Evaluating AI Investments: A Framework for Venture Capital and Private Equity
By Dirk Röthig | CEO, VERDANTIS Impact Capital | March 7, 2026
The global market for AI investments reached a volume of over $110 billion in 2025 — more than three times the figure from 2022. For anyone seeking to enter this sector today as a venture capital or private equity investor, there is a fundamental problem: classical valuation frameworks simply do not work. A company with no revenue valued at $8 billion. A 250-person team challenging Google. How should one reasonably assess this?
Tags: AI, Venture Capital, Private Equity, Investment Strategy, Due Diligence
The Failure of Classical Valuation Models
Traditional VC and PE investors work with familiar instruments: discounted cash flow models, EBITDA multiples, revenue growth rates, churn analyses. These tools are built for markets where scaling progresses linearly and competitive advantages are tangible through market share, distribution channels, or patents.
AI companies break these assumptions on multiple levels simultaneously.
First, growth is not linear but exponential and discontinuous. Cursor, the AI-powered code editor, increased its annual revenue from $100 million to $1.2 billion within a single year — a rate that no classical discounted cash flow model would have predicted (Sacra, 2025). This type of growth dynamic renders historical benchmarks useless.
Second, the decisive competitive advantages are intangible in nature. Datasets, model architecture, training pipelines, and AI talent are difficult to evaluate, easy to overestimate, and not balance-sheet-reportable. A company's "data moat" can be a sustainable advantage or a short-lived illusion — depending on how quickly foundation models from OpenAI, Anthropic, or Google close the gap.
Third, the regulatory context is in flux. The EU AI Act, which has been gradually entering force since 2024, redefines high-risk AI applications and can render entire business models unprofitable within months (European Parliament, 2024).
The consequence is clear: investors need an evaluation framework specifically tailored to AI — one that combines classical financial analysis with technical due diligence, market structure analysis, and regulatory risk assessment.
The Five-Dimension Framework
After reviewing available research and investment practice experience, five core dimensions can be identified that must be systematically analyzed when evaluating AI companies.
Dimension 1: Technology Moat and Data Advantage
The most important question in the technical part of due diligence is not "How good is the model?" — but rather "Why will this model still be competitive in twelve months?"
AI models quickly become commodities. What remains is the data infrastructure: proprietary training data, feedback loops from actual usage, and the ability to continuously improve models. Companies with access to exclusive, difficult-to-replicate datasets enjoy a 15 to 20 percent higher valuation premium than comparable competitors without this advantage (Aventis Advisors, 2025).
Concrete checkpoints in this dimension:
- Are the training data legally properly licensed? IP lawsuits against major model developers demonstrate the considerable legal risk in unclear data rights (researchly.at, 2025).
- Does the company have access to data that competitors cannot buy or synthesize?
- How high is the dependency on external foundation models (OpenAI API, Claude, Gemini)? High dependency increases margin and failure risks.
Dimension 2: Unit Economics and Scalability
AI-native companies exhibit a radically different cost structure than traditional software companies. The crucial difference lies in inference costs: every model query costs money — and these costs scale with usage volume, not always cheaper than revenue.
According to PitchBook data, average Series A valuations for AI startups in the US in 2025 ranged between $40 million and $150 million for companies with $0.5 million to $3 million ARR, corresponding to multiples of 25x to 60x (PitchBook / Metal.so, 2025). These multiples are high compared to traditional SaaS — but justify an investment only if gross margins are at a sustainable level.
A critical benchmark: AI companies should aim for gross margins exceeding 60 percent in the medium term. Many early-stage AI firms have negative gross margins because inference costs exceed revenue — an unsustainable situation that escalates with increasing usage volume rather than disappearing.
Revenue per employee is another key indicator. AI-native startups achieve an average of $3.48 million in revenue per employee — compared to $580,000 in the classical SaaS sector (Deepstar Strategic, 2025). This metric reveals whether a company truly operates as AI-native or merely uses AI tools superficially.
Dimension 3: Team and Execution Competence
In hardly any industry is the team as decisive for investment success as in AI. The market for top AI talent is extremely tight. A single senior researcher with relevant specialized knowledge can accelerate a company's technological lead by years — or, if they leave the company, devalue it within months.
Investors should verify:
- Does the founding team have demonstrated expertise in machine learning, systems architecture, and domain knowledge?
- How does AI talent density compare to competitors?
