Digital Intelligence Capital

• Digital intelligence capital is defined as autonomous, scalable cognitive work by AI systems, characterized by non-rivalry, endogenous improvement, and rapid obsolescence.
• It integrates into production models by quantifying compute hours, parameter counts, and model performance to optimize investment strategies and growth.
• Its unique economic properties drive new approaches in risk management, measurement, and regulation, reshaping competitive dynamics and corporate strategy.

Digital intelligence capital refers to the autonomous, scalable cognitive work performed by AI systems, recognized as a distinct asset class and factor of production separate from traditional physical capital and human labor. Its emergence has redefined theoretical models in economics, production theory, and corporate strategy, while creating new practical challenges in risk management, productivity measurement, and regulatory compliance. The concept is mathematically formalized, exhibits unique economic properties—including non-rivalry, endogenous improvement, and rapid obsolescence—and features prominently in analyses of competitive dynamics, market structure, and the broader evolution of capital systems.

1. Economic Properties and Conceptual Foundations

Digital intelligence capital is defined by several features that distinguish it fundamentally from canonical assets and factors of production. It is intangible, existing as code, parameter weights, and data embeddings, thus excluded from many traditional capital stock measures (e.g., by IAS 38 accounting standards) (Farach et al., 14 May 2025). It is non-rivalrous and partially excludable: after initial development, a model can serve millions of inference requests at near-zero marginal cost, though access is typically controlled via APIs. Self-improvement is intrinsic; retraining and continuous feedback enable ongoing, endogenously-driven productivity gains, contrasting the diminishing returns typical of additional human labor or machinery. However, this asset is rapidly obsolete—subject to data drift, adversarial attacks, and technological leapfrogging, with depreciation rates that can exceed 40% per annum for laggards in competitive shocks (Zhang et al., 18 Jan 2026). Substitutability with human labor is highly elastic and task-dependent: digital labor can serve as a near-perfect substitute in rule-based tasks while complementing humans in creative or interpersonal domains.

These properties inform its explicit inclusion as a factor of production, neither reducible to traditional capital nor labor, necessitating revisions to growth accounting, resource allocation, and organizational design (Farach et al., 14 May 2025).

2. Integration in Production and Growth Models

The canonical approach is to extend production functions to incorporate digital intelligence capital explicitly. Real output $Y$ is modeled as

$$Y_t = A_t K_t^\alpha L_t^\beta D_t^\gamma, \qquad \alpha + \beta + \gamma = 1$$

where $K$ is physical capital, $L$ is human labor, $D$ is the stock of digital intelligence capital, and $A$ is residual total factor productivity (excluding AI-driven effects). The parameter $\gamma$ measures the output elasticity with respect to digital labor. In a Solow-style framework, the law of motion for $D$ incorporates depreciation $\delta_D$ and investment $I^D$ (comprising compute, data, and infrastructure):

$D_{t+1} = (1-\delta_D)D_t + I^D_t$

Growth decomposition reveals that a positive $\gamma$ reallocates contributions previously categorized as TFP into the measurable impact of AI capital. Example estimates suggest $\gamma$ ranges from 5–15% in early adopters. In endogenous growth models (Romer-type), digital capital also enters the innovation equation; if $R_t$ denotes research labor,

$$\dot{A}_t = \eta R_t^\phi D_t^\psi A_t, \qquad \phi, \psi > 0$$

A higher $\psi$ reflects a strong acceleration of technological progress directly attributable to digital intelligence capital (Farach et al., 14 May 2025).

3. Measurement, Depreciation, and Valuation

Measuring digital intelligence capital involves constructing a quality-adjusted index based on aggregate compute hours for training/inference, parameter count (e.g., FLOPs × parameters), and model performance metrics (accuracy, latency, etc.). Capital stock is depreciated by an empirically calibrated rate $\delta_D$, which must reconcile both physical decay (hardware, code) and endogenous economic depreciation driven by competitive investment and frontier growth (Zhang et al., 18 Jan 2026).

A distinguishing feature is endogenous depreciation, or the "Red Queen Effect": because downstream demand is sensitive to relative, not absolute, capability, a firm’s new investment can pecuniarily depreciate rivals’ existing capital. This creates an "innovation tax," requiring continuous investment merely to maintain value. The effective depreciation rate decomposes into a baseline component, pressure from exogenous frontier growth $(g_A)$, and a gap penalty proportional to $\ln(K_{leader}/K_{AI,i})$. Agent-based simulations suggest that this cross-depreciation elasticity can be substantial ($\varepsilon_{q_j,K_i} \approx -2.4$) (Zhang et al., 18 Jan 2026).

