Saturation Room and Firm Value
饱和余量与公司价值:AI 原生公司的估值框架
Working paper (v1) — June 2026 · maymedoc@gmail.com
Abstract
Building on the saturation-decoupling theorem of Paper I, this paper addresses a valuation question: when a firm's value comes from unrealized saturation room—the gap between coverage \(x_t\) and its ceiling \(\bar{x}\)—rather than from current cash flow, how should it be valued? Our central result is a misspecification-cost theorem: any valuation that extrapolates from observed cash flow without explicitly modeling the coverage path \(x_t\to\bar{x}\) underestimates such firms, by an amount that is strictly positive and increases monotonically in the distance to saturation; a correctly specified path-forecasting valuation recovers true value exactly. The underestimation is therefore not a property of discounting but of anchoring. We then define FACE, a valuation re-anchored on net-worth accumulation (\(\mathcal{N}\), assets minus liabilities on a non-monetary balance sheet), and prove a consistency result. Finally, we derive the value–cash-flow lead–lag from an explicit two-rate dynamical system in which execution accumulates exponentially while monetization relaxes toward an Amdahl ceiling; the lead \(\Delta\) is obtained in closed form and is positive exactly when judgment bandwidth binds.
Keywords: firm valuation; intangible assets; endogenous growth; saturation; value–cash-flow lead–lag; AI-native firms. JEL: G31, D21, O33, M21.
AI disclosure. The author used an AI-based assistant (Claude, Anthropic) to support drafting, editing, and the symbolic verification of the closed-form results. All theoretical claims, definitions, proofs, and final text are the author's own and were reviewed and verified by the author, who takes full responsibility for the content.
1.Introduction
A class of firms is becoming common: they accumulate fast yet earn thin current cash flow. They invest in building a capability base (models, data, users, knowledge capital) that can accumulate almost without bound, but for a long time monetize only a small part of it into profit. Standard DCF, as commonly implemented—anchored on current cash flow and extrapolated forward—reaches an uncomfortable conclusion: they "aren't worth much." Yet markets often value them very highly. This gap is not irrationality; it is a mismatch between how the valuation tool is anchored and the firm's structure.
Paper I proved saturation decoupling: when \(K\) is unbounded but value enters through a bounded saturating map \(x=\sigma(K)\to\bar{x}\), firm value is finite for every discount rate. Paper I answered "is value finite." This paper answers the next question: granting finiteness, how should such firms be valued, and precisely where, and why, does the standard method go wrong?
We show the underestimation is the cost of a specific misspecification—failing to model the coverage path \(x_t\to\bar{x}\)—and that this cost is strictly positive, has a closed form, and increases monotonically in distance to saturation. Symmetrically, a valuation that does forecast the coverage path recovers true value \(V\) exactly. FACE is not an alternative to DCF; it is the DCF one obtains by moving the anchor from cash-flow history to the coverage-driving state.
2.The True Source of Value, and the Cost of Mis-Anchoring
Following Paper I: the state \(x_t\) evolves toward a ceiling \(\bar{x}\), value enters instantaneous profit through a bounded nondecreasing map \(\pi(x)\), and firm value is \(V=\int_0^\infty e^{-rt}\pi(x_t)\,dt\). Define the firm's saturation room at time \(t\) as the unrealized coverage gap
\[ S_t \;=\; \bar{x} - x_t \;\ge\; 0 . \]Saturation room is the true source of value: a firm with large \(S_t\) has future profit that will grow as \(x_t\) rises toward \(\bar{x}\); a firm with \(S_t\approx 0\) has no such increment.
2.1 · The misspecification-cost theorem
A valuation is cash-flow-anchored (path-blind) if it extrapolates from current profit without modeling the coverage path; the limiting case is \(V_{\mathrm{DCF}}=\pi(x_0)/r\). A valuation is path-forecasting if it explicitly models \(x_t\to\bar{x}\) and discounts \(\pi(x_t)\) along that path.
