Sovereign wealth funds' AI-infrastructure commitments — Q1.

The headline numbers, decomposed

The aggregate $42B figure across the eight funds is decomposed roughly as follows. CoreWeave-related primary and secondary positions: $7.8B across GIC, NBIM, ADIA, and Mubadala. Crusoe Energy commitments and adjacent power-side positions: $4.2B, primarily ADIA and PIF, with a smaller GIC participation. Together AI direct positions and Series F follow-on: $1.6B, primarily GIC and Future Fund. Direct GP commitments to AI-focused funds — Coatue Tactical Solutions, Khosla Opus III, the Andreessen Horowitz American Dynamism vehicles, and the Stargate-aligned co-investment programme — total $11.4B across the eight funds. Data-centre real-estate positions, both through specialist REITs (Digital Realty, Equinix, the Stargate-aligned development vehicles) and through direct land and power-grid-adjacent assets: $9.6B. Energy-procurement positions including gas turbines, the small modular reactor (SMR) commitments at NuScale, Oklo, and the X-Energy-aligned development programmes: $7.4B. The composition tells a clearer story than the headline. The funds are positioning at every layer of the AI infrastructure stack except the foundation-model layer itself, which they are accessing — when they access it at all — through GP commitments rather than direct positions.

The fund-by-fund breakdown shows the divergence across the cohort. GIC, the most disciplined of the major sovereigns in the AI-infrastructure positioning, allocated approximately $9.2B across its Q1 disclosures, weighted toward CoreWeave (a $2.4B primary and secondary position), the Coatue Tactical Solutions commitment ($1.8B), data-centre real estate ($2.6B through a combination of Digital Realty exposure and direct land positions in Northern Virginia and Singapore), and Together AI ($800M follow-on). The Singapore-resident fund has been explicit, in its public communications, that its AI-infrastructure positioning reflects a thesis on the structural undersupply of inference capacity rather than a bet on the foundation-model layer. The thesis is well-articulated in the GIC chief investment officer's spring letter dated 14 April 2026, which lays out the fund's view that inference compute demand will grow at a 35-45 per cent CAGR through 2030 and that the current supply-side response is materially undersized for that demand curve.

ADIA's positioning, at approximately $8.4B disclosed in Q1, is the most weighted toward the energy layer of the eight funds. ADIA's largest single position is the $2.8B commitment to a Crusoe Energy expansion of its flared-gas compute facility programme, structured as a combination of equity and project-finance debt. The fund has also committed $1.4B to the Oklo SMR development programme through a co-investment structure with the Department of Energy's loan guarantee arm, and approximately $900M to a direct natural-gas turbine procurement programme aimed at supporting the fund's data-centre real estate joint venture with a major US REIT. The energy-heavy weighting reflects ADIA's institutional history and its operating-asset competence: the fund has been investing in energy infrastructure for decades, and the AI-infrastructure thesis maps onto a competence the fund already has rather than requiring the development of new diligence capabilities. The architectural choice — leading with energy, supporting with data-centre real estate, lighter exposure to the compute-services layer — is consistent across multiple fund communications and is best understood as a deliberate institutional strategy rather than as a tactical allocation.

NBIM, the Norwegian sovereign fund, allocated approximately $6.8B to AI-infrastructure positions in Q1, with the heaviest weighting in the data-centre real-estate and public-equity-proxy layers. The fund's Q1 13F-equivalent disclosure shows significant adds to Digital Realty, Equinix, NextEra Energy (positioned as a power-supply play to the AI build-out), and a meaningful new position in Constellation Energy after the company's nuclear-power-purchase agreements with the major hyperscalers. NBIM's AI-infrastructure positioning is the most public-equity-anchored of the eight funds, reflecting the Norwegian fund's structural preference for liquid positions and its disclosure requirements under the Government Pension Fund Global mandate. The fund's chief investment officer Nicolai Tangen, in a 22 April interview with the Financial Times, framed the AI-infrastructure positioning as "selective exposure to the picks-and-shovels of the build-out, declining direct exposure to the model layer for diligence-capacity reasons." The framing is significant. NBIM is the largest sovereign wealth fund in the world by AUM, and its choice to decline direct foundation-model exposure on diligence-capacity grounds is a meaningful market signal.