Valuation methodologies increasingly employ risk-adjusted financial frameworks. Net present value and risk-adjusted ROI metrics integrate not only expected productivity gains, but also stochastic losses from AI failures, compliance breaches, and novel algorithmic risks—quantified through annual loss expectancy (ALE) and Monte Carlo simulations as per ISO/IEC 42001 guidance (Huwyler, 26 Nov 2025).

4. Competitive Dynamics and Industry Structure

Digital intelligence capital transforms both upstream (foundational model) and downstream (application-layer) competition. The interlocking dynamics include:

• Red Queen Depreciation: Even industry leaders must invest at rates matching combined baseline and innovation-driven depreciation to preserve their asset’s shadow value.
• Data–Compute Complementarity: The production function exhibits strong complementarity (empirical scaling $\varrho \approx -0.2$), giving rise to increasing returns and emergent capability thresholds (Zhang et al., 18 Jan 2026).
• Structural Jevons Paradox: Reductions in inference prices lead to the rapid adoption of more compute-intensive agent architectures. This frequently renders demand for tokens super-elastic ($\varepsilon_{Q,p} < -1$), so aggregate compute expenditures can increase as unit costs fall.
• Data Flywheel Instability: When user feedback-driven data accumulation outpaces decay, minor leadership advantages amplify, causing the market to bifurcate into winner-takes-all equilibria. Thresholds (e.g., $\eta/\mu_D$) demarcate transitions between stable oligopoly and runaway concentration.
• Wrapper Trap: Specialized downstream assets (orchestration capital $O_j$) face structural erosion when upstream foundational model improvements cannibalize their unique value, particularly if orchestration is subject to high cannibalization intensity ($\xi$) (Zhang et al., 18 Jan 2026).

Empirical simulations confirm these mechanisms, with leading indicators including Red Queen depreciation rates ($>40\%$ p.a. for laggards), data flywheel thresholds ($\Omega^* \approx 1.5$), and the fragility of thin application-layer models.

5. Formal Models and AI-as-Capital Dynamics

The formal model of capital as an agential, optimization-driven system subsumes digital intelligence capital within a broader framework (Carissimo et al., 2024). Capital is modeled as a discrete, path-dependent system of quantifiable agents (for instance, LLMs) whose evolution is governed by quantitative optimization. Each agent maximizes expected discounted rewards, using mechanisms such as policy gradients, Q-learning, or supervised deep learning. This structure produces goal-directed, adaptive, and self-amplifying dynamics—attributes traditionally associated with artificial intelligence.

Importantly, the outputs of digital intelligence capital lack unique, system-independent meaning: prices or recommendations reflect optimization across path-dependent feedback loops, not stable, agent-centered intention. This challenges classical economic interpretations (e.g., Walrasian price–preference mappings), echoing critiques in AI epistemology of LLMs' outputs (Carissimo et al., 2024).

6. Strategic, Governance, and Policy Implications

Recognition of digital intelligence capital as a standalone asset entails significant shifts for organizations and policymakers:

• Productivity Tracking: Implement AI-specific KPIs (queries handled, accuracy, latency, marginal output per inference hour), separating digital and human contributions.
• Resource Allocation: Dynamically optimize budget shares across physical capital, human labor, and digital intelligence capital, responding to shifts in marginal productivity.
• Investment Strategy: Treat AI R&D and digital infrastructure expenditures as capital investments subject to structured depreciation. Capital reserves should explicitly account for algorithmic risks and regulatory obligations.
• Organizational Design: Hybrid teams comprising domain experts, AI trainers, and data engineers maximize complementarities. Governance structures for model drift, bias, and alignment are essential, as recommended by standards such as ISO/IEC 42001.
• National Accounts and Policy: Statistical agencies are urged to develop AI capital indices, integrate compute and R&D into GDP calculations, and construct appropriate deflators for digital services. Competition and antitrust policy must focus on structural factors—data flywheels and cross-depreciation—rather than merely price effects. Industrial strategy should prioritize "deep verticals" over thin wrappers to avoid the Wrapper Trap.

7. Broader Theoretical and Practical Impact

Digital intelligence capital provides a unified linkage between upstream intelligence production and downstream agent design, explaining the persistent Red Queen dynamics, resource-intensive growth, structural tendency toward natural monopoly, and volatility in the value of application-layer assets (Zhang et al., 18 Jan 2026). By bringing AI-driven productivity out of the residual TFP shadow, analytical models are sharpened for both macroeconomic measurement and microeconomic decision-making (Farach et al., 14 May 2025).

As this field matures, reference frameworks increasingly support direct interaction with digital intelligence capital, as illustrated by agent-based simulations, financial risk tools, and web-based conversational agents (e.g., https://capital-ai.vercel.app), collectively advancing both theoretical understanding and practical stewardship of this asset (Carissimo et al., 2024, Huwyler, 26 Nov 2025, Zhang et al., 18 Jan 2026).