Valtech Published Directions on Valuation of AI-Based Startups, Ventures, and Companies

Valuing an AI company fundamentally involves estimating future cash flows, assessing the risks surrounding those cash flows, and evaluating the durability of its competitive advantage. What changes in the AI sector is not the valuation identity itself, but rather the underlying economic structure. Cash flows are often heavily back-loaded, research and development expenditure is unusually front-loaded, and compute infrastructure and proprietary data may behave similarly to quasi-capital expenditures. In addition, failure probabilities remain materially higher than those of conventional software companies until product-market fit, model reliability, governance standards, and commercial scalability are demonstrated.

Accordingly, the valuation of AI-driven companies remains anchored in intrinsic value principles, but requires greater reliance on scenario analysis, milestone-based assessment, and explicit adjustments for survival probability, dilution risk, and reinvestment intensity.

For many AI firms, particularly application-layer companies, benchmark performance alone is not the primary driver of enterprise value. The emergence of open-weight models has narrowed the performance gap between open and closed models across certain benchmarks, while model capabilities continue to diffuse rapidly and unit economics evolve quickly. As a result, economic value increasingly derives from proprietary workflow data, customer distribution channels, operational integration, switching costs, trust, regulatory compliance, and ecosystem positioning, rather than solely from ownership of a frontier model.

Data itself can be highly valuable; however, its value is context-specific, subject to diminishing marginal returns, and may decay over time. In many cases, recent, legally usable, task-relevant, and high-quality data is significantly more valuable than sheer data volume.

We believe that the discounted cash flow (DCF) method remains the most reliable valuation anchor once a company has established a coherent monetization strategy and forecastable unit economics. Market-based approaches using global peer benchmarking are also useful in forming an indicative valuation range. In contrast, real options analysis and risk-adjusted net present value (NPV) methodologies become increasingly relevant when a substantial portion of value depends on stage-gated technical milestones or uncertain commercialization outcomes. Venture capital and scorecard approaches, meanwhile, are generally more suitable for the earliest-stage companies where financial visibility remains limited.

In practice, robust AI valuation analysis often incorporates a combination of:

• an operating DCF framework;
• market approach benchmarking against comparable companies and transactions; and
• milestone-based or scenario-driven analysis to evaluate strategic optionality and execution uncertainty from multiple perspectives.

A practical framework for AI valuation is to distinguish between capability assets and cash flow translation.

Capability assets may include proprietary data, model architecture, technical talent, distribution networks, intellectual property, and regulatory positioning. However, these assets create enterprise value only when they enhance one or more of the following economic drivers:

• customer acquisition efficiency;
• customer retention;
• pricing power;
• gross margin expansion;
• capital efficiency; or
• the expected duration of competitive advantage.

Professor Aswath Damodaran’s framework for valuing young companies is particularly relevant in this context. For many AI companies, the majority of enterprise value lies in future growth assets. However, growth only creates value when the company can consistently generate returns on reinvestment above its cost of capital.

Data can constitute a meaningful competitive moat, but only under specific conditions. The strongest data assets are not merely large in scale; rather, they are exclusive or difficult to replicate, legally usable, high quality, current, well-labelled, and closely linked to the prediction tasks or operational workflows for which customers are willing to pay.

Similarly, a company that genuinely possesses a superior model stack may deserve a valuation premium. However, such a premium should not be assumed to persist indefinitely, as competitors may close the gap through more efficient training methods, superior fine-tuning techniques, or combinations of open-weight models and proprietary datasets. Consequently, startups should be prepared to demonstrate how they intend to sustain and continuously improve their technological advantage over time.

TAM, SAM, and SOM analysis remains highly relevant in assessing the commercial potential of startups and continues to be applicable in the valuation of AI-related businesses.

Total Addressable Market (TAM)

The total global or industry-wide revenue opportunity for a product or service, without considering geographic, operational, or competitive constraints.

Serviceable Addressable Market (SAM)

The portion of the TAM that aligns with the company’s target market, business model, geographic reach, and operational capabilities.

Serviceable Obtainable Market (SOM)

The realistic share of the SAM that the company can capture within the near to medium term, typically over a one- to three-year horizon, taking into account practical constraints such as competition, marketing resources, staffing, and execution capability.