The Gulf funds: Mubadala and PIF's parallel-but-distinct positioning

Mubadala's Q1 disclosure runs to approximately $5.4B across the AI-infrastructure layer, with a position composition that differs from ADIA's despite the geographic adjacency. Mubadala has been more aggressive at the GP commitment layer than ADIA, with disclosed commitments to Coatue Tactical Solutions ($1.2B), Khosla Opus III ($600M), and the Stargate-aligned co-investment programme ($800M). The fund's direct positions are weighted toward the compute-services layer rather than the energy layer: a $1.4B CoreWeave primary and secondary position, a $400M Together AI participation, and a $400M position in a specialist inference-optimisation company that has not been publicly named. The Mubadala thesis, articulated by Khaldoon Al Mubarak in a 30 April investor communication, emphasises the fund's preference for exposure to "the layer of the AI stack where unit economics will be defensible by 2028." The framing is essentially the same as the GIC framing — undersupplied inference capacity — but the implementation skews toward GP commitments and compute-services direct positions rather than data-centre real estate.

PIF's Q1 positioning, at approximately $4.8B disclosed, is the most concentrated of the eight funds. The fund's largest single position is the $2.2B commitment to the HUMAIN AI infrastructure programme, which sits at the centre of Saudi Arabia's national AI strategy and was structured in late 2025 as a sovereign-backed co-investment vehicle that PIF anchors. The remainder of the PIF allocation is distributed across a $900M position in the NEOM data-centre development programme, a $600M commitment to a Stargate-aligned development vehicle with a Middle East deployment focus, and approximately $1.1B across a small set of US-headquartered AI infrastructure direct positions including Crusoe Energy and an undisclosed power systems specialist. The fund's positioning is the most explicitly national-strategy-aligned of the eight funds. The HUMAIN commitment is the largest single AI-infrastructure commitment any sovereign wealth fund disclosed in Q1, and its size relative to the rest of the PIF AI book signals that the fund is treating AI infrastructure as a strategic-priority allocation rather than a portfolio-construction allocation.

The Gulf-fund coordination patterns, visible in the cap-table-of-record data, are subtle but meaningful. ADIA, Mubadala, and PIF have appeared as co-investors in five named Q1 deals, including the Crusoe Energy expansion, two data-centre development vehicles, and two GP commitments. The frequency of co-investment is higher than would be predicted by random matching among the cohort, suggesting either deliberate coordination or, more likely, a shared deal-sourcing infrastructure that surfaces the same opportunities to all three funds. Multiple sources close to the funds — speaking off-record at the spring sovereign-wealth-fund convening in March — confirmed that informal deal-sharing between the Gulf funds has become more systematic in 2025-2026 than in previous cycles, partly because the deal sizes in AI infrastructure routinely exceed what any single Gulf fund prefers to underwrite alone. The pattern is best understood as natural coordination among funds with adjacent investment mandates rather than as formal joint-venture activity, but the effect is the same: a meaningful subset of the AI-infrastructure deals available at the sovereign-fund scale is being underwritten as Gulf-anchored syndicates.

The Khazanah Q1 disclosure, at approximately $1.6B in AI-infrastructure positions, is the smallest of the eight funds but the most structurally interesting in terms of regional positioning. The Malaysian sovereign fund has concentrated its AI-infrastructure exposure in Southeast Asia, with the largest single position being a $700M commitment to a Malaysian data-centre development programme partnered with a major US REIT. The fund's strategy, articulated by managing director Amirul Feisal Wan Zahir in the Q1 investor letter, is to position Malaysia as a regional AI infrastructure hub by anchoring the capital base for the early-stage build-out. The thesis is regional rather than global, and the position sizes reflect the smaller balance sheet, but the strategic logic is the same as the larger funds: invest in the physical layer of the AI infrastructure stack rather than the model layer, and capture the value that accrues to the underlying capacity rather than to the more contested model-layer competition.