Assume \(\pi\) nondecreasing and \(x_t\) nondecreasing with \(x_t\uparrow\bar{x}\). Then:
(i) Underestimation. \(G = V - V_{\mathrm{DCF}} = \int_0^\infty e^{-rt}[\pi(x_t)-\pi(x_0)]\,dt > 0.\)
(ii) Monotonicity in distance to saturation. \(G\) is increasing in \(S_0=\bar{x}-x_0\). For the benchmark \(\pi(x)=x\), \(x_t=\bar{x}-(\bar{x}-x_0)e^{-kt}\),
\[ G = \frac{k\,(\bar{x}-x_0)}{r\,(r+k)} = \frac{k}{r(r+k)}\,S_0, \qquad \frac{\partial G}{\partial S_0}=\frac{k}{r(r+k)}>0 . \](iii) Path-forecasting recovers \(V\). A valuation that forecasts the true coverage path equals \(V\) exactly. The gap \(G\) is thus the cost of the path-blind misspecification, not a property of discounting.
(i) Subtract \(V_{\mathrm{DCF}}\) from \(V\); since \(x_t\ge x_0\) and \(\pi\) nondecreasing, the integrand \(\ge 0\) with strict inequality on a positive-measure set. (ii) Larger \(S_0\) raises the integrand pointwise; the closed form follows by computing both integrals. (iii) \(\int_0^\infty e^{-rt}[\bar{x}-(\bar{x}-x_0)e^{-kt}]dt=(k\bar{x}+rx_0)/[r(k+r)]=V\). ∎
A correctly implemented multi-stage DCF can forecast a path, and to the extent it forecasts \(x_t\to\bar{x}\) it lands on \(V\). The theorem's content is the converse: underestimation observed in practice is attributable to path-blindness, bounded below by the closed form whenever the practitioner anchors on cash-flow history rather than the coverage-driving state.
3.The Accounts Driving Saturation Room: A Non-Monetary Balance Sheet
What drives the path \(x_t\to\bar{x}\)? We place the non-monetary factors on the books. A non-monetary balance sheet: assets \(A\) accelerate coverage growth, liabilities \(C\) slow it, net worth \(\mathcal{N}=A-C\).
\(A_1\) Data / knowledge-base stock · \(C_1\) Technical debt
\(A_2\) User / distribution network · \(C_2\) Trust / compliance debt
\(A_3\) Judgment capital (taste, decision quality) · \(C_3\) Bandwidth debt (core attention consumed)
\(A_4\) Process / tooling capital · \(C_4\) Dependency debt (over-reliance on external capability)
\(A_5\) Meta-capability (finding & fixing bottlenecks) · \(C_5\) Organizational coordination cost
Let coverage growth be driven by net worth: \(\dot{x}\propto g(\mathcal{N})\cdot(\bar{x}-x)\)—higher net worth means faster evolution toward saturation, still clamped by \(\bar{x}\). Net worth is thus a state variable that leads cash flow.
3.1 · Why these items deserve the names "assets/liabilities"
Asset trajectory: traditional assets peak at purchase then depreciate toward \(0\) (\(dA/dt<0\), driven by time); AI-native assets appreciate through use in a rise-then-saturate S-shape (ceiling exactly Paper I's \(\bar{x}\)), and can decay via obsolescence.
Liability direction: traditional liabilities are contracted and repaid to zero; AI-native liabilities are implicit, off-book, with operations-endogenous compounding, and two-way: left alone they compound, actively serviced they convert into assets.
Marginal cost: traditional marginal cost is hard to split per unit; AI-native variable marginal cost is accountable per token and trends downward.
Convertibility of liabilities (the key to symmetry)
Unlike traditional liabilities whose repayment is a net outflow to zero, servicing an AI-native liability is itself capitalized: addressing technical debt deposits architecture (\(\to A_4\)); unclogging a bandwidth bottleneck deposits "the capability to find and fix bottlenecks" (\(\to A_5\)); internalizing external dependency deposits owned capability (\(\to A_1/A_4\)).
