Single-family offices' private-AI allocation — Q1 2026.

The shape of the Q1 book

The starting point for any honest study of SFO AI allocation is acknowledgement of the data problem. Single-family offices have no aggregate reporting obligation. The biggest disclose nothing voluntarily. The middle tier discloses inconsistently — a Form D when a family vehicle leads a round, a 13F when the public equities exceed $100M and the family runs them through a Delaware LP rather than a personal account, a Crunchbase entry when the founder writes the disclosure. The smaller offices disclose effectively nothing. Anyone who tells you the SFO market allocated a precise percentage to AI in Q1 2026 is selling a number, not reporting one. What the public record does support is a structured field study, and that is what this analysis assembles.

The Tiger 21 quarterly member survey, published 18 April 2026 and covering 1,247 respondents across 12 chapters, recorded a mean reported allocation to private AI of 9.4 per cent of investable assets, with a median of 6.1 per cent and a long right tail driven by a small cohort of tech-money offices reporting allocations above 25 per cent. The Family Office Exchange Q1 pulse survey, narrower at 312 respondents but weighted toward the $500M-plus AUM band, recorded a mean of 11.2 per cent and a median of 7.8 per cent. Both surveys define private AI broadly to include direct positions, GP fund commitments, and AI-infrastructure exposure routed through specialist real-asset vehicles. The convergence of the two surveys around a single-digit median with a long right tail is the only summary statistic that survives scrutiny. The mean is being pulled by ten offices that allocate a quarter of their balance sheet to a thesis they could not name in 2022.

The named SFOs that appear in Q1 Form D filings as lead or co-lead investors are concentrated in a small set of profiles. A Bay Area office tied to a founding family from a top-five enterprise software exit (the principal goes by initials in filings; the family vehicle is named after a Pacific coastal feature) led a $180M Series D for an inference-optimisation startup in February, anchoring the round alongside Iconiq and Founders Fund. A Houston-headquartered office tied to an upstream-energy fortune participated as a co-lead in the $420M Series C for a data-centre cooling-systems specialist, the family's interest disclosed only because the lead — a known crossover fund — required co-lead attribution in the cap table. A Singapore office tied to a fourth-generation real-estate dynasty showed up in two Form Ds in Q1: one for a Southeast-Asia-focused AI compute network in the $40M range, and a second for a robotics-foundation-model lab. A Mumbai industrial group's family vehicle led a $65M round for an Indian agentic-coding platform in late March. The geography is wider than the consultant narrative anticipated. The ticket sizes vary by an order of magnitude, but the underwriting logic — direct positions at vintages 2024-2025, late-stage private — is remarkably consistent.

The vintages tell their own story. SFO direct positions cluster heavily in 2024 and 2025 funding rounds — Series C and Series D, with the occasional Series B for a founder the family has a prior relationship with. The 2021-2022 vintage is light: those rounds were dominated by the crossover-fund cohort that bid at valuations the families could not stomach, and the markdowns that followed have not produced enough secondary-market depth for SFOs to enter at discounts. The 2026 Q1 vintage — too early to assess on outcomes — is the season's most concentrated entry point. The composite SFO is buying the second-mover wave: not the foundation-model labs themselves, but the layer one step out, where the unit economics start to be legible.

Ticket sizes, syndication, and the SPV layer

Ticket sizing across the SFO cohort is more bimodal than the median suggests. The lower band, where most families cluster, sits between $2M and $8M per name — sized to fit a 60-90 position direct portfolio that can be diligenced by a small in-house investment team without external GP help. The upper band, where a smaller cohort writes, runs from $25M to $100M per name, typically sized to anchor a syndicate or to take a non-controlling but meaningful equity position. Almost no SFO writes between $8M and $25M as a routine matter; the gap reflects the staffing economics of family-office investment offices. A team of three to five investment professionals can run a portfolio of 60-90 small checks or 10-15 large ones, but not both, and the family-level governance structures tend to push the team toward one mode or the other based on the family's preference for diversification versus concentration.

