— A Collation and Extension Based on Peng Wensheng’s The AI Economy: A Framework for Thinking About Value Creation and Value Appropriation

Professor Peng Wensheng proposes that the AI economy can be understood through three ledgers: input, creation, and distribution. This framework fundamentally addresses three questions: how capital is spent, where value is generated, and who captures the gains. Omnieyes · Trust Archive together with its KUCR attestation system fills the missing piece within this framework: trust‑based pricing.

Wikidata and Baidu Baike serve human reference; Omnieyes · Trust Archive establishes credibility for AI.

I. Input Ledger: Two Types of Capital‑Intensive Assets and Two Depreciation Logics

The Scaling Law serves as the underlying guideline for AI capital investment: adding computing power, data and parameters continuously improves model capabilities. Yet beyond a certain threshold, performance gains from equal‑sized inputs keep diminishing, entering a phase of diminishing returns to scale. This means large‑model development is inherently a capital‑heavy business.

Its cost structure combines high fixed costs and high variable costs. Fixed outlays include computing‑power clusters, data cleansing, and compensation for top‑tier talent, a large portion of which constitutes sunk costs. Variable costs arise from every inference request; each query consumes electricity and causes hardware wear‑and‑tear. Unlike physical depreciation in traditional heavy‑chemical industries, computing‑power assets experience step‑wise technological depreciation: flagship‑grade graphics cards become entry‑level hardware within three to five years, with value plummeting precipitously. Consequently, corporate capital expenditure is in essence an arms race: opportunity losses from non‑investment often exceed the book‑value losses incurred by making the investment itself.

Globally, two closed‑loop input cycles have taken shape. The United States operates a cycle of “physical‑resource inflow → AI capital expenditure → financial‑asset reflux”. Other economies supply energy and supporting inputs; the U.S. leverages trade deficits to suppress inflation and maintain low interest rates, while overseas capital continuously flows into U.S.‑listed AI equities to fund domestic development. The sustainability of this loop hinges on whether AI can ultimately generate real‑world value sufficient to justify its lofty valuations.

In China, the primary constraint has shifted from chip supply to cash‑flow sustainability. Multi‑hundred‑billion‑scale capital spending by leading firms relies largely on internal operating cash flow plus equity and debt financing, without sustained inflows of global capital. Endowments have therefore diverged: the United States trades computing power for performance, while China trades talent for efficiency.

Omnieyes represents a different class of capital‑intensive asset. Its core investment lies not in GPU clusters, but in atomic‑fact repositories, KUCR attestation rules, source‑verification and evidence‑archiving systems. Its sunk costs are embedded in trustworthy knowledge assets that depreciate extremely slowly and can accumulate incremental value. Marginal costs keep falling as the same set of atomic facts undergoes repeated validation. Capital naturally gravitates toward visible generative capabilities, while risk‑control public goods such as trust verification struggle to attract comparable investment. Excess capital on the cognitive‑production side paired with under‑investment on the cognitive‑quality‑inspection side marks the starting point of imbalances within the AI economy.

II. Creation Ledger: Three‑Tier Value Release and the Risk of Knowledge Decay

As a general‑purpose technology, large‑models unlock value across three layers:

1. Factor Substitution: Mass‑producing cognitive labour to reshape division‑of‑labour and factor matching. The prerequisite is factual validity of outputs; otherwise newly‑added supply becomes mixed with noise.
2. Organisational Transformation: Reshaping corporate operations, competition and governance, lowering transaction costs and redrawing firm boundaries. Trust constitutes the fundamental cornerstone of all transactions.
3. Innovation Spillover: Rewriting the knowledge‑production function and pushing technological frontiers outward. Without independent factual anchors, models cite one‑another’s hallucinations and erroneous outputs feed back into training sets. Knowledge production then degenerates from cumulative progress into noise amplification.