3.2 · Breaking the circularity: exogenous shocks
If the direction of every account is set by \(A_5\), and \(A_5\) is itself an asset, the claim is self-confirming. We break the circularity with an exogenous shock \(Z_t\) (independent of \(A_5\)): (1) model-paradigm shocks (frontier capability jumps, date-stamped by release), (2) upstream-dependency shocks (API pricing/policy, framework deprecation), (3) regulatory shocks (rules revaluing \(C_2\)). The falsifiable content: post-shock recovery should be predicted by a measure of \(A_5\) constructed before the shock (adaptation latency from public changelogs). Because \(Z_t\perp A_5\), this is not circular.
4.FACE: Valuation Re-Anchored on Net-Worth Accumulation
where \(\Pi_0\) is a base scale (currently monetizable profit base), \(\mathcal{N}\) is per-period net-worth accumulation, \(h\) is the number of compounding periods over which saturation room monetizes, and \(\kappa\) is an external/market coefficient. The multiplicative form encodes compounding net-worth accumulation: net worth scales value by \((1+\mathcal{N})\) each period.
Let \(\Pi_0=\pi(x_0)/r\) and \(\kappa=1\). In the benchmark, choose the accumulation–horizon product to match the discounted saturation room,
\[ (1+\mathcal{N})^{h}-1 \;=\; \frac{G}{V_{\mathrm{DCF}}} \;=\; \frac{k}{k+r}\cdot\frac{\bar{x}-x_0}{x_0}. \]Then \(V_{\mathrm{FACE}}=V\) exactly. To first order \(h\,\mathcal{N} = \frac{k}{k+r}\cdot\frac{\bar{x}-x_0}{x_0}\), and the gap from path-blind DCF equals the present value of saturation room: \(V_{\mathrm{FACE}}-V_{\mathrm{DCF}}=G=\frac{k}{r(r+k)}S_0\).
With \(\Pi_0=x_0/r=V_{\mathrm{DCF}}\) and \(\kappa=1\), \(V_{\mathrm{FACE}}=V_{\mathrm{DCF}}(1+\mathcal{N})^h\). Imposing the match, \((1+\mathcal{N})^h=1+G/V_{\mathrm{DCF}}\), gives \(V_{\mathrm{FACE}}=V_{\mathrm{DCF}}+G=V\). ∎
This is a consistency result: there exists a parameterization under which FACE coincides with \(V\). It does not claim uniqueness of the operator, nor that \((\mathcal{N},h,\kappa)\) are uniquely identified. Proposition 1 and the gap are the paper's hard results; FACE is the practical framework, Proposition 3 the bridge.
FACE versus real options — what is added
Real options price the right to discrete future decisions under uncertainty; value comes from volatility. FACE prices a deterministic state trajectory \(x_t\to\bar{x}\); value comes from the bounded map and the distance \(S_0\), not volatility. Real options have no intrinsic ceiling; FACE inherits \(\bar{x}\), giving the monotone-in-\(S_0\) comparative static. Real options are silent on why cash flow lags value; FACE derives the lead \(\Delta\) explicitly. The two are complements, not substitutes.
5.Why Value Leads Cash Flow: An Explicit Two-Rate Derivation
Split the "net worth → delivered value" chain into an accelerable part (execution, share \(p\)) and a non-accelerable part (judgment bandwidth, share \(1-p\)). Model capacity \(a(t)=1-e^{-g t}\) (execution-driven, fast) and realized monetization \(m(t)\) relaxing toward it at bandwidth-limited rate:
\[ \dot m(t) \;=\; \tfrac{1}{\tau}\big(a(t)-m(t)\big), \qquad m(0)=0 . \]The system has closed-form solution (for \(g\tau\ne 1\))
\[ m(t) \;=\; 1 \;-\; \frac{g\tau\,e^{-t/\tau} - e^{-g t}}{g\tau-1}. \]In the bottleneck regime \(\tau>1/g\), the realized curve lags capacity, and the horizontal lead at level \(1-\varepsilon\) is
\[ \Delta(\varepsilon) \;=\; \tau\,\ln\!\frac{g\tau}{g\tau-1} \;+\; \Big(\tau-\tfrac{1}{g}\Big)\ln\frac{1}{\varepsilon} \;>\;0 . \]The lead is strictly positive iff bandwidth binds (\(\tau>1/g\)), increases in \(\tau\), and increases in \(g\).