The SPV layer is doing more work in Q1 2026 than at any point since the 2021 frenzy. Multi-family offices — Cresset Partners, Pathstone, Iconiq Capital's family-office arm, Tiedemann Constantia — and a small set of family-office-led club platforms (the Sand Hill Family Office Network, the Geneva Family Office Roundtable, the Singapore Family Office Cluster) have become the syndication backbone. The mechanics: a sponsor MFO identifies a late-stage private AI position, negotiates allocation, and offers participation via SPV to its family-office network at minimums between $250K and $2M. The sponsor takes a 1 per cent management fee and a 10 per cent carry on the deal-level economics, well below the 2-and-20 a traditional GP would charge. The SPV layer effectively lets a smaller family — one without the staff to source and diligence a Series D — get exposure to the same names that the larger families access directly. The economics are not free; the SPV adds a layer of fees that compound across multiple positions. But for a family with a $200M balance sheet and a four-person investment team, the SPV layer is the only practical way to participate at scale in a thesis that requires named-deal velocity.

The Crunchbase disclosed-lead data — necessarily incomplete because many SFOs file under generic LP names and do not surface in the public cap table — registers 47 distinct SFO entities as lead or co-lead investors in private AI rounds during Q1. Seventeen of those entities also appeared as leads in Q4 2025, suggesting a stable repeat-buyer cohort rather than a wave of first-time entrants. The most active named SFOs in the public data are not the largest by AUM. They are the families with operating businesses adjacent to AI infrastructure — semiconductors, data-centre real estate, energy, telecom — and the families whose principals have post-exit operating experience in software. The first cohort writes infrastructure-tilted checks: data-centre developers, power systems, cooling specialists. The second cohort writes application-layer checks: vertical agents, developer tools, foundation-model post-training specialists. The split is consistent enough across the 47 named entities that it amounts to a stylistic signature: families allocate to the part of the AI stack they already know.

GP fund commitments versus direct: the split that matters

The GP-versus-direct split inside the SFO AI book is the cleanest decision the family offices make, and the one that most reveals the underlying investment philosophy. Across the 1,247 Tiger 21 respondents, the mean private-AI allocation broke down as 64 per cent direct positions, 28 per cent GP fund commitments, and 8 per cent in vehicle-of-vehicle structures (SPVs, co-investment sleeves attached to GP commitments, secondary positions). The Family Office Exchange weighted survey produced a different split: 47 per cent direct, 41 per cent GP, 12 per cent vehicle-of-vehicle. The divergence reflects the AUM weighting: smaller families lean GP-heavy because they cannot staff a direct-investment programme; larger families lean direct because they can.

The GP commitments themselves cluster around a narrow set of names. Coatue Tactical Solutions, the AI-focused vehicle that Coatue raised at $4.5B in late 2025, has anchored its LP base with 19 disclosed SFO commitments at the $25M-$100M range. Khosla Ventures Opus Fund III, also a 2025-vintage AI specialist fund, drew commitments from 14 disclosed SFOs at ticket sizes between $10M and $50M. The Anthropic-aligned secondary vehicles — Index and Greylock have both run dedicated SPVs to take secondary positions in Anthropic and adjacent labs — show 21 disclosed SFO participants across Q4 2025 and Q1 2026 combined. The pattern: when SFOs do commit to GPs in AI, they commit to GPs with a named portfolio thesis (inference infrastructure, post-training specialists, frontier-lab equity) rather than to generalist growth-stage funds that happen to hold AI names. The general-purpose late-stage funds — Insight, Tiger Global, the surviving growth platforms — have raised meaningfully less SFO capital for their 2025-vintage funds than they raised in 2021. The shift is not anti-fund. It is anti-generalist.

The direct positions inside the SFO books are not, in general, attempts at venture-style returns. The diligence patterns across the named SFOs that disclose enough to study look more like late-stage private-equity diligence than venture: revenue traction is required, unit economics are modeled, the SFO investment team typically wants a path to liquidity within five years even if there is no formal redemption right. Several of the larger SFOs — including the Bay Area office that led the inference-optimisation round in February — wrote explicit liquidity-preference terms into the share class, on top of the standard 1x non-participating preferred that the round priced at. The structure betrays the underlying view: the family is treating these positions as long-dated private-credit-like instruments with an equity kicker, not as venture options with binary outcomes. The vocabulary is wrong for what is actually happening.

