AI Reshaping Capital's Trajectory: When Asset Definitions Shift, How Can Investment Logic Be Reconstructed?

Under the impact of the AI wave, we must re-examine the six "self-evident" assumptions that have underpinned investment decisions for decades. First, investment traditionally viewed targets as "collections of people," with human capital as a core component of the production function and a company's moats. For example, valuation and growth models for outsourcing and IT services were built on the human leverage of "recruit—train—price up—replicate." Human capital was considered a depreciable, leverageable, and premium-earning asset. [Verified: Classic Common Sense + FR-12 Production Function Discussion]

Second, value was perceived as inherent in the asset, measurable by discounted future cash flows (DCF) or comparable multiples—the classic financial definition of "forgoing current consumption for future returns." [Verified: Classic Literature (Prevailing Financial Framework)] Third, growth was seen as requiring capital to acquire human resources and operations, with economies of scale stemming from decreasing marginal costs, especially in rapidly replicable domains like software. [Verified: Industry Standard Knowledge + Contrast with Bessemer below]

Fourth, risk was classifiable and compensable through risk premia, such as diversifying portfolios to reduce unsystematic risk, with higher risk corresponding to higher expected returns. [Verified: Classic Literature (Prevailing Portfolio Theory Framework)] Fifth, returns followed a power law, with winners capturing long-term gains through "asset ownership," and exit paths primarily being IPOs or M&A, where a few big winners covered numerous failures. [Verified: Classic Literature (VC Power Law Consensus)] Finally, information asymmetry was primarily addressed through due diligence, with core elements being team, market, and financials ("investing in people, investing in the sector, investing in financial models"). [Verified: Industry Standard Framework] These assumptions collectively formed the "human resources ontology" of investment, wherein investment entities were people and assets driven by people.

Not all technological changes constitute a paradigm revolution. Our judgment of whether a transformation reaches the level of "revolution" depends on whether four core links of the investment chain—"asset ontology (what to invest in), value anchor (why it's valuable), risk sources (what to fear), and return structure (how to profit)"—undergo fundamental shifts, and whether the "rights" acquired by investors change. If AI is merely used to enhance efficiency or assign higher valuations to AI companies, it is merely "deepening" within the old framework; however, if the asset definition, valuation anchors, nature of risk, and even investor rights fundamentally transform, it signifies a "revolution."

For instance, using AI to improve efficiency for portfolio companies is deepening, but investing in "AI-enhanced production units" (whose value derives from leverage structure and boundaries rather than human capital) is revolutionary. Adjusting DCF discount rates or adding AI-specific due diligence checklists is deepening, while shifting valuation anchors to "verifiable unit economics × reusability × convexity × boundary enforceability" is revolutionary. Similarly, traditional risks centered on execution, market, and management, but when the primary subjects of risk in the AI era transform into "capability depreciation, model drift, compute concentration, verification tax, and liability exposure," this portends a profound reconstruction.

The core lies in this: when execution is no longer scarce, when revenue decouples from headcount, and when the most valuable assets can be openly replicated, the investment target ontology shifts from "collections of people" to "leverage and boundaries," and the value anchor moves from discounted cash flows to convexity and trustworthiness. This is not merely a change in valuation figures but a fundamental reshaping of the underlying logic, defining what is "investable" and what is "valuable."

The rise of AI does not merely boost efficiency; it fundamentally undermines the cornerstones of the old paradigm through seven mechanisms. First is the collapse of execution costs, downgrading human labor from "irreplaceable execution units" to "compute-substitutable capacity." The marginal execution costs for tasks like content generation, code writing, and customer service response have dramatically decreased, shattering the old equation that "more people equals more output." As concluded in FR-10, the scarcity of talent lies in problem definition, tool mastery, and accountable verification, not simple execution. [Inference, based on FR-10 + FR-12]