Temasek and the Future Fund: the technology-fluent sovereigns

Temasek's Q1 AI-infrastructure positioning, at approximately $3.4B disclosed, sits closer to the model layer than any of the other seven funds. The Singapore-resident fund's largest Q1 commitments include a $900M follow-on in Anthropic via a secondary purchase from earlier-stage investors, a $400M direct position in xAI via the Series F primary round, a $600M commitment to the Khosla Opus III fund, and approximately $1.5B across infrastructure-layer positions including CoreWeave and a specialist GPU-procurement financing vehicle. The Temasek positioning is structurally different from the rest of the cohort because the fund has, alone among the eight, retained meaningful direct exposure to the foundation-model layer through both primary and secondary positions. The choice reflects the fund's longer history of technology-direct investing and its institutional comfort with model-layer diligence — Temasek has been investing in tech-direct positions for more than a decade and has built the staffing to underwrite at the model layer in a way that several of the other sovereigns have not.

The Future Fund of Australia, at approximately $2.6B disclosed in Q1, has positioned similarly to Temasek but at a smaller scale. The fund's largest position is a $800M commitment to a co-investment vehicle anchored on Anthropic and adjacent secondaries, structured by Future Fund alongside Coatue and a single multi-family office sponsor. The remaining Future Fund positions distribute across data-centre real-estate exposure ($600M, weighted to Australia-located developments), GP commitments to Coatue Tactical Solutions and a regional growth fund ($700M combined), and approximately $500M across a small set of US-headquartered direct positions in AI-infrastructure adjacent companies. The fund's chief investment officer Raphael Arndt has been more publicly vocal than most sovereign CIOs about the AI investment thesis; his April 28 portfolio update letter to the Australian Treasurer included a multi-paragraph discussion of the fund's AI-infrastructure framework that is the most structured public articulation of any of the eight funds' positioning approaches. The Australian fund's willingness to engage publicly with the thesis reflects both Arndt's personal style and the broader Australian sovereign-fund tradition of relatively higher disclosure than the median in the global cohort.

The Temasek and Future Fund positioning, taken together, raises a structural question for the rest of the cohort. The two technology-fluent sovereigns are the only ones in the cohort writing meaningful direct foundation-model exposure, and their willingness to do so reflects institutional staffing and competence that the other funds either do not have or do not yet have. If the next twelve to eighteen months produce the kind of foundation-model-layer return outcomes that several of the technology funds are now projecting in their LP marketing — the kind of outcomes that would meaningfully reward direct exposure relative to infrastructure-only positioning — then the rest of the sovereign-fund cohort will face pressure to either build the staffing required for direct foundation-model diligence or to expand their GP commitment programmes substantially to access the layer indirectly. The early signs from the spring convenings suggest that the GP commitment route will be the dominant response. The infrastructure-only positioning of GIC, NBIM, ADIA, and Mubadala is unlikely to extend to a wholesale exclusion of the foundation-model layer if the layer continues to produce above-market returns.

A subtler signal comes from the secondary-market activity that the technology-fluent sovereigns have been driving. Temasek and the Future Fund have both been active buyers in the Anthropic secondary market across late 2025 and early 2026, picking up positions from earlier-stage funds that have been crystallising IRR in advance of their fund's harvest period. The secondary trades have priced at modest premiums to the prior primary round, suggesting that the marginal sovereign bid is supporting the foundation-model layer's valuation environment in a way that the broader cohort's positioning would not predict. The pattern is one of the more interesting market microstructure features of the current cycle: the largest pool of long-duration capital in the world is providing the marginal bid for the foundation-model layer's secondary liquidity, even as the modal sovereign-fund positioning is officially infrastructure-only.

The energy-procurement angle: gas turbines, SMRs, and the power-side bet

The energy-procurement layer of the sovereign-fund AI-infrastructure book is the part of the positioning that the consultant-class narrative most consistently understates. ADIA's $900M direct gas turbine procurement programme is the most visible position, but the pattern extends across multiple funds. GIC has disclosed a $400M commitment to a natural-gas-turbine financing vehicle that supports a major US REIT's data-centre development pipeline. Mubadala has participated in a $600M co-investment with a US utility holding company that is building dedicated combined-cycle gas capacity for hyperscaler power-purchase agreements. PIF's NEOM data-centre development programme includes direct gas turbine procurement at the project level. The aggregate gas turbine and natural-gas-power exposure across the eight funds, including both direct equipment and the project-finance debt against gas-fired generation, exceeds $4.2B in Q1 alone. The number is large enough to be material to the gas turbine OEMs' order books and to the natural-gas supply chain that feeds the new generation capacity.