Solving the linear ODE with integrating factor \(e^{t/\tau}\) and \(a(t)=1-e^{-gt}\) gives the solution. The tail of \(1-m(t)\) is dominated, when \(g\tau>1\), by the slower mode \(e^{-t/\tau}\). Solving \(a(t_a)=1-\varepsilon\) and the dominant balance for \(t_m\), then \(\Delta=t_m-t_a\). ∎
Amdahl's law supplies only the steady-state ceiling \(1/(1-p)\). The lead is a transient property requiring the ODE. The sharp, testable prediction is counter-intuitive: accelerating execution \(g\) lengthens the lead, because it widens the gap between what capability can deliver and what bandwidth can monetize. This is the mechanistic root of "value systematically leads cash flow."
6.Testable Implications
We give a refutable interface with operational proxies. Sampling frame: U.S.-based AI-native firms with sufficient public operating trace (disclosure, not valuation, is the inclusion criterion). Exogenous shocks are date-stamped and dense (frontier-model releases now occur on a cadence of order weeks). Three-layer non-circular design: frame = disclosure density; treatment / pre-shock \(A_5\) = adaptation latency to prior model-paradigm shocks (measured before the focal shock); outcome = post-shock recovery (valuation enters only as predicted outcome).
Proxies for distance-to-saturation \(S_0\)
(1) Penetration: \(1-(\text{penetration}/\text{TAM})\). (2) Marginal-cost-to-frontier: gap of realized capability to a frontier benchmark. (3) Accumulation-velocity (inverse): growth rate of intangible-capital stock. Use at least two to guard against mechanical relation to valuation.
Predictions
P1. Controlling for cash flow, market value increases in the net-worth proxy, and the premium increases in \(S_0\). P2. Net worth leads cash flow; the distinctive cross-restriction is \(\partial\Delta/\partial S_0>0,\ \partial\Delta/\partial g>0,\ \partial\Delta/\partial\tau>0\)—what no generic intangibles-lead model predicts. P3. Lower industry marginal cost \(\Rightarrow\) higher \(g\) \(\Rightarrow\) longer lead and more severe underestimation. P4. Post-shock recovery predicted by pre-shock adaptation latency (\(Z_t\perp A_5\), non-circular); the dense release cadence permits a repeated-shocks panel—the natural first empirical target.
7.Conclusion
The underestimation of high-accumulation, low-current-cash-flow AI-native firms is the cost of a specific misspecification—anchoring on cash-flow history rather than the coverage path—strictly positive, closed-form, monotone in distance to saturation. FACE operationalizes the correctly anchored valuation; its gap from path-blind DCF equals the present value of saturation room. The value–cash-flow lead follows in closed form from a two-rate model with an Amdahl ceiling, yielding the testable comparative static that faster execution lengthens the lead.
(a) FACE's algebraic form is an organizing framework, not a unique operator. (b) The balance-sheet taxonomy is proposed; measurability awaits empirics. (c) The predictions require panel data—the bridge to Paper III (the new-species paradigm). (d) The coverage-dynamics kernel's rigor is governed by Paper I.
References (selected)
Amdahl, G. M. (1967). Validity of the single processor approach to achieving large scale computing capabilities. AFIPS Conference Proceedings.
Cunningham, W. (1992). The WyCash portfolio management system. OOPSLA Experience Report.
Dixit, A. K., & Pindyck, R. S. (1994). Investment under Uncertainty. Princeton University Press.
Haskel, J., & Westlake, S. (2018). Capitalism without Capital: The Rise of the Intangible Economy. Princeton University Press.
Lev, B. (2001). Intangibles: Management, Measurement, and Reporting. Brookings Institution Press.
Romer, P. M. (1990). Endogenous technological change. Journal of Political Economy, 98(5), S71–S102.
Mei, J. (2026). Saturation Decoupling: Finite Firm Value under Unbounded Capital Accumulation. Working paper (Paper I).