Second, revenue has decoupled from headcount, giving rise to investable entities characterized by "minimal units, maximal output." For instance, according to KuCoin/AIMPACT, citing Epoch AI's estimates, the per-employee revenue for Anthropic, OpenAI, and Nvidia reached approximately $9 million, $5.6 million, and $5.1 million, respectively ([KuCoin/AIMPACT citing Epoch AI, 2026-05](https://www.kucoin.com/news/flash/anthropic-generates-9-million-in-revenue-per-employee-surpassing-openai-and-tech-giants)). Furthermore, Midjourney is estimated to have achieved approximately $500 million in annualized revenue with around 40 employees and zero external funding ([Value Add VC, 2026-07](https://valueaddvc.com/blog/how-does-midjourney-make-money-500m-arr-zero-funding-and-the-ai-image-business-model-explained)). This renders traditional "per-headcount valuation" methods obsolete, necessitating new anchors for asset value, consistent with FR-13's assertion that "what is truly accumulable is process capital and trust capital." [Inference, based on M2 + FR-13]

Third, value anchors are migrating from "cash flow/multiples" to "capability × reusability × verifiability × boundaries." When generation is cheap and replication costs are near zero, scarcity shifts from production to verification, attribution, distribution, and trust. Due diligence thus transforms to focus on "whether unit economics are verifiable, how high reusability is, and whether boundaries are enforceable." For instance, Tomasz Tunguz notes that by late January 2025, the median forward ARR multiple for public AI software companies was approximately twice that of non-AI software companies (11.2x vs 5.5x), and AI companies saw a significant increase in both their share of B2B Series A funding and valuation premium ([Tomasz Tunguz, 2025-02-03](https://tomasztunguz.com/private_valuations_2025/)). This could either be a premium within the old paradigm or the pricing of a new paradigm revolution; the key lies in whether the premium purchases cash flows from human replacement or convexity and boundaries. [Inference]

Fourth, risk sources are shifting from "execution/human capital" to "capability depreciation/model drift/compute concentration/verification tax/liability." An AI-enhanced asset requires continuous operation, monitoring, and re-verification to retain value. Model upgrades, data distribution shifts, and API changes can all render yesterday's assets obsolete today. FR-12 has already argued that "cost disease migrates to the verification end," thus investment risk is being rewritten as capability depreciation rates, drift rates, concentration on single model/compute providers, verification tax, and liability exposure in case of errors. In the old paradigm, "people" were relatively stable, trainable, and replaceable assets, whereas in the new paradigm, "models" are rapidly depreciating assets that can be open-sourced, replicated, and subject to regulatory scrutiny. [Inference, based on FR-12/FR-13]

Fifth, the return structure is shifting from "sharing cash flows" to "sharing convexity." The extreme characteristics of non-rivalrous digital products mean that successful upside is virtually limitless, while the marginal cost of failure is limited. AI enables "successful products" to be replicated at near-zero cost, making the return function more akin to options than bonds. This explains why capital is willing to pay for AI assets with "high multiples, low gross margins, and high uncertainty": it is buying convexity. Concurrently, market data shows capital betting earlier and more concentratedly—"fewer bets, more capital" ([TechCrunch citing Carta data, 2026-03-20](https://techcrunch.com/2026/03/20/ai-startups-are-eating-the-venture-industry-and-the-returns-so-far-are-good/)). [Inference, based on M5]

Sixth, capital intensity returns: Contrary to the intuition that "AI makes everything lighter," AI infrastructure is capital-intensive. TrendForce has revised its forecast for the combined capital expenditure of the top nine global cloud vendors in 2026 to approximately $830 billion, a year-on-year increase of about 79% (with Microsoft at approximately $190 billion, AWS exceeding $230 billion, Google at $180-$190 billion, and Meta at $125-$145 billion) ([Evertiq citing TrendForce, 2026-05-06](https://evertiq.com/news/2026-05-06-ai-boom-pushes-hyperscaler-capex-towards-usd-830-billion-in-2026)). This implies that the investment paradigm is not unilaterally "de-capitalizing"; instead, the application layer is extremely light, while the infrastructure layer is extremely heavy, with capital allocation logic diverging between the two layers. [Inference]

