Sovereign Wealth Funds Allocate One Percent to Digital Infrastructure by 2028

On July 21, 2026, MGX, the AI Infrastructure Partnership and BlackRock's Global Infrastructure Partners completed a $40 billion acquisition of Aligned Data Centers, among the largest data-center transactions on record. That single deal is roughly a quarter of one percent of the $17.2 trillion in combined assets across the ninety funds in Invesco's sovereign study. Which means this thesis is either long since settled or barely started, depending on a definition nobody states.

The Thesis Is Already True, or Nowhere Near

Read digital infrastructure as compute and the question is closed. Sovereign funds have committed an estimated $350 billion or more to AI infrastructure since 2025, led by MGX's $49 billion fund close and Saudi Arabia's Public Investment Fund deploying $36.2 billion in a single year, with roughly $120 billion going specifically to data centers, chip fabrication and high-performance computing across 2025 and 2026.

The allocation data says the same thing structurally. Average infrastructure allocation across the Invesco sample has nearly doubled to 9 percent since 2022, allocations to artificial intelligence and technology have risen from 5 percent in 2020 to 15 percent in 2026, and GIC alone reports $15 billion in digital infrastructure globally.

Read digital infrastructure as digital-asset rails, which is what the phrase usually means in this market, and the picture inverts. None of that capital went to custodians, settlement networks, chains or stablecoin issuers. It went to data centers, semiconductors and sovereign compute platforms. On that reading the allocation is not approaching one percent; it is not a recognized category.

What Must Be True for the Other Reading

Three things, and they are not equally far along. First, the exposure has to be investable at sovereign ticket size. A fund writing checks measured in billions cannot deploy meaningfully into a sector whose largest private operating companies raise in the hundreds of millions, which is a capacity constraint rather than a conviction one.

Second, the asset has to produce the cash-flow profile these mandates are built around. Data centers won this capital because they offer contracted revenue, inflation-linked escalators and a long asset life. Custody and settlement businesses are fee-based operating companies with regulatory capital requirements and no contracted revenue, which is a different instrument entirely.

Third, and this is the one that actually binds, the ownership has to be permissible at the stake size required to matter. That is not a market question and it is where the examiner enters.

The Examiner's Question Is Control

A sovereign wealth fund is not a supervised institution, so an examiner never looks at its portfolio. The examiner appears on the other side of the transaction, and the question is not what was bought but who now controls what.

When sovereign capital takes a stake in a regulated entity, a trust company, a chartered custodian or a permitted stablecoin issuer, the supervisory question becomes whether a foreign government has become a controlling person. That triggers change-in-control review, source-of-funds analysis, and potentially an entirely different regulatory regime for the target, with the acquirer's own governance, sanctions exposure and affiliate structure pulled into scope. The relevant threshold is a stake large enough to presume control, not a majority.

A data center has no such gate. It is an unregulated operating asset, and a sovereign fund can own 40 percent of a hyperscale campus without a banking supervisor forming a view. A chartered custodian cannot be acquired on the same terms by the same investor. That asymmetry is not a detail of the thesis; it is the mechanism that routes sovereign capital toward compute and away from rails.

Why the Capital Went to Compute

Honesty requires conceding the simpler explanation first. Data centers offer contracted revenue, escalators and scale, and infrastructure mandates were built for exactly that shape. Sovereign funds may simply be buying the better asset, and roughly 30 percent of AI allocations are directed at nation-building to create domestic champions, which is a strategic motive with nothing to do with regulation.

But the control gate does additional explanatory work, and it predicts the observed structure rather than just rationalizing it. Sovereign participation in regulated financial infrastructure, where it exists at all, arrives as limited partner interests in funds, minority co-investments below control thresholds, and commitments to sponsors who hold the operating stake. The pattern looks exactly like capital routing around a permissibility constraint.

For the agent-rails question this is the load-bearing distinction. If autonomous software needs settlement rails, those rails are licensed operating companies, and licensed operating companies are precisely what a sovereign investor cannot take a controlling position in without inviting a different supervisor into the room. Sovereign capital can fund the compute layer of that stack directly and can only fund the settlement layer indirectly.

Invalidation, and What to Watch

The thesis fails on any of three findings. If no sovereign fund takes a disclosed direct stake above a control threshold in a chartered digital asset institution by 2028, then the rails reading is wrong regardless of how much compute gets bought. If sovereign allocations keep arriving as fund commitments rather than direct holdings, the one percent will be real on paper and will represent no governance position in any rail. And if the compute reading is the one intended, the thesis was satisfied before it was published, which makes it a description rather than a forecast.

Four things are trackable monthly and all are public. Change-in-control filings and charter applications naming a sovereign entity as an acquirer. The direct-versus-fund split in sovereign digital infrastructure commitments. Whether any national trust or payment-stablecoin charter is granted to an entity with disclosed sovereign ownership. And whether a sovereign fund publishes digital assets as a reporting line distinct from technology or infrastructure, which is the cleanest signal that it has become a category at all.

The constructive reading is that this capital has already shown appetite for the technology stack at enormous scale, which disposes of the conviction question. What remains is a permissibility question. The honest restatement is that sovereign funds pass one percent of assets in digital infrastructure well before 2028 and probably already have, that essentially none of it sits in regulated digital-asset rails, and that the constraint keeping those two facts apart is a control threshold rather than a view on the asset.

For informational purposes only. Not an offer to buy or sell any security. Available only to accredited investors who meet regulatory requirements.

‍