饱和空间与企业价值
Saturation Room and Firm Value: A Valuation Framework for AI-Native Firms
工作论文 (v1) — 2026 年 6 月 · maymedoc@gmail.com
摘要
建立于 Paper I 的饱和解耦定理之上,本文处理一个估值问题:当一家企业的价值来自未实现的饱和空间——覆盖度 \(x_t\) 与其上限 \(\bar{x}\) 之间的差距——而非来自当前现金流时,应如何估值?我们的核心结果是一个误设代价定理(命题 1):任何从已观测现金流外推、而不显式建模覆盖路径 \(x_t\to\bar{x}\) 的估值都会低估此类企业,低估幅度严格为正,且随距饱和距离单调上升;而一个能正确预测路径的估值则精确回收真实价值。因此,低估不是折现本身的性质,而是锚定的性质。随后我们定义 FACE,一个重新锚定于净值累积(\(\mathcal{N}\),非货币资产负债表上的资产减负债)的估值,并证明一个一致性结果(命题 3)。最后,我们从一个显式双速率动力系统推导价值–现金流领先滞后结构:其中执行项指数累积,而货币化以受带宽限制的速率向 Amdahl 上限弛豫;领先量 \(\Delta\) 以闭式求出,且当且仅当判断带宽绑定时为正(命题 4)。
关键词:企业估值;无形资产;内生增长;饱和;价值–现金流领先滞后;AI-native 企业。 JEL:G31, D21, O33, M21。
AI 使用披露。 作者使用了 AI 辅助工具(Claude,Anthropic)协助起草、编辑以及对闭式结果的符号化验证。所有理论主张、定义、证明与最终文本均为作者本人所作,并经作者审阅与核验;作者对全部内容负全部责任。
1.引言
一类企业正变得普遍:它们累积迅速,却只赚取微薄的当前现金流。它们投资于构建某种能力底座(模型、数据、用户、知识资本),这种底座几乎可以无界累积,但在很长时间里只把其中一小部分货币化为利润。标准 DCF,按常见实施方式——锚定于当前或近期现金流并向前外推——对此类企业得出一个令人不安的结论:它们"不值多少钱"。然而市场往往给予它们极高估值。这一缺口不是非理性,而是估值工具的锚定方式与企业结构之间的失配。
Paper I 证明了饱和解耦:当一个累积要素 \(K\) 无界、但价值通过一个有界的饱和映射 \(x=\sigma(K)\to\bar{x}\) 进入利润时,企业价值对任意折现率都有限。Paper I 回答了"价值是否有限"。本文回答下一个问题:在有限性成立的前提下,此类企业应如何估值,以及标准方法究竟在哪里、为何出错?
我们证明低估是一个特定误设的代价——未能建模覆盖路径 \(x_t\to\bar{x}\)——且该代价严格为正、有闭式、并随距饱和距离单调上升。对称地,一个确实预测覆盖路径的估值会精确回收真实价值 \(V\)。FACE 不是 DCF 的替代品;它就是把锚从"现金流历史"移到"驱动覆盖度的状态"之后所得到的那个 DCF。
2.价值的真正来源,与错误锚定的代价
沿用 Paper I:状态 \(x_t\) 向上限 \(\bar{x}\) 演化,价值通过一个对 \(x\) 非递减的有界映射 \(\pi(x)\) 进入瞬时利润,企业价值为 \(V=\int_0^\infty e^{-rt}\pi(x_t)\,dt\)。定义企业在时刻 \(t\) 的饱和空间为未实现的覆盖差距
\[ S_t \;=\; \bar{x} - x_t \;\ge\; 0 . \]饱和空间是价值的真正来源:\(S_t\) 大的企业,其未来利润将随 \(x_t\) 向 \(\bar{x}\) 上升而增长;\(S_t\approx 0\) 的企业则没有这样的增量。
2.1 · 误设代价定理
一种估值若从已观测/当前利润外推而不建模覆盖路径,则称为现金流锚定(路径盲);其极限情形是 \(V_{\mathrm{DCF}}=\pi(x_0)/r\)。一种估值若显式建模 \(x_t\to\bar{x}\) 并沿该路径贴现 \(\pi(x_t)\),则称为路径预测。
设 \(\pi\) 非递减,\(x_t\) 非递减且 \(x_t\uparrow\bar{x}\)。则:
(i) 低估。 \(G = V - V_{\mathrm{DCF}} = \int_0^\infty e^{-rt}[\pi(x_t)-\pi(x_0)]\,dt > 0.\)
(ii) 关于距饱和距离的单调性。 \(G\) 关于 \(S_0=\bar{x}-x_0\) 递增。对基准 \(\pi(x)=x\),\(x_t=\bar{x}-(\bar{x}-x_0)e^{-kt}\),
\[ G = \frac{k\,(\bar{x}-x_0)}{r\,(r+k)} = \frac{k}{r(r+k)}\,S_0, \qquad \frac{\partial G}{\partial S_0}=\frac{k}{r(r+k)}>0 . \](iii) 路径预测回收 \(V\)。 一个预测真实覆盖路径的估值精确等于 \(V\)。差额 \(G\) 因此是路径盲误设的代价,而非折现的性质。
(i) 从 \(V\) 中减去 \(V_{\mathrm{DCF}}\);由 \(x_t\ge x_0\) 及 \(\pi\) 非递减,被积函数 \(\ge 0\) 且在正测度集上严格大于零。(ii) 更大的 \(S_0\) 逐点抬高被积函数;闭式由计算两个积分得到。(iii) \(\int_0^\infty e^{-rt}[\bar{x}-(\bar{x}-x_0)e^{-kt}]dt=(k\bar{x}+rx_0)/[r(k+r)]=V\)。∎
一个正确实施的多阶段 DCF 能够预测一条增长路径,且只要它正确预测了 \(x_t\to\bar{x}\),它就落在 \(V\) 上。定理的内容恰是其逆否:实践中观察到的低估应归因于路径盲,且只要从业者锚定于现金流历史而非驱动覆盖度的状态,低估就被闭式从下方界定。
3.驱动饱和空间的账户:一张非货币资产负债表
什么驱动了路径 \(x_t\to\bar{x}\)?我们把支配覆盖度增长的非货币要素记入表内。一张非货币资产负债表:资产 \(A\) 加速覆盖度增长,负债 \(C\) 减缓之,净值 \(\mathcal{N}=A-C\)。
\(A_1\) 数据 / 知识库存量 · \(C_1\) 技术债
\(A_2\) 用户 / 分发网络 · \(C_2\) 信任 / 合规债
\(A_3\) 判断资本(品味、决策质量) · \(C_3\) 带宽债(核心注意力被占用)
\(A_4\) 流程 / 工具资本 · \(C_4\) 依赖债(过度依赖外部能力)
\(A_5\) 元能力(发现并修复瓶颈) · \(C_5\) 组织协调成本
设覆盖度增长由净值驱动:\(\dot{x}\propto g(\mathcal{N})\cdot(\bar{x}-x)\)——净值越高,向饱和演化越快,仍被 \(\bar{x}\) 钳制。净值因此是一个领先现金流的状态变量。
3.1 · 为何这些项目配称"资产/负债"
资产轨迹:传统资产在购入时见顶,随后单调折旧趋于 \(0\)(\(dA/dt<0\),由时间驱动);AI-native 资产通过使用而增值,呈先升后饱和的 S 形(上限恰为 Paper I 的 \(\bar{x}\)),并可因过时衰减。
负债方向:传统负债有合约、偿还时趋于零;AI-native 负债是隐性、表外、复利率内生于运营的,且双向:搁置则复利累积,主动处理则转化为资产。
边际成本:传统边际成本难以按单位拆分;AI-native 的可变边际成本可按 token 核算并趋势性下降。
负债的可转化性(对称性的关键)
与偿还即趋零净流出的传统负债不同,服务一项 AI-native 负债本身即被资本化:处理技术债沉淀出更好的架构与可复用工具(\(\to A_4\));疏通带宽瓶颈沉淀出"发现并修复瓶颈的能力"(\(\to A_5\));内化外部依赖沉淀出自有能力(\(\to A_1/A_4\))。
3.2 · 打破循环:外生冲击
若每个账户的方向都由 \(A_5\) 决定,而 \(A_5\) 本身又是一项资产,该主张就自我确认。我们通过一个外生冲击 \(Z_t\)(独立于 \(A_5\))来打破循环:(1) 模型范式冲击(前沿能力跃迁,由发布日期定日),(2) 上游依赖冲击(API 定价/政策、框架弃用),(3) 监管冲击(重估 \(C_2\) 的规则)。可证伪内容:冲击后的恢复应由一个在冲击之前构造的 \(A_5\) 度量(由公开更新日志读取的适配延迟)预测。由于 \(Z_t\perp A_5\),这并不循环。
