Why this matters now
AI capital concentration due diligence is the systematic audit performed by limited partners (LPs) and institutional investors to measure, evaluate, and mitigate portfolio risk arising from the disproportionate accumulation of venture capital within a small cluster of frontier model developers and their underlying compute infrastructure. As capital flows consolidate heavily into a few mega-cap labs, institutional portfolios face severe systemic risk, hidden asset overlap, and compressed valuation multiples if foundation models commoditize or fail to convert paper valuations into tangible cash distributions. Conducting rigorous concentration due diligence allows LPs to evaluate fund-level exposure, test revenue durability, and preserve long-term liquidity.
Venture capital markets have entered an era of unprecedented funding asymmetry. A remarkably small number of mega-cap AI laboratories are capturing the vast majority of growth capital across global private markets. Limited partners evaluating venture capital fund allocations must look beyond headline portfolio performance and examine how deeply their capital is tied to the capital-intensive foundation model race. During the first half of 2026, OpenAI and Anthropic alone received over 60% of all venture capital dollars committed to U.S. startups. This extraordinary pooling of funds means that many LPs who believe they have bought a diversified index of early-stage software innovation are in reality holding concentrated exposure to a few massive model builders.
This structural concentration introduces acute risks when market assumptions shift. The emergence of high-performing open-weight and open-source models has begun to challenge the pricing power and scarcity premium historically enjoyed by closed frontier labs. When open alternatives achieve near-parity in core capabilities at a fraction of the inference cost, the high revenue multiples assigned to proprietary model developers come under immediate pressure. If a fund's paper gains rely heavily on unrealized markups in one or two frontier labs, any valuation reset instantly impacts the entire fund vintage. Institutional investors evaluating 2021 vintage funds and subsequent vehicles must audit whether underlying gains represent sustainable cash-generating businesses or temporary markups vulnerable to technological commoditization.
- Massive Capital Pooling: A dominant share of venture investment is absorbed by fewer than five foundation model companies, inflating paper valuations across growth funds.
- Open-Source Commoditization Pressure: Rapid improvements in open-weight models erode the pricing power and gross margins of proprietary frontier labs.
- Concentrated Distribution Bottlenecks: Fund distributions (DPI) become contingent on multi-billion-dollar liquidity events for a tiny cohort of mega-cap entities.
- Systemic Portfolio Overlap: Indirect exposure via portfolio startups building on top of or re-selling underlying models amplifies fund-level concentration.
Consequently, LPs must move beyond basic sector categorization and implement granular exposure mapping across their private market portfolios. Understanding how capital is distributed across the AI stack provides the foundation for evaluating fund risk profiles.
The main practical framework
To perform effective AI capital concentration due diligence, institutional investors must analyze portfolio assets through a three-tiered structural framework: the model layer, the application layer, and the compute infrastructure layer. Capital efficiency, gross margin structures, and competitive moats vary dramatically across these three segments, dictating how risk accumulates within a fund.
Categorizing the AI Value Chain
At the model layer, capital intensity is extreme. Frontier labs require tens of billions of dollars in initial funding to cover hardware training runs, talent acquisition, and safety alignment. While headline revenue can scale rapidly, gross margins are frequently constrained by heavy ongoing compute requirements and continuous model re-training cycles. Evaluating startups in mega-seed capital efficiency shows that early capital burn can easily outpace long-term unit margins if inference costs remain elevated.
Conversely, the application layer features traditional software unit economics, but faces distinct replication risks. Application wrappers that simply interface with third-party model APIs risk rapid margin erosion if the underlying model provider releases competing features or if customers switch to cheaper open-source models. Investors conducting software moat due diligence must verify whether an application company possesses proprietary workflow integration, proprietary data loops, or deep system-of-record status.
The infrastructure layer encompasses physical data centers, specialized chips, and specialized cloud hosts. While infrastructure providers benefit from surging compute demand, they face substantial capital expenditure requirements, rapid GPU hardware depreciation, and complex debt structures. Investors examining neocloud due diligence must evaluate how utilization rates and power capacity agreements impact cash flows.
The three layers therefore behave like three different asset classes. The model layer is the most capital-hungry: training runs, talent and alignment work consume funding at a scale no application company approaches, and margins are held down by the compute bill that never stops, so the dominant diligence question is cost per training cycle against revenue growth and the risk of open-weight commoditization. The application layer needs comparatively little capital and keeps the healthiest unit economics of the three, although inference now sits in cost of goods sold and pulls margins below classic software benchmarks; here the tests are net retention, workflow depth and how exposed the company is to a single model vendor. The infrastructure layer is capital intensive in the physical sense, funded largely with debt against depreciating hardware and contracted power, so utilization rates and debt payback schedules matter more than headline growth. LPs should ask for each of these metrics layer by layer rather than accepting a blended portfolio view.