The SMR commitments are smaller in current dollar terms but structurally more interesting. ADIA's $1.4B Oklo commitment is the largest single SMR-related disclosure across the eight funds. NBIM has disclosed approximately $400M in NuScale equity, structured through the fund's standard public-equity allocation. Mubadala has committed approximately $300M to the X-Energy-aligned development programme through a co-investment with a US-headquartered energy fund. The Future Fund has disclosed a $200M commitment to an Australia-domiciled SMR development programme. The SMR positions are positioned as long-dated infrastructure plays — none of the SMR developers will produce commercial-scale power before 2028-2030 — and the funds are explicit in their communications that the SMR commitments are not expected to deliver near-term returns. The positioning is closer to a structural option on the energy supply for AI infrastructure than to a near-term return-generating investment. The optionality is meaningful: if the SMR developers reach commercial scale in the late 2020s, the early commitments will compound substantially. If they do not, the positions are written down. Both scenarios are priced into the fund-level expected return profile.

The transmission-and-grid-adjacent exposure is a quieter component of the energy positioning that deserves attention. Several of the funds have disclosed positions in transmission-and-grid services companies — utilities that own the grid connections that hyperscalers must secure to operate data centres at scale — and the positions are structured to capture the rent that transmission constraints are increasingly extracting from the AI infrastructure build-out. NBIM's add to NextEra Energy is the largest single position in this category; GIC has disclosed a smaller position in a Northern Virginia-area utility holding company that has been a primary beneficiary of the Northern Virginia data-centre demand surge. The transmission-and-grid layer is less visible than the gas turbine and SMR layers, but the structural rents it captures are arguably more durable: the transmission capacity is the bottleneck that the AI infrastructure build-out cannot easily route around, and the holding companies that own the relevant assets have meaningful pricing power that should compound through the build-out cycle.

The energy positioning, taken in aggregate, signals a thesis that the sovereign funds are betting more on than they are saying publicly. The thesis: AI infrastructure demand will run into a power-supply constraint within 24-36 months that the gas turbine and SMR cohort will partially relieve, and the relief will be priced significantly above the marginal cost of supply because the demand side will be willing to pay almost any price to maintain its build-out cadence. The positioning of the sovereign funds on the power side is consistent with this thesis. The position sizes — $7.4B aggregate across the energy layer in Q1 alone, with the gas turbine and SMR components leading — reflect the level of conviction the funds are willing to express on the thesis. The conviction is meaningful. It is also one of the cleanest expressions of the AI infrastructure investment thesis available in the public record, and it deserves more attention than the consultant-class narrative has given it.

Inter-fund coordination patterns and the sovereign syndication layer

The cap-table-of-record data across the Q1 disclosures registers more inter-fund co-investment than any quarter on record. The eight named funds participated together in 14 named transactions across Q1, with five of those transactions involving three or more of the funds as named co-investors. The Crusoe Energy expansion alone drew named participation from ADIA, PIF, and GIC. The Coatue Tactical Solutions commitment closed with five of the eight funds as named LPs (GIC, ADIA, Mubadala, Future Fund, and a sixth fund not in this analysis cohort). The Khosla Opus III fund closed with four named sovereign LPs (GIC, ADIA, Mubadala, Future Fund). The pattern of overlap is statistically distinct from random co-occurrence among the cohort, and the consistency of the named-fund clustering across multiple unrelated deals suggests either deliberate coordination or the operation of a shared deal-sourcing infrastructure that surfaces the same opportunities to the same set of funds.

The shared deal-sourcing infrastructure is the more likely explanation. The sovereign-wealth-fund community has been investing in shared deal-flow platforms for the better part of a decade — the most-cited platform is the IFSWF (International Forum of Sovereign Wealth Funds) deal-sharing network, but several smaller informal networks operate at the chief-investment-officer level. The IFSWF spring meeting in March 2026, attended by senior representatives of all eight funds in this analysis, produced what one attendee described as "the most active deal-sharing dynamic the forum has supported in any quarter." The specific deals discussed are not public, but the spring meeting was followed by a measurable spike in cross-fund co-investment activity over the subsequent six weeks. The temporal pattern is consistent with the spring meeting having functioned as a deal-sharing accelerant rather than a formal coordination mechanism.

