Large Model Valuations Are Shifting from Storytelling to Financial Statements
This article explains why investors are increasingly valuing large language model companies on their financial data rather than on technical narratives. It covers five core metrics for assessing AI companies, three principles for reading financial reports, how the industry has moved from narrative to financials, practical takeaways for Chinese model vendors, and a checklist for verifying valuations.
1. Why Valuation Logic Is Changing
Valuations of large model companies have long rested on storytelling. Investors focused on how capable a model was, how large the opportunity for general-purpose agents might be, and whether leaderboard rankings could be turned into funding leverage. This framework worked in the early stage of the industry. Companies had not yet built meaningful revenue, so markets could only judge them on technical prospects.
As leading companies began to face price discovery and scrutiny from public markets, financial transparency became far more important. Once a company goes public, investors need to know where its revenue comes from, whether growth is sustainable, and when losses will narrow. Financial information has become a precondition for pricing.
The shift affects Chinese model vendors more directly. In the past, vendors often used release frequency and leaderboard results to prove their strength, but this approach is losing effect. A slowdown in release cadence may suggest that vendors have realized model launches alone no longer translate directly into valuation.
2. Five Core Metrics for Assessing an AI Company’s Value
When applying traditional software valuation methods to large model companies, the following five metrics are the most common starting points.
| Metric | What to Look At | Common Risks |
|---|---|---|
| Revenue growth and quality | Sequential and year-over-year growth; whether revenue comes from customers who keep paying | One-time contracts or subsidized revenue inflating growth |
| Billing structure | Share of usage-based billing, subscriptions, and enterprise licenses | Usage grows but unit prices fall, making revenue hard to sustain |
| Gross margin | Inference cost as a share of revenue; whether it declines with scale | Compute costs grow in line with revenue, leaving gross margin unimproved over the long term |
| Compute commitments | Total value and maturity schedule of future purchase, lease, and cloud service contracts | Commitment terms longer than the revenue growth cycle, creating rigid expenses |
| Operating cash flow and profit basis | Differences between accounting-standard and adjusted profit figures, and their causes | Focusing only on adjusted profit and ignoring the effects of equity incentives and depreciation |
These five metrics are interrelated. If revenue grows quickly but gross margin does not improve, scale has not diluted compute costs. If compute commitments are high while revenue growth slows, cash pressure will gradually emerge.
3. Three Principles for Reading Financial Data
First, distinguish adjusted metrics from accounting-standard metrics. Adjusted profit usually excludes equity incentives, some depreciation, and one-time items. It can show the profitability trend of the core business, but it cannot replace figures on an accounting-standard basis. The larger the gap between the two, the more important it is to ask where the gap comes from.
Second, distinguish commitments by type. Terms such as “contractual obligations and other commitments” often combine several kinds of arrangements, including purchase commitments, operating leases, and cloud service agreements. Each has different performance conditions, consequences of breach, and timing. Treating the total commitment as cash that is already locked in can overstate or misjudge the risk.
Third, watch the direction of change rather than a single number. If the loss rate falls while the absolute loss widens, revenue is growing faster than costs, but the company still needs outside funding to support expansion. Changes in loss amount and loss rate carry different investment implications, so both should be compared across the same time series.
4. From Narrative to Financials: What Changed in the Industry
Over the past year or so, the industry’s progression shows a clear path. Early in the year, a breakout agent product led the market to imagine the commercial prospects of general-purpose agents. Coding then became the focus of competition in model capabilities. In summer, a dense wave of model releases followed, and each launch and leaderboard ranking could directly affect funding negotiations.
Narratives have their own life cycle. When most vendors tell similar stories and show similar leaderboard results, the narrative itself loses its ability to differentiate. Investors turn to hard indicators that set companies apart, such as revenue mix, customer retention, and cost curves.
Pricing in public markets reflects the same trend. The same company can be valued differently in private and public markets. Public market investors are more willing to pay for verifiable revenue, while premiums for future potential tend to shrink.
5. Three Takeaways for Chinese Model Vendors
First, commercial capabilities must be built in advance. The credibility of a revenue growth curve directly affects a company’s ability to raise funds. Enterprise customer expansion, product designs with usage-based billing, and renewal and upsell systems all need to be in place before the financing window tightens.
Second, compute investment should match the pace of revenue. Large compute commitments are an important risk signal. Once infrastructure spending exceeds what revenue growth can support, cash flow pressure can build quickly. Vendors with limited capital reserves can prioritize using mature cloud services on demand to obtain compute, which reduces the rigid burden of heavy self-built infrastructure. When choosing a cloud service, compare billing methods, elasticity, and breach terms rather than looking only at unit prices.
Third, differentiation should rest on quantifiable scenario value. The marginal value of leaderboard scores is declining. Customers usually pay based on the cost reduction and efficiency gains they see in specific business scenarios. Vendors need to translate model metrics into business metrics and prove results with auditable data.
6. Reader Verification Checklist
To check whether a model company’s valuation is supported by data, review the following questions in order:
- Is revenue growth over the last four to eight quarters stable, and can changes in growth be explained?
- What share of revenue comes from usage-based billing, subscriptions, and enterprise contracts, and how concentrated is the customer base?
- Does gross margin improve as scale grows, and can declines in inference cost be broken down by specific component?
- Are contract liabilities growing, and can advance payments support future revenue recognition?
- What is the ratio of total compute commitments to annual revenue, and how are the maturity dates distributed?
- What causes the gap between adjusted profit and accounting-standard profit: equity incentives, depreciation, or other items?
- Is operating cash flow positive, and how does the financing scale compare with the pace of cash burn?
Answers to these questions should come from company announcements, prospectuses or offering documents, regulatory disclosures, or authoritative media reports. Data from unknown sources should be treated as unverified until confirmed.
7. Risk Disclosure
This article provides an analytical framework only and does not constitute investment advice. Financial reporting bases, accounting policies, and disclosure timing vary across companies, so the same metric may not be directly comparable across reports. Financing conditions and market sentiment can change quickly, so valuation judgments should be updated as new information emerges. Before citing any specific figures, refer to the original announcements and official disclosure documents.