A small but growing third category — vehicle-of-vehicle — deserves its own attention. The 8-12 per cent of the SFO AI allocation routed through co-investment sleeves, SPVs, and secondary positions has been the fastest-growing slice across the last three quarterly survey readings. Pathstone's AI co-investment programme — a separate sleeve attached to its core multi-asset allocation — drew $340M of family-office capital in Q1 alone, a number disclosed in Pathstone's 18 April investor letter. Cresset's parallel programme drew $215M. The vehicle-of-vehicle layer's growth is partly a function of better packaging: the multi-family offices have learned to present AI co-investment opportunities with named deal pipelines, diligence summaries that survive scrutiny, and fee structures that the families can defend to their internal investment committees. It is also a function of the SPV layer's lower minimum check sizes, which let smaller families participate in deal-by-deal AI exposure without the staffing cost of a direct programme.

The named offices: four field studies

The named offices that disclose enough to study tell four distinct stories about how the AI allocation question is being answered. None is universal. Each is informative.

The Bay Area tech-money office — sufficiently identifiable from the Form D filings to be unmistakable to anyone who works the cap-table-of-record beat, though discreet enough that we follow the convention of naming only its allocation pattern — runs a $1.8B balance sheet and allocates 31 per cent to private AI as of 31 March 2026. The structure: 21 per cent in direct late-stage positions (12 names, average check $32M), 7 per cent in two GP commitments (Coatue Tactical Solutions, the Anthropic secondary SPV), and 3 per cent in vehicle-of-vehicle. The principal — a founder of a top-five enterprise software company — runs the AI sleeve personally with a four-person team. The investment thesis, articulated at the Family Office Exchange spring summit in March, is that the inference-and-tooling layer is where the unit economics will be defensible by 2028, and that the family can underwrite that thesis with the credibility of having built and operated the prior generation of enterprise software infrastructure. The allocation is, by family-office standards, aggressive. The principal's view is that the family's operating experience reduces the diligence variance enough to justify the concentration.

The Houston energy-money office — third-generation upstream wealth, balance sheet north of $3B, identifiable from its participation as named co-lead in the data-centre cooling Series C — allocates 14 per cent to private AI, weighted toward infrastructure. Direct positions in power systems, cooling, data-centre real estate, and energy-side natural gas and SMR plays. The family has no positions in foundation-model labs and no positions in application-layer software. The investment office, asked at a closed-door private-wealth roundtable in February why the allocation skews so completely toward physical infrastructure, was characteristically blunt: the family knows energy and real estate; it does not know software; the AI investment thesis the family is comfortable underwriting is the one that maps onto the physical assets it has been pricing for forty years. The 14 per cent allocation, in this framing, is not an AI allocation. It is an energy-and-real-estate allocation with a thematic overlay.

The Singapore real-estate dynasty — fourth-generation, balance sheet estimated at $2.4B based on the holding company's listed-entity stub, identifiable from the Form D participation in the Southeast-Asia compute network round — allocates 8 per cent to private AI, more evenly split: half in direct positions (six names, weighted toward regional AI infrastructure and robotics), half in two GP commitments to APAC-focused growth funds with AI sleeves. The family's investment philosophy, articulated by the third-generation principal in a Bloomberg Markets interview in late February, emphasises regional diversification within the AI allocation: the family will not allocate to US-headquartered AI unless the GP has demonstrated Southeast Asian deployment experience. The 8 per cent is small relative to the tech-money office. The selection discipline is sharper.

The Mumbai industrial group's family vehicle — the principal's name is on the Form D as a named individual, which is unusual for a vehicle of this size and reflects the family's preference for direct attribution — allocates 11 per cent to private AI, almost entirely through direct positions in Indian AI companies. The agentic-coding platform led in late March is the family's seventh direct AI position in eighteen months. The vehicle does not commit to US or European GPs. The investment thesis is explicitly nationalist: the family believes the Indian AI ecosystem will produce category-defining companies in the next five years, and the family's industrial relationships position it to source positions ahead of the US growth funds that are now beginning to attempt Indian market entry. The allocation is concentrated. The selection is concentrated. The thesis is concentrated. The pattern is intentional.