Finally, the unit of account is migrating from seat/hour/EBITDA to outcome/credit/result. The per-seat SaaS model is being eroded by intelligent agents; the more capable the agent, the fewer seats a customer requires. Bain analysis shows that 65% of SaaS vendors have adopted a hybrid "seat + usage" model, and companies like monday.com have introduced "seat + credits" models, indicating that the basis for revenue measurement is being renegotiated, and investment valuations must ultimately follow this shift in units of account ([UC Today citing Bain, 2026-05-12](https://uctoday.com/is-per-seat-saas-pricing-dead-monday-coms-consumption-based-pricing-pivot)). [Inference]

The mechanisms described above are not merely theoretical deductions but are substantiated by the actual flow of capital, indicating a structural reconfiguration of investment logic. According to the OECD's "AI Venture Capital 2025" report, AI companies attracted approximately 61% of global VC investment throughout 2025, totaling about $258.7 billion, dominating the global aggregate of $427.1 billion. [Verified: Headline Checked/Second-hand Reporting/Primary Verification Pending] This clearly demonstrates that capital is concentrating in the AI sector at an unprecedented pace and scale, with its intensity and speed far exceeding any previous technology cycle.

The speed of this concentration is even more striking. Data from Reuters and PitchBook shows that in Q1 2025, AI investments accounted for approximately 57.9% of global VC investment (around $73.1 billion), whereas in the same period the previous year, this proportion was only about 28%. This means that within just twelve months, AI investment nearly doubled its share in the global VC landscape, reflecting capital's extreme bullishness and accelerated deployment towards AI's potential. [Verified: Headline Checked/Second-hand Reporting]

Furthermore, the phenomenon of "fewer bets, more capital" is becoming increasingly pronounced. TechCrunch, citing Carta data, reported that AI investments constituted approximately 41% of the total $128 billion raised on the Carta platform in 2025, further confirming the trend of high capital concentration in a few leading AI projects. [Verified: Headline Checked/Second-hand Reporting] Tomasz Tunguz, in his February 2025 report, also noted that the median forward Annual Recurring Revenue (ARR) multiple for public AI software companies was twice that of general software companies (11.2x vs 5.5x). Concurrently, AI companies accounted for as much as 70% of B2B Series A funding (up from only about 40% in early 2024), and AI Series A valuations were, on average, 40% higher. These data collectively paint a clear picture: capital is reconfiguring its allocation strategies in preparation for the new forms of value in the AI era. [Verified: Primary Blog (author is VC, data self-reported)]

Facing the erosion of the old paradigm and the emergence of new evidence, we need to establish a new investment logic. Under the new paradigm, the investment target ontology is redefined as an "AI-enhanced asset"—a production unit capable of transforming "rentable AI capabilities" into "verifiable outcomes." It is composed of four key elements: human agency (defining goals, verifying results, assuming responsibility), AI capabilities (rentable, replaceable), proprietary flows (data, processes, memory, distribution), and boundaries (exclusivity, verifiability, attribution). The investment key is no longer the stock of human capital or equipment it possesses, but rather the leverage effect of this production unit and the strength of its boundaries. [Inference, based on M1–M4]

In terms of value assessment benchmarks, the DCF method is gradually relegated to merely a "lower bound test," no longer serving as the dominant anchor. A new approximate valuation formula becomes: [verifiable unit economics × reusability] + convexity (option value) × boundary enforceability × distribution/trust depth. Traditional multiples like P/E ratios are no longer valuation methods but rather market pricing signals for an asset's "convexity" and "scarce boundaries." Concurrently, the definition of risk has fundamentally shifted, no longer primarily revolving around execution risk or management risk, but rather focusing on capability depreciation rates, model drift rates, concentration on single model/compute providers, verification tax, and potential liability exposure. Risk shows a negative correlation with the "thickness of the control layer" (e.g., verification, traceability, compliance, responsibility assumption).