By mapping every underlying fund holding to these three layers, LPs can quickly establish whether a venture portfolio represents genuine technological diversification or a single macro bet on continuous compute expansion.
What investors are really testing
When institutional LPs audit AI allocations during primary fund commitments or secondary portfolio reviews, they are testing the operational durability and cash conversion potential of the underlying assets. The central question for investment committees is whether impressive headline revenue growth translates into defensible, long-term enterprise value.
Bridging the Pilot-to-P&L Gap
A major area of scrutiny is the growing disconnect between enterprise pilot adoption and sustainable revenue. During the initial surge of generative AI interest, enterprise buyers allocated innovation budgets to test dozens of tools. However, converting pilot enthusiasm into permanent software line items has proven far more difficult than expected. In an analysis of enterprise generative AI deployments, MIT's NANDA initiative found that roughly 95% of enterprise generative AI pilots delivered little to no measurable profit and loss (P&L) impact, with only about 5% achieving rapid revenue acceleration.
For LPs evaluating fund portfolios, this pilot-to-production gap represents a critical valuation risk. When enterprise contracts expire, buyers routinely consolidate software seats, negotiate lower usage fees, or replace proprietary tools with open-source models. If a general partner's portfolio companies are reporting ARR built largely on temporary pilot budgets, subsequent revenue contraction will lead to sharp valuation write-downs.
Evaluating Portfolio Overlap and Secondary Market Exposure
LPs must also test for hidden portfolio overlap across their fund-of-funds, primary fund allocations, and co-investment vehicles. Because top-tier VC funds frequently participate in the same mega-funding rounds for leading AI labs, an LP invested in five different venture funds may unknowingly hold quadruple-concentrated exposure to the exact same frontier entities.
This hidden overlap becomes especially problematic when seeking liquidity. In the secondary market, institutional buyers penalize funds with heavy exposure to illiquid mega-cap private entities that lack immediate IPO timelines. Investors evaluating AI secondary liquidity or reviewing late-stage secondary shares must price in potential buyer discounts when a portfolio's net asset value (NAV) is anchored by a few illiquid mega-rounds.
- Contract Quality Audit: Distinguishing between multi-year committed enterprise contracts and short-term pilot credits.
- Model Switchability Assessment: Determining how easily a portfolio company's enterprise customers can substitute its underlying model for a cheaper alternative.
- Hidden Multi-Fund Overlap: Mapping duplicate exposure across primary, secondary, and co-investment holdings to prevent single-asset saturation.
- Secondary Discount Risk: Stress-testing portfolio liquidity scenarios if mega-cap AI IPOs are delayed or priced at a discount.
What companies, funds, or platforms are expected to show
To satisfy LP diligence standards in a concentrated market, venture capital funds and growth platforms are expected to provide granular disclosures regarding valuation methodologies, liquidity pathways, and competitive defensibility.
Standardizing LP Disclosure Expectations
GPs can no longer rely on aggregated valuation figures that obscure asset-level concentration. LPs are increasingly requiring funds to report top-five asset concentration metrics, detail exact capital invested in frontier labs, and separate core operating revenue from subsidized API usage credits. Furthermore, funds must demonstrate clear pathways to realized cash distributions rather than relying perpetually on paper markups.
Defensibility audits require funds to prove that their portfolio applications possess durable moats against the dominant market leaders. Market power in general-purpose foundation models is concentrated among a handful of players. An analysis highlighted by Brookings notes that the top three AI model providers, Google, OpenAI, and Anthropic, control nearly 90% of the enterprise general-purpose LLM market. When three suppliers dominate model delivery, application startups built on those models must demonstrate deep proprietary differentiation to avoid being displaced by provider platform updates.
Additionally, funds investing in compute platforms must disclose exposure to physical bottleneck constraints, including power access, GPU cluster depreciation, and data center financing terms. LPs evaluating credit and equity exposure should review infrastructure financing risks to verify that asset debt structures are resilient against unexpected drops in compute utilization.
- Top-Asset Exposure Reporting: Explicit disclosure of the percentage of total fund NAV held in the top three AI holdings.
- Underlying Gross Margin Breakdown: Detail on gross margins net of third-party model inference costs and cloud infrastructure spend.
- DPI Realization Timeline: Realistic multi-year forecasts mapping expected cash distributions against historical distribution benchmarks.
- Data Rights and IP Provenance: Proof that portfolio companies hold full legal rights to their training datasets and customer outputs.