The implications of the inter-fund coordination patterns are worth considering. The first is that the sovereign-fund cohort is, in aggregate, exerting more pricing pressure on the AI-infrastructure layer than a less-coordinated group of capital allocators would. The funds are sharing diligence, sharing pricing intelligence, and increasingly anchoring transactions together at scales that smaller pools of capital could not support alone. The second implication is that the AI-infrastructure layer's primary-round valuations are increasingly being set by sovereign-anchored syndicates rather than by traditional venture or growth-equity capital. The pricing dynamics that prevail in a sovereign-anchored round are different from the pricing dynamics in a venture-fund-anchored round: the sovereign funds are more willing to accept lower implied IRRs in exchange for size, predictability, and the structural protections that come with being the largest LP in any round. The pattern is reshaping the AI-infrastructure primary market in ways that the consultant-class narrative has not yet caught up to.

The third implication is the most subtle. The inter-fund coordination patterns suggest that the sovereign-fund cohort is treating the AI-infrastructure layer as a shared institutional priority rather than as a competitive opportunity. The collaborative posture is structurally different from the more competitive posture the same funds have historically adopted in technology investing — particularly in the late 2010s, when several of the same funds competed aggressively for direct positions in late-stage tech companies. The shift to a collaborative posture in AI infrastructure reflects both the scale of the capital required (which exceeds what any single fund prefers to underwrite alone) and the structural nature of the asset class (which rewards long-duration patient capital and is less well-suited to the competitive bidding dynamics that characterised the late-2010s tech-direct cycle). The collaborative posture is one of the cleanest signals of how the sovereign funds are thinking about the asset class strategically. They are not competing for it. They are building it together.

Data-centre real estate: the layer where the sovereign funds are most aligned

The data-centre real-estate layer, at approximately $9.6B in aggregate Q1 commitments across the eight named funds, is the most cohesively positioned segment of the sovereign-fund AI-infrastructure book. Every one of the eight funds disclosed meaningful data-centre real-estate exposure in Q1, and the structural choices across the cohort have been more similar than the broader positioning patterns. The exposure is split roughly evenly between specialist REIT positions (Digital Realty, Equinix, CyrusOne in its post-take-private continuing-investor structure, and the Stargate-aligned development-stage REITs that closed primary equity raises in late 2025), direct land and grid-adjacent asset positions (Northern Virginia, Quincy, Singapore, Frankfurt, Dublin, and an emerging set of secondary markets including Phoenix, Austin, and several Northern European cities), and direct joint ventures with hyperscaler development partners that the funds have entered into through structured co-investment vehicles.

The structural attractiveness of the data-centre real-estate layer to the sovereign-fund cohort reflects several factors. First, the asset class is operationally familiar: the sovereign funds have been investing in commercial real estate for decades and have built the diligence and asset-management capabilities that data-centre real estate requires, with the AI-specific operational requirements (cooling, power, network connectivity) being incremental additions to the standard commercial-real-estate diligence rather than fundamentally novel competences. Second, the revenue streams are typically structured as long-dated lease commitments with creditworthy hyperscaler tenants, which produces the kind of stable cash flow that sovereign funds explicitly value. Third, the asset class has demonstrable pricing power in the current cycle: the lease rates being achieved on new data-centre development have risen materially across 2024-2025 and into Q1 2026, reflecting the supply-demand imbalance in the underlying inference compute market. The combination of operational familiarity, stable cash flow, and pricing power makes the asset class one of the most institutionally appropriate AI-infrastructure exposures available to the sovereign-fund cohort.