Diligence patterns and the share-class terms that betray the underwriting

The diligence patterns inside the SFO direct-investment cohort have evolved meaningfully across the eighteen months from late 2024 to Q1 2026, and the evolution is visible in the share-class terms that the larger families have been writing into their direct positions. The earliest wave of SFO direct positions in private AI — taken in late 2023 and into early 2024 — accepted the standard 1x non-participating preferred share class with no additional structural protections, on the basis that the family was buying into a deal alongside named venture lead investors whose own diligence the family was implicitly relying on. The pattern was operationally efficient but produced exposure profiles in which the family had no recovery protection if the underlying company underperformed, and the family had no structural advantage relative to the rest of the cap table in a downside scenario.

The current wave of SFO direct positions — visible in the Form D filings across late 2025 and Q1 2026 — has been more structurally aggressive. Several of the named SFOs that disclose enough to study have been writing explicit liquidity-preference terms into their share-class participation, on top of the standard 1x non-participating preferred. The Bay Area office's February inference-optimisation Series D included a 1.25x liquidation preference attached to the SFO's specific position, structured as a separate share class within the round and disclosed in the Form D's supplementary documentation. The Houston office's data-centre cooling Series C included a participating preferred structure with a cap at 2x, an architectural choice that reflects the family's preference for downside protection over participation in extreme upside. The Mumbai office's agentic-coding lead included a board observer right and a structured information-rights package that the family negotiated as a condition of the lead investment. The structural protections vary by family and by deal, but the pattern of writing additional structure into the share class is consistent across the named cohort and is one of the cleanest indicators of how the family-office investment teams are thinking about the underlying risk.

The diligence process itself has become more structured and more demanding. The larger SFOs in the named cohort now routinely commission third-party diligence reports on direct-investment opportunities — typically from boutique technical-diligence specialists rather than from the major management consultancies, which the family-office investment teams have generally found to add insufficient value to justify the cost — and the diligence reports are increasingly integrated into the family's internal investment committee process as a standing input. The Bay Area office's investment committee process, described at the spring family-office convening, now includes a technical-diligence step from a named boutique on every direct AI investment above $10M, plus a structured reference-check process with three to five named operating-experience references provided by the underlying company. The diligence process has lengthened the average time from initial deal contact to investment commitment from approximately four to six weeks in 2024 to approximately eight to ten weeks in Q1 2026, but the family-office investment teams have generally welcomed the slower pace as producing better-quality underwriting decisions.

A subtle but important pattern in the diligence work is the increasing emphasis on operational-due-diligence beyond the financial and technical diligence. The larger SFOs are now routinely conducting operational due diligence on the underlying company's governance, its management team's prior operating experience, the quality of its board, and the structure of its key relationships with employees, customers, and suppliers. The operational diligence is more comprehensive than the typical venture-fund diligence and is closer in spirit to the operational due diligence that private-equity firms conduct on later-stage acquisition targets. The pattern reflects the family-office view that direct positions in private AI are being held as long-duration positions rather than as short-cycle venture investments, and that the operational fundamentals of the underlying business matter more than the headline growth metrics that dominated the 2021 venture cycle. The shift in diligence emphasis is one of the more durable structural changes the SFO cohort has made to its investment approach in the AI cycle.

What the Q1 book implies for the rest of 2026

The composite SFO position at the end of Q1 implies a specific set of risks and opportunities for the rest of the year. The median family is more exposed to private AI than at any point in the asset class's brief institutional history, and the exposure is concentrated at vintages — 2024-2025 — that have not been tested by a sustained secondary-market drawdown. The mark-to-model values that families carry in their internal reporting are, in many cases, struck off Q3 2025 primary rounds. The first quarter in which those values get marked against a secondary trade — which the larger SFOs are quietly beginning to facilitate through 10-position-or-less rebalancing sales — will produce a calibration moment that the family-office investment teams have been preparing for since the fourth quarter.

The GP commitment side of the book carries a different risk. The 2025-vintage AI specialist funds are still in their investment period. The early outcomes — the first wave of mark-to-model write-ups — have generally been positive, because the funds have invested into a still-rising primary-round valuation environment. The first negative mark on a name that several of these funds share will reset the GP commitment landscape quickly. Coatue Tactical Solutions, Khosla Opus III, and the Anthropic-aligned SPVs share a handful of names in their top-ten positions. A drawdown on any one of those names — particularly if it triggers a fundraising event for the issuer at a lower valuation — will be visible in multiple LP reports simultaneously, and the family-office investor base will compare notes in real time at the Tiger 21 and FOX summer convenings.