Red-flag table and data-room checklist
Institutional LPs must employ systematic red-flag criteria during fund review to isolate operational and structural vulnerabilities early. Identifying warning signs in fund disclosures allows investment committees to adjust commitment sizes or negotiate tailored reporting covenants.
| Red Flag Area | Specific Warning Sign | Underlying Risk | Required LP Action |
|---|---|---|---|
| Model Dependency | Core product functionality relies almost entirely on a single proprietary model API | Extreme vulnerability to API price increases or feature replication | Require API diversification plan and cost-to-serve analysis |
| Concentrated Fund NAV | A single AI lab holding dominates total fund net asset value | High fund vintage volatility tied to one asset's exit outcome | Apply portfolio concentration limits and demand stress-test models |
| Unfunded Compute Commitments | Startup has signed multi-year compute commitments without matching ARR | Balance sheet distress if end-user revenue growth decelerates | Audit cloud vendor contracts and minimum spend liabilities |
| Opaque Valuation Marks | Refusal to disclose valuation mark assumptions or internal discount rates | Distorted TVPI figures that do not reflect secondary market realities | Request independent valuation review and secondary pricing comps |
LP Data-Room Due Diligence Checklist
When reviewing fund data rooms or co-investment documentation, LPs should systematically verify the following core evidence items to ensure complete exposure visibility:
- Fund-Level Concentration Schedule: Complete breakdown of total capital deployed into model, application, and infrastructure layers.
- Top Holdings Overlap Analysis: Cross-fund matrix identifying duplicate company exposure across existing LP commitments.
- Vendor Contract and Credit Terms: Full terms of cloud compute commitments, GPU hosting debt, and vendor credit agreements for major portfolio assets.
- Customer Churn and Usage Metrics: Cohort-level retention figures distinguishing between active software usage and idle seat allocations.
- IP and Data Provenance Audit: Documented verification of proprietary data ownership, training licenses, and data privacy compliance.
How to use this in your next diligence workflow
Integrating AI capital concentration checks into standard LP due diligence workflows enables investment committees to make informed, risk-adjusted allocation decisions. By standardizing questions during annual fund reviews and initial manager selection, institutional investors can effectively hedge against systemic market resets.
Actionable Steps for Investment Committees
First, investment teams should update their standard due diligence questionnaires (DDQs) to mandate explicit disclosures on AI stack distribution, top-holding concentration thresholds, and model dependency risks. Questionnaires should require GPs to report unit economics net of compute costs and provide detailed sensitivity models showing how valuation marks adjust under varying market conditions.
Second, investment committees should establish strict fund concentration limits, capping exposure to any single frontier model lab across primary and co-investment pools. When evaluating portfolio rebalancing, LPs can streamline decision-making by utilizing IC memo automation to synthesize complex manager reports and flag concentration anomalies instantly.
Finally, LPs should incorporate forward-looking reserve checks to verify that GPs have retained sufficient capital to support top application-layer assets through extended growth cycles. Reviewing structured strategies for portfolio reserve models ensures that funds are prepared to maintain equity ownership without over-allocating capital to hyper-expensive model rounds.
- Update LP DDQs: Add mandatory questions covering AI concentration, compute contract liabilities, and underlying gross margins.
- Map Portfolio Overlap: Perform cross-fund analysis to quantify total indirect exposure to major frontier model labs.
- Enforce Asset Cap Rules: Apply explicit percentage caps on single-asset concentration within individual funds and broader LP portfolios.
- Audit Realization Pathways: Stress-test GP distribution projections against secondary market pricing and public market valuation benchmarks.
How Plausity supports the workflow
Evaluating AI capital concentration across complex fund structures, LP reports, and extensive data rooms requires automated analytical tools capable of processing thousands of unstructured financial and legal documents. Plausity provides investment teams with an integrated platform built to streamline fund due diligence and surface hidden concentration risks.
At the core of the platform, the AI-Analysis Engine ingests, parses, and cross-references thousands of fund documents, quarterly reports, and portfolio financial models in minutes. Data Room Ingestion connects directly to virtual data rooms so investment professionals can scan complex LP disclosures, cap tables, and side letters without manual data entry overhead.
To isolate structural risks, the Risk Radar automatically detects concentration vulnerabilities across private equity and venture capital portfolios. It identifies hidden asset overlap across multiple funds, flags excessive reliance on third-party model APIs, and evaluates compute debt commitments against reported revenue. Through automated Findings & Risk Intelligence capabilities, the system highlights materiality scores and alerts investment committees to hidden liabilities before capital is committed.
Once analysis is complete, teams can utilize the Report Builder to automatically generate structured, investor-ready diligence summaries and IC memos with full source traceability back to raw data room documents. Furthermore, the Collaboration Hub provides a unified, secure workspace where deal teams and LP investment committees can coordinate workstreams, share real-time findings, and finalize portfolio risk assessments efficiently.