The specific positioning within the data-centre real-estate layer reveals subtle preferences across the cohort. GIC has been the most aggressive in Asian markets, with significant Q1 positions in Singapore, Tokyo, and Mumbai-area developments. ADIA and Mubadala have concentrated more in the North American market, with both funds holding meaningful exposure to Northern Virginia, the Texas Triangle, and the emerging Phoenix and Salt Lake City markets. NBIM, constrained to liquid positions by its mandate, has focused on the major US-listed specialist REITs (Digital Realty, Equinix) and a small set of EU-listed equivalents. PIF's NEOM data-centre development programme is the largest single direct-development position any sovereign fund has disclosed in Q1, with the programme structured around the Saudi state's strategic positioning of NEOM as a regional AI infrastructure hub. The Future Fund and Khazanah have concentrated in their respective home regions (Australia and Southeast Asia), reflecting both the funds' geographic mandates and their assessment that the regional opportunity set is sufficiently attractive to justify the focused positioning.

A subtle but important pattern is the increasing willingness of the sovereign-fund cohort to take direct development risk rather than purchase stabilised assets. The traditional sovereign-fund real-estate exposure has been weighted toward stabilised assets — completed buildings with lease income — and away from development-stage exposure, on the basis that stabilised assets fit the funds' risk profile better. The Q1 2026 data-centre real-estate positioning shows a measurable shift toward development-stage exposure across at least five of the eight funds, with the development positions typically structured as joint ventures with experienced operating partners. The shift reflects both the supply-side constraint in the data-centre market (there are not enough stabilised assets available to absorb the capital the cohort wants to deploy) and the cohort's view that the development-stage returns on data-centre real estate are sufficiently attractive to justify the additional risk. The architectural choice is a meaningful departure from the cohort's historical real-estate positioning and is one of the cleanest signals of how the AI infrastructure investment thesis is reshaping sovereign-fund allocation patterns.

What to watch

The Q1 2026 sovereign-fund commitments to AI infrastructure mark the largest single-quarter allocation the cohort has disclosed. The next three quarters will tell us whether the positioning is structural or vintage-specific.

• Whether GIC, NBIM, ADIA, and Mubadala expand their direct foundation-model exposure through the rest of 2026, or whether the infrastructure-only positioning holds; the Temasek and Future Fund willingness to write direct foundation-model positions is producing performance optionality that the infrastructure-only funds will face pressure to access if the layer continues to perform.
• Whether the gas turbine and SMR commitments produce the kind of supply-side relief that the sovereign positioning is implicitly betting on; the gas turbine deliveries are scheduled to come online in 2026-2028 and the SMR programmes are 2028-2030 commercial-scale targets, and the actual delivery cadence will determine whether the power-side bet pays off or whether the positions become long-dated underperformers.
• Whether the inter-fund coordination patterns sustain through a more competitive primary-round environment; the spring 2026 deal flow has been characterised by relatively coordinated sovereign positioning, but a sharper competitive dynamic in the second half of the year — particularly if a major foundation-model-layer fundraise draws bidding from multiple sovereigns at differentiated terms — would test whether the collaborative posture holds.
• Whether NBIM's selective public-equity-only positioning produces meaningfully different returns from the more privately-anchored positioning of GIC, ADIA, and Mubadala; the NBIM mandate constrains the fund to liquid positions, and the resulting structural under-exposure to the private market may produce a meaningful return differential over the next 18-24 months that the fund's stakeholders will have to interpret either as a feature of the mandate or as a problem with it.
• Whether the PIF HUMAIN commitment and the NEOM data-centre development programme deliver on their stated timelines; the PIF positioning is the most strategy-aligned of the eight funds and the largest single AI-infrastructure commitment any sovereign disclosed in Q1, and the execution risk on both programmes is meaningfully higher than the execution risk on the more conventionally structured commitments at GIC, ADIA, or NBIM.

Frequently asked

Why are the sovereign funds favouring AI infrastructure over foundation-model direct exposure?

The cohort's collective answer, articulated across multiple Q1 investor communications, has two components. First, the AI-infrastructure layer's unit economics are more legible to sovereign-fund diligence: the asset class has identifiable revenue streams (compute capacity at known prices, power generation at known prices, real-estate rents at known prices) and a clearer link between capital expenditure and asset value than the foundation-model layer. Second, the staffing required to underwrite the foundation-model layer is significantly more specialised than the staffing required to underwrite the infrastructure layer, and several of the larger funds have not yet developed the in-house competence to do model-layer diligence at scale. The result is a structural preference for the infrastructure layer that reflects both institutional risk preference and institutional capability.

