$16T+
Traditional private capital under management.
AFAQ AI · Special Report
How artificial intelligence is being absorbed into the infrastructure of private capital — and the standard of evidence that will decide which of it endures.
Contents
Foreword
Private capital has become one of the largest pools of institutional money in the world, and the least legible. The systems that record what it owns, what those holdings are worth, and how that figure was reached were built in layers over two decades. Artificial intelligence is now arriving on top of that inheritance.
This report is written for the people who carry the consequence of a valuation '—' investment teams, finance functions, and the committees they answer to. It takes the position that the interesting question is not how capable the models are, but which of their output an institution can actually stand behind.
The argument runs in three movements. Scale and illiquidity have made valuation the central problem of private markets. AI compresses much of the analytical work but leaves the standard of evidence exactly where it was. And the durable advantage belongs to whoever can pair the speed of the model with a record disciplined enough to be questioned.
Executive summary
Seven observations frame the report.
Traditional private capital under management reached roughly sixteen trillion dollars. As exits slowed and holding periods lengthened, the interim mark — not the eventual sale — became the figure institutions are asked to trust.
Data capture, administration, monitoring, valuation, and reporting were each solved separately. The largest platforms are now absorbing the rest.
Private data arrives unstructured, on a quarterly cadence. Extraction and retrieval are where AI is applied first — the highest manual cost, the lowest judgement.
Almost every large manager runs generative AI. Only about a quarter report substantial change. The constraint is data, governance, and trust — not the model.
The standard a valuation must meet — traced inputs, recorded assumptions, a result that holds under audit — does not move because the draft was produced faster.
A number without a source is a liability long before it is an error. The discipline that matters is provenance, and keeping a model out of the path that computes the figure.
Models improve and converge; provenance, determinism, and an immutable record solve a governance problem that improvement does not erase.
Part I
Private capital grew into an institutional asset class faster than the discipline to value it. Scale, and the illiquidity that came with it, moved the burden of proof onto the mark.
Scale
Traditional closed-end private funds hold on the order of sixteen trillion dollars, and capital in newer structures has grown alongside them into the mid-single-digit trillions. The moment private capital crossed into institutional scale, it created a measurement problem that only systems could carry.
Illiquidity
Capital has been raised more quickly than it can be deployed, and the undeployed balance has aged. When an asset is held for seven years instead of four, it is revalued many more times before it is ever sold '—' and each interim mark is read more closely than the last.
Part II
A stack assembled in layers, by different vendors, for different parts of the workflow — data capture, administration, monitoring, valuation, reporting. It is consolidating, but the hardest layer resists it.
The data problem
Private-market data does not arrive ready to use. It comes as documents '—' capital-call notices, financial statements, board packs '—' issued quarterly and formatted however the manager chose. Extraction, classification, and retrieval are where AI has moved first. It is real progress, but it is the foundation of the building, not the building.
Part III
Of the five layers, valuation is the hardest to evidence and the most consequential to get wrong. It is the node where AI, regulation, and governance all converge.
The hierarchy
Accounting standards rank a valuation by how observable its inputs are. Private holdings sit at the bottom '—' Level 3 '—' valued on inputs that cannot be observed in any market, where the burden of evidence is heaviest.
Pressure
The conditions that lifted private returns for a decade have passed. Top-quartile buyout performance in 2025 fell to a single-digit pooled return, less than half of what public indices delivered '—' exactly the condition under which a disciplined method matters most.
Part IV
AI compresses the analytical layer of valuation. The evidence is clear that this is real, and equally clear that it has not yet shown up as results.
The gap
Almost every large manager now runs generative AI across multiple use cases. Far fewer can point to a substantial change in their results. The constraint is not access to capable models; it is data that is not clean, processes that are not governed, and outputs that cannot yet be trusted in a regulated decision.
Part V
The standard a valuation must meet is fixed by regulation and governance, not by tooling. Speed does not move it.
The real risk
The larger problem is quieter than invented facts: a figure that is plausible, useful, and wrong about its own origin '—' an estimate that has quietly become a fact because nothing recorded that it was ever assumed. The discipline that answers this is provenance: every input carries its state '—' cited, inferred, or missing, and the three are never blurred.
Part VI
For sovereign and central capital, the constraints are sharper. Data cannot leave the jurisdiction, governance is non-negotiable, and generic tools stop at the door.
The constraint
For sovereign wealth funds and the institutions that serve them, the choice of AI is a matter of permission, not preference. A capable model that requires sending holdings to an external endpoint is, for these institutions, simply unavailable. The infrastructure has to run where the data lives, with provenance, determinism, and an immutable record present from the first day.
Part VII
The durable layer is not a model and not a feature. It is infrastructure: computation under fixed rules, inputs traced to source, and a record built to be questioned.
The principle
A valuation an institution can stand behind has three properties, and a system that lacks any one of them is a tool, not infrastructure.
The difference
The contrast is not speed. A capable team with a spreadsheet can produce a number. What it cannot reliably produce, across hundreds of holdings and many reporting cycles, is the same number twice, traced to its sources, and recoverable when a board asks why it moved.
AFAQ AI builds toward the right-hand column for sovereign and institutional portfolios, from Muscat. It gives the judgement of the people who manage capital a record that holds up under the people who will examine it.
Part VIII
The next few years will be decided less by which models win than by which firms can make their output accountable.
Models converge and commoditise; provider choice matters less each year.
Document extraction becomes table stakes, not advantage.
Agentic workflows handle more of the analytical draft.
The stack consolidates onto fewer, broader platforms.
The standard of evidence a board and auditor require.
The need to trace every input to its origin.
The discipline of keeping a model out of the final number.
The demand, for sovereign capital, that infrastructure be governable and local.
In closing
Osama Al Zadjali, Founder and Chief Executive Officer
Methodology and sources
Figures are drawn from recognised third-party research and are rounded. Market-level data is taken from the McKinsey Global Private Markets Report and from Preqin and Bain. AI-adoption data is taken from EY, ThoughtLab and Grant Thornton. The fair-value framework referenced is IFRS 13; private-asset valuation practice follows the IPEV guidelines. Charts described as frameworks illustrate structure, not data. Nothing in this document constitutes investment advice, an offer of any security, or a recommendation.
Published by AFAQ AI. Part of AFAQ AI Insights, written for the people who carry the consequence of an institutional valuation.
AFAQ AI · Muscat, Sultanate of Oman · AFAQ / SR · 01