How to Value Large Model (LLM) Companies: The Shift in Valuation Logic and an Evaluation Pitfall Guide
Large model (LLM) companies are seeing a shift in pricing power: in private markets, valuation methods that rely on narrative, sector hype, and funding cadence are giving way to public-market scrutiny of revenue quality, cost structure, cash flow, and technical barriers. This article explains why traditional software valuation frameworks fall short for LLM companies, then walks through three consecutive stages of valuation migration — the supply-scarcity premium, the ARR growth narrative, and global comparison with profitability quality. It also provides a practical evaluation checklist for developers, investors, and enterprise decision-makers, an illustrative Python function for tracking key valuation inputs, and a rundown of common mistakes to avoid. The content is an analytical framework and method examples only, and does not constitute investment advice.
Large model companies are undergoing a shift in pricing power: in private (primary) markets, valuation methods that rely on narrative, sector hype, and funding cadence are giving way to public-market scrutiny of revenue quality, cost structure, cash flow, and technical barriers. For developers, investors, and enterprise decision-makers, understanding this shift helps avoid mistaking short-term capital behavior for long-term value, and helps build more stable criteria for financing, partnerships, procurement, and technology roadmaps.
1. Why Large Model Companies Need a New Valuation Framework
Traditional software companies are typically valued around revenue growth, profit margins, customer retention, and cash flow. Their product forms are relatively mature, revenue recognition is fairly clear, and the number of variables is limited. Large model companies are different: R&D investment is front-loaded and keeps rising; revenue structures are more complex and can combine subscription, usage-based billing, and custom project delivery; and technical iteration happens on a monthly or even weekly basis. Inference cost, compute, and engineering costs all directly affect gross margin. Applying the traditional tech-stock template directly can therefore misjudge their real value.
From private to public markets, the pricing logic is also switching. In the private stage, information transparency is lower, and valuations often rely on consensus among a few investment institutions, sector narrative, and future expectations. Public markets place more emphasis on liquidity, comparable companies, real performance, and continuous validation. Every expectation must repeatedly withstand scrutiny from financial reports, user growth, gross-margin changes, and technical progress. This switch has given rise to three consecutive migrations of valuation logic.
You can summarize the three stages with a table first:
| Migration stage | Core metrics | Main risks | Applicable scenarios |
|---|---|---|---|
| Stage 1: Supply-scarcity premium | Free float ratio, index inclusion, passive capital | Lock-up expiry, share supply expansion, more comparable targets | Scarce targets, strong market sentiment |
| Stage 2: ARR growth narrative | ARR, NRR, subscription share, collections | Project-based noise, low-quality growth, declining gross margin | Verifiable revenue, clear subscription model |
| Stage 3: Global comparison and profitability quality | Gross margin, inference cost, open-source disruption, unit economics | Price wars, technology homogenization, cost overruns | Well-developed competition, capital focused on profitability |
2. How to Read Stage 1: Supply-Scarcity Premium
In the early days after listing, public markets typically have few targets whose core business is a general-purpose foundation model. If capital wants to allocate to this direction, the limited options mean scarcity itself brings a valuation premium. This premium does not come entirely from fundamentals; it can also come from insufficient supply, concentrated holdings, and capital allocation demand.
If some companies have a low free float ratio, a small amount of tradable shares facing relatively high subscription demand makes prices very sensitive to marginal buying and selling. In this situation, first-day performance, short-term volatility, and turnover may be amplified. What investors should note is that such price changes mostly reflect the share structure and capital behavior, and do not mean the company’s fundamentals have already pulled ahead by an equivalent margin.
Passive capital from index inclusion and cross-market connectivity mechanisms can also push valuations higher. When a target is added to a mainstream tech or broad-based index, passive funds that track the index may buy it mechanically, reinforcing short-term price performance. But this capital belongs to trading structure and allocation demand; it cannot directly prove a company’s long-term profitability.