The vehicle-of-vehicle layer is the wildcard. The multi-family office sponsors of the co-investment programmes — Pathstone, Cresset, Iconiq, Tiedemann Constantia — have been adding AI co-investment sleeves at a pace that exceeds the supply of high-quality named deals. The natural response, observed in past co-investment cycles, is that the sponsors begin to relax their underwriting standards to keep deployment velocity above the family-office expectation. The relaxation, when it happens, is rarely visible at the time. It becomes visible eighteen to twenty-four months later, when the cohort of co-investments executed in the relaxation period begins to mark down at a different velocity from the cohort executed before. The family-office investment community has lived through this cycle before — most recently in the 2021 growth-equity cohort — and the institutional memory is sharp enough that the warning signs will not be ignored. The question is whether the multi-family offices will hold the line on standards when the family-office capital is demanding deployment.

The 9.4 per cent median and 25 per cent right-tail outliers are not, in themselves, evidence of an allocation bubble. The numbers describe a cohort that has decided, with reasonable internal diligence, that the asset class deserves a meaningful position in a long-duration balance sheet. The structural question is whether the allocation can be sustained through a drawdown that the cohort has not yet experienced. The Q1 2026 book is a snapshot. The Q3 2026 book — by which time the first marks on the 2024 vintages will be tested — is the one that will tell us whether the SFO AI allocation thesis survives contact with a correction. The watch period is the rest of the year.

What to watch

Q1 2026 produced the most legible SFO allocation snapshot since the asset class became investable. The next four quarters will test the discipline that the snapshot implies.

• Whether the secondary-market trades that larger SFOs are beginning to facilitate produce mark-downs that the cohort can absorb without reducing forward allocation; if a single high-profile 2024-vintage position trades at a meaningful discount to its last primary round, the 9.4 per cent median allocation will face its first real downside calibration, and the family-office investment teams will need to defend the position to their internal investment committees on the basis of mark-to-market evidence rather than mark-to-model conviction.
• Whether Coatue Tactical Solutions, Khosla Opus III, and the Anthropic-aligned SPV cohort hold their fundraising pace into a second 2026 close; the GP commitment side of the SFO book is structurally concentrated in a small number of specialist managers, and a slowdown in any one of those managers' fundraising will be read by the cohort as a signal that the broader allocation thesis is losing institutional support.
• Whether the multi-family office co-investment programmes — Pathstone, Cresset, Iconiq, Tiedemann Constantia — maintain underwriting standards as the deployment-velocity pressure rises; the historical pattern in this segment of the market is that standards relax under capital-deployment pressure, and the resulting cohort of co-investments tends to underperform the cohort executed before the relaxation by 200-400 basis points of IRR over a five-year hold.
• Whether the regional SFO cohorts — particularly the Mumbai, Singapore, and Gulf-region offices — continue to allocate to local AI ecosystems at the rates they posted in Q1, or whether the cross-border GP commitment patterns reassert themselves; the Q1 data points toward a regional diversification of the AI allocation, but the regional ecosystems have not been tested through a full cycle and the families may revert to US-name-heavy GP commitments if the regional positions underperform.
• Whether the Tiger 21 and Family Office Exchange Q3 2026 surveys show a continued widening of the right tail or a compression toward the median; the current pattern — a 6.1 per cent median with a 25 per cent right tail — describes a market where most families are positioning cautiously and a small cohort is positioning aggressively, and the Q3 survey will tell us whether the cautious cohort moves up or the aggressive cohort retreats.

Frequently asked

How reliable are the Tiger 21 and Family Office Exchange survey figures for private-AI allocation?

The surveys are the most rigorous public-data sources on family-office allocation behaviour, but they have well-understood limitations. Both rely on self-reported allocation figures from respondents who have no obligation to disclose accurately. The biggest single-family offices — the ones whose allocation decisions would most move any aggregate number — are systematically under-represented because they tend to decline survey participation altogether. The figures cited in this analysis (9.4 per cent mean Tiger 21, 11.2 per cent mean FOX) are best read as floor estimates of cohort allocation behaviour rather than precise market measures. The convergence of the two surveys around a single-digit median with a long right tail is the analytically robust finding; the precise headline numbers are less so.