4.FACE:重新锚定于净值累积的估值
其中 \(\Pi_0\) 是基准尺度(当前可货币化的利润基底),\(\mathcal{N}\) 是每期净值累积,\(h\) 是净值累积的复利期数(饱和空间货币化的视野长度),\(\kappa\) 是外部/市场系数。乘法形式编码"复利式净值累积":净值每期把价值放大 \((1+\mathcal{N})\) 倍。
设 \(\Pi_0=\pi(x_0)/r\)、\(\kappa=1\)。在基准下,选取累积–视野乘积以匹配贴现饱和空间,
\[ (1+\mathcal{N})^{h}-1 \;=\; \frac{G}{V_{\mathrm{DCF}}} \;=\; \frac{k}{k+r}\cdot\frac{\bar{x}-x_0}{x_0}. \]则 \(V_{\mathrm{FACE}}=V\) 精确成立。一阶匹配条件为 \(h\,\mathcal{N} = \frac{k}{k+r}\cdot\frac{\bar{x}-x_0}{x_0}\),且其与路径盲 DCF 的差额等于饱和空间的现值:\(V_{\mathrm{FACE}}-V_{\mathrm{DCF}}=G=\frac{k}{r(r+k)}S_0\)。
以 \(\Pi_0=x_0/r=V_{\mathrm{DCF}}\)、\(\kappa=1\),有 \(V_{\mathrm{FACE}}=V_{\mathrm{DCF}}(1+\mathcal{N})^h\)。代入匹配式,\((1+\mathcal{N})^h=1+G/V_{\mathrm{DCF}}\),故 \(V_{\mathrm{FACE}}=V_{\mathrm{DCF}}+G=V\)。∎
命题 3 是一个一致性结果:存在一种参数化使 FACE 与 \(V\) 重合。它不主张该算子唯一,也不主张 \((\mathcal{N},h,\kappa)\) 被唯一识别。命题 1 与差额是本文的硬结果;FACE 是据此提出的实践框架,命题 3 是桥梁。
FACE 对实物期权——多出了什么
实物期权定价的是在不确定性下进行离散未来决策的权利;价值来自波动率。FACE 定价的是一条确定性状态轨迹 \(x_t\to\bar{x}\);价值来自有界饱和映射与距离 \(S_0\),而非波动率。实物期权没有内在上限;FACE 继承 \(\bar{x}\),给出关于 \(S_0\) 单调的比较静态。实物期权对滞后机制保持沉默;FACE 显式导出领先量 \(\Delta\)。二者互补而非替代。
5.为何价值领先现金流:一个显式双速率推导
把"净值 → 交付价值"链条拆为可加速部分(执行,份额 \(p\))与不可加速部分(判断带宽,份额 \(1-p\))。建模能力 \(a(t)=1-e^{-g t}\)(执行驱动,快)与已实现货币化 \(m(t)\) 以带宽受限速率向其弛豫:
\[ \dot m(t) \;=\; \tfrac{1}{\tau}\big(a(t)-m(t)\big), \qquad m(0)=0 . \]上述系统有闭式解(\(g\tau\ne 1\))
\[ m(t) \;=\; 1 \;-\; \frac{g\tau\,e^{-t/\tau} - e^{-g t}}{g\tau-1}. \]在瓶颈区间 \(\tau>1/g\),已实现曲线滞后于能力曲线,在饱和水平 \(1-\varepsilon\) 处的水平领先量为
\[ \Delta(\varepsilon) \;=\; \tau\,\ln\!\frac{g\tau}{g\tau-1} \;+\; \Big(\tau-\tfrac{1}{g}\Big)\ln\frac{1}{\varepsilon} \;>\;0 . \]领先量当且仅当带宽绑定(\(\tau>1/g\))时严格为正,并随 \(\tau\) 增大、随 \(g\) 增大。
以积分因子 \(e^{t/\tau}\) 解该线性 ODE 并代入 \(a(t)=1-e^{-gt}\) 得解。\(1-m(t)\) 的尾部当 \(g\tau>1\) 时由较慢模式 \(e^{-t/\tau}\) 主导。解 \(a(t_a)=1-\varepsilon\) 与主导平衡得 \(t_m\),则 \(\Delta=t_m-t_a\)。∎