How significant is the $42B aggregate Q1 commitment relative to the funds' total AUM?

The aggregate $42B figure is meaningful in absolute terms but small relative to the cohort's total AUM. The eight funds in this analysis manage approximately $9.4 trillion in aggregate AUM. The Q1 AI-infrastructure commitment of $42B therefore represents approximately 0.45 per cent of aggregate AUM in a single quarter, an annualised pace of roughly 1.8 per cent — well within the funds' normal portfolio-rotation cadence. The figure is more interesting as a directional signal than as an absolute allocation: it tells us the cohort is positioning meaningfully in the AI-infrastructure layer, but it does not indicate the layer is dominating the funds' overall allocation profile.

What does the gas turbine and SMR positioning imply about the cohort's view of AI infrastructure power supply?

The positioning reflects a thesis that the AI infrastructure build-out is power-supply constrained within a 24-36 month horizon and that the gas turbine and SMR cohort will provide partial relief of the constraint at prices that meaningfully exceed the marginal cost of supply. The thesis is consistent with current AI-infrastructure demand forecasts (35-45 per cent inference compute demand CAGR through 2030, per several published industry forecasts) and with the observed grid-interconnection queue lengths in major data-centre markets (interconnection requests in PJM and ERCOT now exceed 200 GW in aggregate). The sovereign positioning effectively expresses a bet that the power-supply layer will be the binding constraint and the highest-rent layer of the build-out, and the $7.4B aggregate Q1 commitment to energy-layer positions reflects the size of the bet the cohort is willing to express.

How does the inter-fund coordination affect AI-infrastructure pricing?

The inter-fund coordination patterns produce a different pricing dynamic in sovereign-anchored AI-infrastructure rounds than would prevail in venture-fund-anchored rounds. Sovereign funds are typically more willing to accept lower implied IRRs (in the high-single-digit to low-double-digit range, versus the 20-25 per cent venture-fund target) in exchange for size, predictability, and structural protections. The result is that AI-infrastructure primary rounds anchored by sovereign syndicates tend to price at higher valuations than equivalent rounds anchored by venture or growth-equity funds would. The pricing differential is partly responsible for the strong primary-round valuation environment in AI infrastructure across late 2025 and early 2026, and it is one of the structural features of the current cycle that the consultant-class narrative has not fully integrated.

Why is Temasek's positioning so different from the other sovereigns'?

Temasek's institutional history positions the fund differently from the other seven sovereigns in technology investing. The Singapore-resident fund has been investing in tech-direct positions for more than a decade — including direct positions in Alibaba, Bytedance, Stripe, Indigo Agriculture, and others — and has built the staffing required to underwrite at the model layer in a way that several of the other sovereigns have not. The fund's Q1 AI-infrastructure positioning preserves the historical tech-fluent posture while extending it into the AI-specific opportunity set, including the Anthropic secondary purchases that no other sovereign matched in scale. The Future Fund's positioning is similar in posture but at smaller scale, reflecting both the smaller balance sheet and the more conservative Australian sovereign-fund tradition. Both funds are positioned to capture model-layer returns that the other six are accessing only indirectly through GP commitments.

What happens to the sovereign positioning if AI-infrastructure returns disappoint over the next 18-24 months?

The cohort's positioning is structured to be durable through a return drawdown. The infrastructure-layer assets (data centres, gas turbines, SMR development positions) are long-duration physical assets with multi-decade payoff profiles, and the sovereign funds are explicit that they are underwriting on a 10-20 year horizon rather than on a near-term return basis. The GP commitments are similarly long-duration, with the typical AI-specialist fund having a 10-year primary fund life plus extensions. The infrastructure positioning is therefore less vulnerable to near-term return disappointment than the equivalent venture or growth-equity positioning would be. The cohort's primary risk is not return-driven exit pressure but capital-deployment-pace risk: if the AI-infrastructure capital expenditure programmes accelerate beyond the supply-side response capacity, the cohort's existing positions will gain in value but the marginal new allocation will face less attractive pricing. The funds' Q1 disclosures suggest they are managing this risk by accelerating their existing relationships rather than chasing new ones.