What truly tests how much scarcity is embedded is lock-up expiry and free-float expansion. Lock-up expiry means more tradable shares. If the earlier high valuation relied mainly on scarcity, then once the float expands, the valuation may face recalibration. One illustrative scenario: a company has a low free float ratio and the market assigns it a high valuation; as lock-up expiry approaches, even if some long-term shareholders signal that they will hold, investors may still worry about future selling pressure and liquidity changes, affecting the short-term price. This kind of scenario does not represent any specific company and does not constitute a judgment on any individual stock.
So, when facing the supply-scarcity premium, investors should focus on a few questions: How large is the target’s free float ratio? Is lock-up expiry approaching? How much of the non-fundamental gain came from index inclusion and passive capital? If the float expands and comparable targets increase, can the original scarcity premium still hold?
3. How to Read Stage 2: The ARR Growth Narrative
As the scarcity premium fades, the market needs a new valuation anchor. Annual Recurring Revenue (ARR) has become an important metric for valuing large model companies. ARR emphasizes the sustainability, predictability, and subscription nature of revenue, and can relatively clearly reflect the speed of a company’s growth.
But ARR must use a rigorous definition. Generally, ARR should be annualized and should exclude non-recurring project revenue, one-time delivery revenue, and revenue that does not have a continuous subscription nature. If a company counts short-term projects, custom development, or one-time revenue as ARR, it will overstate the true quality of growth. A higher subscription share usually means more stable revenue and is more easily accepted by public markets.
Around ARR, the market typically focuses on several dimensions:
First, the absolute size of ARR and its year-over-year growth. Size determines the market space, and growth rate determines growth elasticity, but high growth must be looked at together with renewals and collections.
Second, net revenue retention (NRR). NRR should be calculated on the same cohort of existing customers, reflecting whether existing customers keep expanding, upgrading, or adding purchases. If NRR is persistently below 100%, it means existing customers may be churning or spending less, and growth has to rely on continuously adding new customers, with higher cost pressure.
Third, revenue structure. The share of subscription revenue, usage-based revenue, and project-based revenue determines the predictability of revenue. Project-based revenue may bring short-term scale, but if it relies on manual delivery, gross margin and scalability are limited.
Fourth, the collection cycle and bad-debt risk. If revenue growth is not accompanied by cash inflow, its quality should be discounted. A common problem for large model companies is that contract amounts are large, but delivery cycles are long and collections are slow, ultimately affecting cash flow.
The ARR narrative also has a ceiling. It can only prove that a company grows fast; it cannot automatically prove that it grows steadily or makes money. Project-based delivery, deep customization, and trading price for scale can all push up ARR while crushing gross margin. When every player can report similar growth rates, ARR itself becomes homogeneous and loses its differentiating power.
So how do you judge the quality of a large model company’s ARR? Look at three signals: first, whether renewals and upsells are real; second, whether subscription revenue can be annualized; third, whether growth comes at the cost of low gross margin. Common mistakes include: looking only at ARR growth while ignoring gross margin; counting one-time projects as ARR; ignoring lock-up expiry and the float; and mistaking low-price customer acquisition for healthy growth.
4. How to Read Stage 3: Global Comparison and Profitability Quality
The third migration of valuation logic is to re-price within the coordinates of global competition and profitability quality. Large models are a global competition: a company must compare not only with peers in its own market, but also with global open-source models, closed-source models, and rapidly iterating technology routes.
When open-source models approach the capability of closed-source models on specific tasks, and offer lower inference cost and more flexible deployment, the “technology-leadership premium” of closed-source models is compressed. This should not point to any specific model or company, but it can be observed: once the open-source ecosystem matures quickly, the pricing power, user stickiness, and gross-margin room of closed-source vendors may all come under pressure. This judgment is only an industry-analysis example and does not constitute investment advice.
Profitability quality is a hard constraint on the valuation center. Capital will ultimately ask: how much revenue can each unit of compute investment bring back? Is the ratio of inference cost to revenue reasonable? Can gross margin improve as scale grows? If high revenue is accompanied by low gross margin, high customer-acquisition cost, and high inference cost, the valuation lacks support.