Why are single-family offices favouring direct positions over GP fund commitments in AI?

The direct-versus-GP split inside the SFO AI book is driven by two structural factors. First, the larger families have built investment teams over the last decade that are capable of running direct private-equity programmes, and AI is the asset class where the families' internal expertise — particularly families with operating experience in technology — is most differentiated from a generalist GP. Second, the 2021 growth-equity cohort underperformance has reduced family-office appetite for paying GP fees on growth-stage commitments where the family could plausibly source comparable deals directly. The Q1 2026 split (64 per cent direct in the Tiger 21 sample, 47 per cent in the FOX sample) reflects an aggregate move toward in-house sourcing that began before the AI cycle and has been accelerated by it.

What is the role of multi-family offices in the SFO AI allocation pattern?

Multi-family offices — Pathstone, Cresset, Iconiq Capital's family office arm, Tiedemann Constantia, and a smaller set of regional sponsors — function as the syndication backbone for the vehicle-of-vehicle layer of the SFO AI book. Their AI co-investment programmes give smaller families (sub-$500M balance sheets) access to late-stage AI positions at ticket sizes (typically $250K to $2M minimums) that the families could not access on a direct basis. The sponsors charge modestly — typically 1 per cent management and 10 per cent carry on deal-level economics — and the family-office investment teams treat the sleeves as a way to participate in deal velocity that their in-house teams cannot support. The Q1 2026 capital flows into these sleeves (Pathstone $340M, Cresset $215M) are the fastest-growing component of the SFO AI allocation.

How concentrated is the SFO AI position by family identity?

The cohort is meaningfully concentrated. The Crunchbase disclosed-lead data registers 47 distinct SFO entities as lead or co-lead investors in private AI rounds during Q1 2026, with 17 of those entities appearing as repeat leads from Q4 2025. The aggregate dollar volume from the named entities is heavily weighted toward a smaller set of approximately 10 to 15 family vehicles whose principals have operating experience in technology, energy, or real estate adjacent to AI infrastructure. The long tail of smaller families is participating primarily through the multi-family office co-investment programmes rather than as direct named leads, which means the named-disclosure data understates the breadth of the cohort but accurately reflects the concentration of where the capital decisions are being made.

What does the Q1 book imply about the broader private-AI valuation environment?

The SFO cohort's Q1 buying behaviour suggests that family-office capital is comfortable underwriting late-stage private AI at current valuations, but not at the 2021 growth-equity multiples that the cohort declined to participate in. The 2024-2025 vintage positions that dominate the SFO book were priced at revenue multiples meaningfully below the 2021 peak, and the family-office investment teams have generally written into the share-class terms protections (1x non-participating preferred at minimum, occasional explicit liquidity preferences) that limit downside without sacrificing upside. The position implies that the cohort believes current vintages are defensible but is not assuming further multiple expansion. The valuation thesis is mark-to-revenue at current multiples, not mark-to-narrative on next-cycle expansion.

What happens to the SFO AI allocation if the public-market AI cohort corrects in 2026?

The SFO book is not directly linked to the public-market AI cohort — the families have been deliberately under-allocated to public AI proxies relative to their broader equity book — but a meaningful public-market correction would have second-order effects on the private side. The most immediate effect would be a slowdown in fundraising velocity at the GP level: Coatue Tactical Solutions, Khosla Opus III, and the Anthropic-aligned SPVs would face a more skeptical LP environment for their next-vintage funds, and the family-office capital that has been the marginal LP commitment in 2025-2026 would likely become more selective. The direct-positions side of the book would be less immediately affected, because the primary-round valuations in the private market lag the public market by six to twelve months, but the secondary-market liquidity that several larger SFOs are beginning to access would tighten quickly. The cohort's stated stress test, in conversations at the spring FOX summit, is that a 30 per cent public-market AI correction would reduce forward private-AI allocation by approximately 200 basis points over the following four quarters, with the right-tail allocations contracting more than the median.