Amdahl 定律只提供稳态上限 \(1/(1-p)\)。领先量是一个瞬态性质,需要微分方程。经济上尖锐且可检验的预测是反直觉的:加速执行 \(g\) 反而拉长领先,因为它扩大了"能力所能交付"与"带宽所能货币化"之间的差距。这是"价值系统性领先现金流"的机制根源。
6.可检验的推断
我们给出一个可证伪接口并配以可操作代理。样本框:有足够公开运营痕迹的美国 AI-native 企业(以披露、而非估值为纳入标准)。外生冲击有日期戳且密集(前沿模型发布如今以周为量级出现)。三层非循环设计:样本框 = 披露密度;处理 / 冲击前 \(A_5\) = 对此前模型范式冲击的适配延迟(在焦点冲击之前测量);结果 = 冲击后恢复(估值仅作为被预测的结果进入)。
距饱和距离 \(S_0\) 的代理
(1) 渗透率:\(1-(\text{渗透}/\text{TAM})\)。(2) 距前沿成本:已实现能力与前沿基准之差。(3) 累积速度(反向):无形资本存量增速。使用一个以上代理以防止与估值机械相关。
预测
预测 1. 控制现金流后,市值随净值代理上升,且溢价随 \(S_0\) 上升。预测 2. 净值领先现金流;区分性交叉约束为 \(\partial\Delta/\partial S_0>0,\ \partial\Delta/\partial g>0,\ \partial\Delta/\partial\tau>0\)——任何通用"无形领先"模型都不预测。预测 3. 行业边际成本越低 \(\Rightarrow g\) 越高 \(\Rightarrow\) 领先越长、低估越严重。预测 4. 冲击后恢复由冲击前适配延迟预测(\(Z_t\perp A_5\),非循环);密集节奏允许重复冲击面板——自然的首个实证目标。
7.结论
高累积、低当前现金流 AI-native 企业的低估是一个特定误设的代价——锚定于现金流历史而非覆盖路径——它严格为正、有闭式、随距饱和距离单调上升。FACE 把那个正确锚定的估值操作化;其与路径盲 DCF 的差额等于饱和空间的现值。价值–现金流领先从一个带 Amdahl 上限的双速率模型以闭式导出,得到一个可检验的比较静态:加速执行反而拉长领先。
(a) FACE 的代数形式是组织性框架,而非唯一算子。(b) 资产负债表分类是提议性的;可计量性尚待实证。(c) 预测需要面板数据——通往 Paper III(新物种范式)的桥梁。(d) 覆盖动力学内核的严谨性由 Paper I 管辖。
参考文献(节选)
Amdahl, G. M. (1967). Validity of the single processor approach to achieving large scale computing capabilities. AFIPS Conference Proceedings.
Cunningham, W. (1992). The WyCash portfolio management system. OOPSLA Experience Report.
Dixit, A. K., & Pindyck, R. S. (1994). Investment under Uncertainty. Princeton University Press.
Haskel, J., & Westlake, S. (2018). Capitalism without Capital. Princeton University Press.
Lev, B. (2001). Intangibles: Management, Measurement, and Reporting. Brookings Institution Press.
Romer, P. M. (1990). Endogenous technological change. Journal of Political Economy, 98(5), S71–S102.
Mei, J. (2026). Saturation Decoupling: Finite Firm Value under Unbounded Capital Accumulation. Working paper (Paper I).