Wall Street analysts and fund managers are quietly overhauling the frameworks they use to value artificial intelligence companies, acknowledging that traditional metrics developed for software businesses do not adequately capture what AI companies are building or how they make money. The shift has significant implications for how AI stocks are priced, which companies get rewarded by public markets, and what the next generation of AI businesses will look like as they are built with an eye toward how investors will eventually evaluate them.

Why Old Valuation Models Do Not Work for AI

The software-as-a-service (SaaS) playbook that Wall Street used to value tech companies for the past two decades was built around recurring subscription revenue, high gross margins, and predictable customer retention. Price-to-revenue multiples, rule of 40 calculations, and net revenue retention metrics made sense for companies selling software licenses to enterprises.

AI companies break most of these assumptions. The largest AI labs spend billions of dollars on compute to train models that are then made available for pennies per API call. Gross margins are often dramatically lower than traditional software because compute costs are real and significant. Revenue models are still evolving – some companies charge subscriptions, some charge per token, some license model weights, and some are still trying to figure out monetization entirely. Traditional valuation frameworks were not built for this business model.

• Analysts are increasingly using ‘compute efficiency’ metrics – revenue or capability generated per dollar of training compute – as a differentiator between AI companies.
• The distinction between AI ‘infrastructure’ companies (selling compute and model access) and AI ‘application’ companies (selling specific solutions) is becoming a critical valuation frame.
• Customer concentration risk is being scrutinized heavily after several AI companies revealed that a small number of large customers account for the majority of revenue.
• The pace of model capability improvement creates obsolescence risk that traditional software did not face – last year’s frontier model is this year’s commodity.

The New Metrics Emerging on Wall Street

Several new metrics are gaining traction among analysts covering AI companies. ‘Token economics’ – the relationship between the cost of generating a token and the price charged for it – is becoming a standard part of AI earnings analysis. ‘Model depreciation’ is being discussed as a real line item, since AI models lose competitive value faster than software products did historically. ‘Agent utilization’ is emerging as a key metric for companies selling AI agents, measuring how much of the agent’s theoretical capacity is being productively used by customers.

What This Means for AI Company Strategies

The shift in how Wall Street values AI companies is already influencing how those companies talk about their businesses. CEOs are emphasizing metrics they know analysts care about, and companies are making product and pricing decisions with an eye toward improving those numbers. This creates both healthy discipline – forcing AI companies to think more rigorously about unit economics – and potential distortions, as short-term metric optimization can conflict with the long-term investments needed to stay at the frontier.

Frequently Asked Questions

Are AI stocks overvalued?

Opinions differ sharply. Bulls argue that AI’s significant potential justifies premium valuations. Bears argue that current prices discount an implausibly optimistic scenario. The honest answer is that valuing AI companies involves more genuine uncertainty than most asset classes, making confident declarations of over or undervaluation unreliable.