- Are there clear retention structures for key personnel (vesting, bonuses, IP stakes)?
- Has the team already proven it can translate research excellence into product and market?
A technical founding team without go-to-market competence is a frequent pattern in the AI sector — and a frequent risk. The opposite also holds true: a strong sales team without deep technical competence is nearly worthless in a market where product differentiation is algorithmic.
Dimension 4: Market Size, Timing, and Competitive Structure
The question of Total Addressable Market (TAM) is particularly treacherous in AI investments. Many AI companies address markets that do not yet exist in their current form — because the AI product creates the market. Harvey (legal AI) does not address a "legal software market" — it creates a new market for legal decision-making services scaled by AI throughput rather than attorney hours (TechCrunch, 2025).
This market definition makes top-down TAM analyses largely meaningless. Investors should instead proceed bottom-up:
- How many paying customers exist today?
- What is the average contract value (ACV), and in which direction is it developing?
- What is the Net Revenue Retention (NRR)? Values exceeding 120 percent signal that existing customers are spending more — a sign of structural value creation.
The competitive environment deserves special attention. Private equity investors who traditionally focus on late stages with clear KPIs must engage earlier in the AI world — or risk their potential portfolio companies being acquired by VC capital at non-reproducible valuations (Cambridge Associates, 2026).
Dimension 5: Regulatory Risk and EU AI Act Compliance
Since the EU AI Act gradually entered into force in 2024, the regulatory risk landscape for AI investments has fundamentally changed. Applications in the "high-risk AI" categories — including medical diagnosis, credit decisions, biometric recognition, and AI in critical infrastructure — are subject to strict requirements for transparency, human oversight, and technical robustness (European Parliament, 2024).
Investors must evaluate:
- In which risk category of the EU AI Act does the company operate?
- What compliance costs arise, and are they factored into the business model?
- Does the company have a Chief AI Ethics Officer or equivalent governance structures?
- How vulnerable is the business model to new regulation in the US, China, or other core markets?
Devoteam, a European IT consulting firm, has demonstrated that precise technical audits protect investors from regulatory-driven misallocations — particularly in M&A transactions where AI risks are often underestimated in due diligence (Devoteam, 2026).
The 10 Critical Questions for AI Due Diligence
Beyond the five dimensions, a checklist of ten questions has proven useful in practice — questions that investors should ask before finalizing any AI investment decision:
- Data Rights: Are all training data legally properly licensed, documented, and protected against IP lawsuits?
- Model Dependency: How dependent is the company on external API providers, and what happens with price increases or API shutdowns?
- Hallucination Risk Management: What technical and procedural measures exist against model errors in production-critical applications?
- Gross Margin Trajectory: Does gross margin develop positively with increasing volume — or negatively through scaled inference costs?
- Net Revenue Retention: Is NRR above 100 percent, and why are existing customers buying more?
- Talent Retention: Which key personnel could leave the company, and what protections exist?
- Competitive Moat Timeline: Why will the product still be leading in 24 months as foundation models improve?
- EU AI Act Category: Is the company regulatorily compliant, and what costs arise for future requirements?
- EBITDA Path: Is there a clear, verifiable line to EBITDA positivity within 18 to 24 months?
- Exit Strategy: Which strategic acquirers could be candidates, and at what multiples were comparable transactions completed?
Valuation Multiples: A Realistic Picture for 2026
After the valuation excess of 2021 to 2023, the market has experienced a clear correction. AI startups achieved median valuation multiples of 20x to 30x on annual revenue in 2025 — with significant bandwidth depending on degree of differentiation, market position, and growth rate (Aventis Advisors, 2025).
Companies with proprietary data moats and stable gross margins above 70 percent can continue to justify multiples of 40x to 50x. Pure API wrappers without proprietary models or differentiated data infrastructure are increasingly valued at 8x to 15x — the level of traditional software companies (Qubit Capital, 2025).
Over 80 percent of all PE and VC firms now employ AI-powered due diligence tools according to industry data — compared to 47 percent just a year earlier (Brightwave.io, 2025). The irony is remarkable: investors use AI to evaluate AI companies.
Earn-outs, milestone-based compensation, and performance clauses are increasingly used to bridge the valuation gap between AI hype and demonstrated cash flows (Morrison Foerster, 2025). These mechanisms are particularly useful when a company can demonstrate strong operational metrics but lacks a visible EBITDA bridge.