Regarding return structure, the return function in the AI era exhibits more power-law characteristics. Returns equal cash flow (lower bound) plus convexity (upper bound) minus verification and replication costs. This implies that return distribution will be more skewed towards a few "boundary nodes," with potential decreases in diversification benefits at the portfolio level, while control quality and in-depth research for single-point investments become more critical. Furthermore, the core of investor rights is no longer solely traditional equity and ownership, but rather the right to access, orchestrate, verify, and settle transactions within the "boundary layer" of core AI capabilities. The role of capital also shifts from "purchasing human labor and scale" to "purchasing time and certainty"—that is, leveraging capital to accelerate iteration speed, improve data accuracy, build trust, and expand distribution networks, transforming uncertainty into verifiability.

The profound implication of this investment paradigm revolution is that the core scarcity in the economy is undergoing a fundamental shift from "labor" and "capital" to "trust" and "verification." As AI renders execution capabilities nearly infinite and inexpensive, true value capture will concentrate among individuals or entities capable of clearly defining problems, effectively mastering tools, accurately verifying results, and assuming responsibility for them (per FR-10). This means that in an AI-driven economic ecosystem, trust capital and verification capabilities will become new core competencies, and the accumulation and management of these intangible assets will become paramount.

This shift will inevitably lead to the ultra-concentration of capital. In a digital world of near-zero marginal cost replication, only a few "boundary nodes" with powerful technology, unique data, and brand moats will be able to capture the convex returns of extreme power-law distributions. These nodes can effectively establish exclusive advantages and continuously iterate, while downstream segments of the value chain will face increasingly fierce price competition and margin compression. This will reshape industrial landscapes and potentially exacerbate wealth inequality, prompting significant changes in socio-economic structures.

Concurrently, our traditional economic statistics and growth models will also face severe challenges. For instance, indicators like GDP and per capita output may struggle to accurately reflect new value creation models and wealth accumulation methods driven by AI. AI's utility often manifests as an exponential increase in production efficiency rather than linear human input, which may lead traditional metrics to underestimate true output. Therefore, developing new indicators such as "trust capital" and "verification tax" to more accurately measure economic activity and its cost-effectiveness will be crucial for understanding the healthy functioning of the AI economy; they represent the intangible yet decisive costs underpinning social cooperation and market fairness in the AI era.

In the face of AI-driven investment paradigm shifts, all participants must adjust their strategies. For entrepreneurs, the key lies in redefining "assets" and "moats." Focus should be on delivering "verifiable outcomes," rather than merely tools or capabilities, and on constructing "boundary layers" such as unique orchestration rights, access rights, verification rights, or distribution rights, to establish exclusive advantages. Product design should maximize "convexity" and "reusability," ensuring unit economics can be replicated at near-zero cost. Concurrently, when AI results may err, clarifying liability boundaries and building customer trust is in itself valuable "trust capital."

Investors, in turn, need to reconstruct their screening checklists and valuation frameworks. They should move beyond DCF and traditional multiples, treating DCF as a lower-bound test, and instead evaluate assets based on "verifiable unit economics × reusability × convexity × boundary enforceability × distribution/trust depth." Risk due diligence should focus on capability depreciation rates, model drift rates, concentration on single suppliers, verification tax, and liability exposure. In investment agreements, beyond equity, investors should strive for rights to access, orchestrate, verify, and settle transactions within the core "boundary layer." In a power-law distributed world, it is crucial to identify early a few winners with true "convexity" and "boundaries" and dare to concentrate investments, while also re-evaluating investment opportunities in infrastructure such as power, chips, and data centers.

For researchers and policymakers, the immediate task is to define new metrics and guide a new order. New economic indicators, such as "trust capital" and "verification tax," need to be developed to more accurately measure AI-driven economic activities. Furthermore, proactive legal and ethical frameworks should be established to address issues like liability attribution, data privacy, and compute monopolies brought about by AI, thereby safeguarding market fairness and public interest. In-depth research into the interplay and value creation mechanisms of different capital forms—such as equity, access rights, and data usage rights—within the context of "rentable capabilities" is fundamental to understanding future economic structures.