To turn these qualitative judgments into computable metrics, here is an illustrative code snippet for systematically tracking a large model company’s key valuation inputs:
def evaluate_foundation_model_company(arr_t0: float,arr_t1: float,gross_profit: float,revenue: float,market_cap: float,inference_cost: float,):"""Inputs:- arr_t0 / arr_t1: opening and closing annual recurring revenue (annualized);- revenue: annual or trailing-12-month revenue in the same currency;- market_cap: current market cap, must use the same currency as revenue;- gross_profit / inference_cost: same currency and same period basis."""if arr_t0 <= 0 or arr_t1 <= 0:raise ValueError("arr_t0 and arr_t1 must be greater than 0")if revenue <= 0:raise ValueError("revenue must be greater than 0")if market_cap < 0 or inference_cost < 0:raise ValueError("market_cap and inference_cost cannot be negative")arr_growth = (arr_t1 - arr_t0) / arr_t0gross_margin = gross_profit / revenueps_multiple = market_cap / revenuearr_multiple = market_cap / arr_t1inference_ratio = inference_cost / revenuereturn {"arr_growth": round(arr_growth, 4),"gross_margin": round(gross_margin, 4),"ps_multiple": round(ps_multiple, 2),"arr_multiple": round(arr_multiple, 2),"inference_ratio": round(inference_ratio, 4),}
This code itself does not predict stock prices; instead, it compresses four kinds of information — growth speed, profitability quality, valuation level, and cost pressure — into comparable metrics. Only by looking at ARR growth, gross margin, and the share of inference cost together is it less easy to be led astray by a single narrative.
For example: suppose a company has annual revenue of 1 billion RMB, opening ARR of 600 million, closing ARR of 800 million, gross profit of 240 million, current market cap of 6 billion, and inference plus compute cost of 120 million. Then the ARR growth rate is about 33.33%, the gross margin about 24%, the price-to-sales ratio about 6x, the ARR multiple about 7.5x, and the inference-cost-to-revenue ratio about 12%. If revenue is 0, or arr_t0 or arr_t1 is 0, the code raises a ValueError to avoid division by zero or invalid computation. If the currencies are inconsistent, you should convert everything to a common currency before calculating.
5. An Evaluation Checklist for Developers and Enterprise Decision-Makers
When facing a large model company, work through the following questions step by step:
First, ask about scarcity. How large is the target’s free float ratio? Is lock-up expiry approaching? How much gain came from index inclusion and passive capital? If the price is driven mainly by the share structure, you should reduce linear extrapolation about long-term value.
Second, ask about growth quality. Is ARR annualized and does it exclude non-recurring project revenue? Is the subscription share high enough? Is NRR calculated on the same cohort of customers and kept healthy? Does growth come from low prices, project-based work, or one-time delivery?
Third, ask about profitability and competition. What is the gross-margin trend? Can inference cost fall as scale grows? Facing the cost shock from open-source models, how much pricing power does the company still have? If technology leadership cannot translate into a cost advantage or pricing power, the long-term valuation is hard to keep solid.
Common mistakes include: focusing only on ARR growth while ignoring gross margin; counting one-time projects as ARR; ignoring lock-up expiry and float changes; mistaking a rise driven by passive capital for fundamental improvement; and linearly extrapolating a closed-source model’s high premium under the shock of open-source models.
6. Conclusion and Risk Warning
For large model companies, the pricing logic — from the supply-scarcity premium, to the ARR growth narrative, and then to global comparison and profitability quality — is essentially a process of valuation shifting from expectation-driven back to value-driven. Scarcity can only explain the moment; ARR can prove growth speed; but the long-term valuation center will ultimately be jointly determined by technical strength, revenue quality, gross margin, the path to profitability, and the global competitive landscape.
Understanding this migration path not only helps judge the long-term value of large model companies, but also helps enterprise managers understand how capital markets evaluate technology routes, revenue models, and cost efficiency. It should be specially noted: this article provides only an analytical framework and method examples, and does not constitute any investment advice. The market phenomena and calculation examples mentioned here are used only to illustrate the logic; specific company situations should be based on publicly disclosed information and verifiable data.