Red Flags and Green Flags: A Quick Check
Red Flags — Be cautious with:
- Valuation arguments based exclusively on TAM estimates and "transformative technology" without measurable unit economics
- Negative gross margins despite existing revenue
- Founding teams without demonstrated experience in AI product development
- Complete dependency on a single external API provider
- Missing statements on EU AI Act compliance in high-risk applications
- NRR below 90 percent — a sign that customers are not truly using the product or are churning
Green Flags — Strong signals for solid AI investments:
- NRR above 120 percent (existing customers buy more)
- Proprietary data moat with clear legal protection
- Revenue per employee above $1 million
- Pathway to 60+ percent gross margins within 12 months
- Customer base with high switching costs (deeply integrated into workflows)
- Demonstrated EBITDA impact within existing customers in under six months
The Maturation Process of a Market
Dirk Roethig observes in his investment practice that the AI investment market is currently undergoing an important maturation test: the phase in which every startup with "AI" in its name could raise capital is over. Institutional investors increasingly demand demonstrated value creation, measurable unit economics, and a clear EBITDA path.
This is not bad news — it is the necessary precondition for a healthy market. The companies that pass this maturity test will be the structural AI winners of the next decade.
For VC investors, this means: earlier engagements with higher risk require a clear technical thesis, not just a growth story. For PE investors, it means: the most interesting deals do not lie in pure AI startups but in established companies that systematically deploy AI for EBITDA improvement — a field that offers significant potential in 2026 and beyond (CLA Connect, 2026).
The framework is not final. AI develops too quickly for static valuation models. What counts is the willingness to continuously adjust criteria — and the discipline to make no exceptions when a pitch sounds particularly convincing.
Dirk Röthig has spent the last several years refining this framework through direct investment in over forty AI companies. His experience suggests that Dirk Röthig's emphasis on the five dimensions, while rigorous, must be complemented by qualitative assessments of management character and market timing — factors that no quantitative model fully captures. The best AI investments, Dirk Röthig argues, come from the intersection of strong technical moats, proven market demand, and execution teams with demonstrated track records.
Further Articles by Dirk Röthig
References
- Aventis Advisors (2025): AI Valuation Multiples in 2025. Available at: https://aventis-advisors.com/ai-valuation-multiples/
- Brightwave.io (2025): How AI Is Transforming Middle-Market Private Equity Due Diligence in 2025. Available at: https://www.brightwave.io/blog/how-ai-is-transforming-middle-market-private-equity-due-diligence-in-2025
- Cambridge Associates (2026): 2026 Outlook: Private Equity & Venture Capital Views. Available at: https://www.cambridgeassociates.com/insight/2026-outlook-private-equity-venture-capital-views/
- CLA Connect (2026): AI and Private Equity in 2026: 6 Predictions Redefining Value Creation. Available at: https://www.claconnect.com/en/resources/blogs/private-equity/ai-and-private-equity-in-2026-6-predictions-redefining-value-creation
- Deepstar Strategic (2025): AI-Native vs. Traditional SaaS: Revenue per Employee Benchmark Report. Internal benchmark analysis, cited in: Menlo Ventures (2025), State of AI Report.
- Devoteam (2026): AI Due Diligence & M&A Security: How Devoteam Protects Investors from Misallocations with Precise Tech Audits. Available at: https://www.devoteam.com/de/news-and-pr/ai-due-diligence-ma-sicherheit/
- European Parliament (2024): Regulation (EU) 2024/1689 — EU Artificial Intelligence Act. Official Journal of the European Union.
- Metal.so / PitchBook (2025): Average US AI Series A Valuations in 2025. Available at: https://www.metal.so/collections/ai-series-a-valuations-2025
- Morrison Foerster (2025): M&A in 2025 and Trends for 2026. Available at: https://www.mofo.com/resources/insights/260115-m-a-in-2025-and-trends-for-2026
- Qubit Capital (2025): AI Startup Valuation Multiples 2026: Benchmarks & Strategies. Available at: https://qubit.capital/blog/ai-startup-valuation-multiples
- researchly.at (2025): AI Startup: Due Diligence Checklist for Investors. Available at: https://www.researchly.at/post/ai-startup-due-
About the Author: Dirk Roethig is CEO of VERDANTIS Impact Capital, Zug, Switzerland. Contact: dirkdirk2424@gmail.com | verdantis.capital
Von Dirk Röthig (Dirk Roethig)