 Substitution, Not Speculation: Why the Artificial Intelligence Investment Cycle Need Not Be a Bubble, and Why Its Robustness Accelerates the Commoditization of the BuildArthur de Miranda Neto ()
Working Papers from HAL Abstract: The 2025–2026 debate over an "AI bubble" has generated abundant heat and comparatively little classificatory clarity. This paper does not argue that valuations are modest, that froth is absent, or that no capital will be destroyed; it argues something more specific and more defensible, that the prevailing analogy to the dot-com episode of 1999–2001 misclassifies the object being repriced, and that once the error is corrected the cycle need not be read as a bubble at all. The dot-com episode was the speculative repricing of a medium: the internet, however transformative, transports, stores, and recombines value created elsewhere but does not itself perform productive work. The present cycle is the repricing of a production factor: a synthetic, scalable, and increasingly capable substitute for qualified human labor. The distinction is not rhetorical: a medium is valued on the speculative expectation that others will build productive activity upon it (a link long, contingent, and easily over-extrapolated, which is why media bubbles are common) whereas a factor is valued on the marginal product it delivers directly, and the link between an AI system and the cognitive task it performs (drafting, translating, coding, reviewing, summarizing) is immediate and measurable. Building on the author's Commoditization Stack framework and on the network-effects and platform architecture developed in the Data-Driven Platform of Platforms (DPoP) model, the paper advances four propositions. First, the concentration of venture capital in frontier laboratories is a rational repricing of a labor-substituting factor, not speculative mania. Second, the observable weakness in software-as-a-service (SaaS) valuations is not a symptom of an AI bubble but a confirmation of the commoditization gradient: layer-4 application software is being substituted by agentic AI precisely as the Stack predicts. Third, AI's dual-use character and its centrality to the strategic posture of developed states place it in a different risk class than purely commercial speculative assets. Fourth, the genuine pathologies of the cycle (circular vendor financing, the enterprise deployment gap, energy constraints, and the contested economic life of AI hardware) are best read as the friction of a mispriced-but-real factor market rather than as evidence of a content-free bubble. An illustrative, data-calibrated demonstration for the Brazilian services economy makes the mechanism concrete: as AI capability commoditizes, a measurable share of a national wage bill is converted into payments for an imported factor that migrate toward the bloc hosting the frontier laboratories (surfacing there as revenue and valuation) while a substantial premium is retained domestically. Because it models a single economy and only the labor-substitution channel, omitting the revenue AI earns as an embedded component of countless digital products, the demonstration is a conservative lower bound on a mechanism that operates across every importing bloc. Benchmarked against the actual Brazil–United States trade balance, that flow would, on the recent pattern, become a first-order driver of the bilateral imbalance rather than a marginal one. The paper closes with a falsifiable diagnostic that specifies the conditions under which the substitution thesis would fail, and is careful to separate two questions it does not conflate: whether the asset is real, which it answers in the affirmative, and whether current valuations are correct, on which it takes no position, a real production factor and a financial bubble being able to coexist. It situates the argument within the broader project of valuing intangible, knowledge-layer assets in the AI era. Its aim is analytical and constructive rather than advisory: it is written for entrepreneurs and intrapreneurs deciding what to build and where to defend it, not for investors deciding what to buy, and it supplies elements for build-and-innovation strategy (where value accrues, which of the stack's seven layers are defensible, and how differential commoditization reprices them) applying the framework's layered valuation method to the frontier firm itself, which it finds exposed to rather than insulated from commoditization. It offers no recommendation on any security. Its central twist is that this is not reassurance: because the frontier factor is robust and keeps advancing, the build accelerates, flattening the cost gradient of construction and speeding the commoditization of the layers on which most ventures are built, so the robustness of AI is itself the strategic problem the paper hands the entrepreneur. A network-effects reading converges with the same conclusion: the durable effects (data, platform, expertise, and protocol) accrue not to commoditizing capability access but to the defensible upper layers of the stack, which is also where future revenue compounds. Keywords: Artificial Intelligence; Asset Bubbles; Commoditization Stack; Labor Substitution; Network Effects; Venture Capital Concentration; Dual-Use Technology; SaaS; Platform of Platforms; Technology Valuation (search for similar items in EconPapers) Downloads: (external link) Related works: Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text Persistent link: https://EconPapers.repec.org/RePEc:hal:wpaper:hal-05693944 Access Statistics for this paper More papers in Working Papers from HAL |
Is your work missing from RePEc? Here is how to contribute.
Questions or problems? Check the EconPapers FAQ or send mail to .
EconPapers is hosted by the
School of Business at Örebro University.