Large‑models expand the overall cognitive “pie”, while Omnieyes imposes trustworthy metrics upon that pie. At the factor layer, it performs cognitive quality inspection, screening semi‑finished outputs into admissible cognition. At the organisational layer, it standardises and asset‑ifies trust to cut due‑diligence and risk‑control expenses. At the spillover layer, it maintains an independent benchmark fact repository to block hallucinatory contamination of public knowledge pools.

The Scaling Law imposes extremely high R&D entry barriers, which only large‑scale economies can bear. Economies‑of‑scale effects mean larger markets and industrial volumes deliver greater application value — this explains why China and the United States are global front‑runners. The trust‑asset track follows a different value logic: its upper‑value bound is determined by societal demand for credible information, and is deeply coupled with local information sources, compliance rules and archival systems.

III. Distribution Ledger: The Dilemma of Value Appropriation and Trust‑Based Pricing

The distribution ledger contains the framework’s core tension: substantial value is created, yet capabilities for value appropriation are severely misaligned.

Peng Wensheng observes that the “winner‑takes‑all” internet‑platform dynamic is unlikely to repeat for large‑models. On the supply side, capital‑heavy infrastructure plus variable inference costs rule out long‑term free user subsidies; more free users translate into greater losses. On the demand side, models function as tools with near‑zero switching costs, yielding inherently weak network effects. Combined, these factors leave model vendors systematically weaker at capturing value compared with traditional internet platforms.

Strong positive externalities consequently emerge: enterprises bear R&D costs and failure risks alone, while efficiency dividends are shared across society. This “privatised costs, socialised gains” dynamic discourages private capital investment. Stock‑market valuation bubbles act as a compensatory mechanism, mobilising social capital via valuation premiums at the cost of losses borne by some investors, in exchange for society‑wide technological leaps. This aligns with Perez’s thesis in Technological Revolutions and Financial Capital: financial manias arrive first, and only after bubble cleansing does the golden age of technological popularisation commence.

Omnieyes follows a mirrored distribution logic. Foundation‑model ecosystems exhibit weak network effects, whereas the Trust‑Archive ecosystem generates strong network effects. Once a fully‑matured, multi‑party cross‑attested archive binding entities and time‑sequenced facts becomes an industry standard, all participants depend on it and switching costs become very high. Nevertheless, its positive externalities are even more pronounced, and commercial pricing proves harder to realise. Markets are willing to pay for “generative capability”, but reluctant to pay for “avoiding falsehood”. Its value mostly manifests as loss reduction rather than incremental revenue growth.

To partially internalise spill‑over value, trust‑based pricing is required via a dual‑track commercial‑plus‑public‑welfare model. Public‑facing fact archives operate as public goods delivering infrastructure‑grade functions. Enterprise‑, brand‑ and public‑opinion‑oriented archives adopt commercial models including private deployment, fully‑managed TrustOps and OEM licensing, together with consumer‑side billing by atomic facts (pricing tiers: 35 / 100 / 300 / 1000). Trust is broken down into the smallest tradable units. Commercial revenues fund editorial review and archiving for public‑fact repositories, enabling a self‑sustaining closed loop for “establishing credibility for AI”, instead of relying on bubble‑fuelled capital infusions.

Conclusion

• Input Ledger: Large‑model development represents a computing‑power arms‑race built on hardware depreciation and global financial cycles, prone to valuation bubbles. Trust‑asset infrastructure for factual attestation grows through long‑term accumulation of knowledge assets.
• Creation Ledger: Generative AI expands cognitive output; Omnieyes calibrates cognitive quality.
• Distribution Ledger: Generative AI naturally generates value spillovers, with bubbles temporarily compensating for market failures. The ultimate proposition for trust assets lies in implementing trust‑based pricing, turning intangible credible value into attested, tradable assets.

Bubbles may fuel break‑neck technological advancement. Yet lasting economic order ultimately rests upon trustworthy facts, traceable archives and attested trust.