MW of next-generation AI data-center and compute infrastructure inside Saudi Arabia, and to deploy xAI's Grok models there. The $3 billion equity stake came on top of that buildout. This is not portfolio investment that sits passively in a brokerage account; it is a transfer of physical compute — the racks, the cooling, the network ingress and egress — into a jurisdiction the investor controls.
That distinction matters because data centers are not neutral boxes. Whoever holds the keys to the building holds the power to compel access, to mandate logging, to condition operating licenses on lawful-intercept hooks, and to inspect what crosses the border. Frontier AI compute located inside a state becomes, in a real sense, an instrument of that state.
Why the buyer's record is the whole story
The strongest case for welcoming this capital is straightforward, and we should state it plainly. Sovereign wealth diversifies the investor base for hugely capital-intensive AI infrastructure, reduces dependence on a handful of US hyperscalers, and spreads compute geographically in ways that can lower latency and cost for billions of users outside North America. Capital is capital, and a more multipolar compute map is not inherently bad.
The problem is the specific buyer. Saudi Arabia has one of the most thoroughly documented records of digital surveillance against dissidents anywhere. In January 2020, the University of Toronto's Citizen Lab reported that a Saudi-linked operator it named KINGDOM had targeted New York Times journalist Ben Hubbard with NSO Group's Pegasus spyware. The same operator was tied to the surveillance of Saudi dissidents and human-rights workers in the period surrounding the October 2018 murder of Jamal Khashoggi. This is not contested fringe allegation; it is forensic research that has since been litigated in open court.
And it is still being litigated. In October 2024, the UK High Court granted permission for Saudi human-rights defender Yahya Assiri to bring a legal challenge against Saudi Arabia over spyware — including Pegasus and QuaDream technology — that he says targeted him between 2018 and 2020. The point is not that surveillance happened once; it is that it is an ongoing, court-acknowledged pattern of transnational repression.
The surveillance pattern is current, not historical
It would be comforting to treat the spyware era as the past. It is not. On May 20, 2026, a coalition of twelve civil-society organizations — including Access Now, the EFF, and ALQST — condemned Meta for geo-blocking human-rights accounts inside Saudi Arabia at the government's request. Since April 30, 2026, the Facebook accounts of ALQST, Democratic Diwan, researcher Abdullah Alaoudh, and — again — Yahya Assiri were rendered "unavailable" to Saudi audiences. Per Meta's own content-restriction reports, over 100 Facebook pages and Instagram accounts had been restricted since March 2026. Notably, when Riyadh asked X to geo-block prominent Saudi activists, X had not complied as of that date.
That contrast is instructive. The same government now acquiring physical control of frontier AI compute is, in the same season, actively pressuring US platforms to silence its critics — and platforms are partially obliging. Put a Grok-grade model and 500MW of compute inside that government's borders, under its licensing and lawful-access regime, and the leverage runs the other direction far more cheaply than spyware ever did.
A proportionate fix, not a wall
The pro-innovation answer is not to slam the door on Gulf capital, which would be both futile and economically self-defeating. It is to attach the investment to the infrastructure's governance, not just its financing. Three guardrails would go most of the way:
- Operational firewalling. Model weights, training data, and inference logs for frontier systems should be contractually and technically walled off from host-state access, with the burden on the operator to prove no lawful-intercept hooks exist.
- Disclosure. US-listed companies — SpaceX/xAI now among them — should disclose foreign-state data-residency and access terms in filings, so investors and regulators can price the surveillance risk rather than discover it in a prospectus footnote.
- CFIUS-style review of the bundle. Where an equity stake is conditioned on co-located compute in a country with a documented record of targeting dissidents, the data-center clause deserves the same scrutiny as the equity.
None of this blocks a single dollar. It simply insists that frontier compute carry the same accountability we would demand of any dual-use export. Saudi Arabia's money can help build the AI buildout. Its surveillance apparatus should not get a free ride inside it.