The Economics of Builder Saturation in Digital Markets

Abstract: Recent advances in generative AI systems have dramatically reduced the cost of digital production, fueling narratives that widespread participation in software creation will yield a proliferation of viable companies. This paper challenges that assumption. We introduce the Builder Saturation Effect, formalizing a model in which production scales elastically but human attention remains finite. In markets with near-zero marginal costs and free entry, increases in the number of producers dilute average attention and returns per producer, even as total output expands. Extending the framework to incorporate quality heterogeneity and reinforcement dynamics, we show that equilibrium outcomes exhibit declining average payoffs and increasing concentration, consistent with power-law-like distributions. These results suggest that AI-enabled, democratised production is more likely to intensify competition and produce winner-take-most outcomes than to generate broadly distributed entrepreneurial success. Contribution type: This paper is primarily a work of synthesis and applied formalisation. The individual theoretical ingredients - attention scarcity, free-entry dilution, superstar effects, preferential attachment - are well established in their respective literatures. The contribution is to combine them into a unified framework and direct the resulting predictions at a specific contemporary claim about AI-enabled entrepreneurship.

• The paper formalizes the Builder Saturation Effect, showing that free entry and scarce attention drive average attention per builder to k/p while falling build costs expand entry rather than producer profits.
• The paper’s reinforcement model shows that stronger cumulative advantage sharply increases inequality, with simulations raising the top 1% attention share from 4.8% to 62.7% as reinforcement reaches its maximum.
• The paper calibrates its model to the iOS App Store, where sub-linear reinforcement reproduces observed concentration, including roughly 69% of downloads for the top 1% and nearly 23% of apps receiving fewer than 100 downloads.

Overview and motivation

"The Economics of Builder Saturation in Digital Markets" (2603.23685) addresses a specific contemporary claim: that AI-assisted software creation, by collapsing the cost of building digital products, will produce a proliferation of economically successful companies. The paper argues that this claim conflates an expansion of productive capacity with a proportional expansion of realized value, because it overlooks human attention as the binding scarce resource. The author formalizes this as the Builder Saturation Effect: in markets with near-zero marginal reproduction costs and free entry, increases in the number of producers dilute average attention and returns per producer even as total output expands, and under quality heterogeneity with reinforcement dynamics, outcomes become increasingly concentrated.

The paper is explicit about its epistemic status. It is presented as a work of synthesis and applied formalization rather than a source of new theoretical primitives. The individual ingredients—attention scarcity (Simon), information-goods economics (Shapiro and Varian), monopolistic competition (Spence; Dixit–Stiglitz), superstar effects (Rosen), preferential attachment (Simon 1955; Barabási–Albert), fitness-based condensation (Bianconi–Barabási), and network externalities (Katz–Shapiro)—are all well established. The contribution is to combine them into a single attention-constrained entry framework and direct its predictions at the "billions of companies" narrative associated with figures such as Sam Altman. The paper also distinguishes three tiers of claims: results proven within its own symmetric model, distributional results imported from network science, and interpretive implications about market structure, which it labels structured conjectures rather than proven conclusions.

Empirical motivation

The introduction assembles evidence consistent with elastic supply, inelastic attention, and concentrated outcomes. It cites the 2025 Stack Overflow survey (84% of developers using or planning to use AI coding tools), GitHub Octoverse data (46% of new code AI-generated), and Y Combinator's Winter 2025 batch (25% of startups with codebases at least 95% AI-generated). On the demand side, the Apple App Store hosts roughly 1.9 million apps, yet nearly a quarter have fewer than 100 downloads; the top 1% of monetizing U.S. publishers capture approximately 94% of revenue, and the top 1% of all publishers account for about 70% of downloads. Average per-user engagement—roughly 10 apps per day and 30 per month—has remained stable despite continuous supply growth. GitHub maintainers' reports of being overwhelmed by AI-generated contributions, likened to a "denial-of-service attack on human attention," are cited as qualitative corroboration. These figures are drawn from secondary industry sources rather than original empirical analysis, which bounds the strength of the motivating evidence.

The baseline model

The environment consists of $M$ consumers, each endowed with attention budget $a$, so aggregate attention is $A = Ma$; $B$ builders produce digital artifacts with fixed entry cost $k > 0$ and negligible marginal cost. Attention allocation follows a logit rule in which each product's share depends on its quality $q_i$ relative to all alternatives and to an outside option capturing inertia (non-adoption or incumbent usage), with effective outside-option weight $z = e^{\beta(q_0 - q)}$ under symmetry.

In the symmetric benchmark, average attention per builder is $\bar{s}(B) = A/(B+z)$, which is strictly decreasing and convex in $B$, with elasticity $-B/(B+z)$ approaching $a$0 when $a$1. Free entry drives profit $a$2 to zero, yielding equilibrium entry

$a$3

Two results carry most of the argumentative weight. First, the zero-profit identity: at equilibrium, average attention per builder equals $a$4, independent of total attention $a$5. Demand expansion is therefore fully absorbed by additional entry rather than improved producer outcomes. Second, the comparative statics imply that as AI tools drive $a$6, equilibrium entry grows without bound while equilibrium attention per builder vanishes and profits remain zero throughout. This directly contradicts the narrative that lower build costs translate into broadly distributed entrepreneurial success: cost reduction removes supply-side barriers without addressing demand-side constraints.

The welfare analysis adds a business-stealing result: because entrants do not internalize the dilution externality they impose on incumbents, free entry generically exceeds the social optimum $a$7, so saturation involves excess entry, not merely neutral redistribution.

Heterogeneity, reinforcement, and concentration

The symmetric model shows dilution but not concentration. To capture heavy-tailed outcomes, the paper extends the framework with heterogeneous quality $a$8 and reinforcement dynamics: a fraction $a$9 of attention is reallocated each period with probability proportional to $A = Ma$0, where $A = Ma$1 governs preferential attachment. This specification nests the static logit ($A = Ma$2), linear preferential attachment ($A = Ma$3, homogeneous quality, yielding $A = Ma$4), and the Bianconi–Barabási fitness model ($A = Ma$5, heterogeneous quality, yielding fitness-dependent power-law exponents and condensation phases).

The distributional propositions are explicitly imported rather than re-derived, and the paper flags this honestly. Within its own analysis, it proves that interior fixed points exist for $A = Ma$6 with shares proportional to $A = Ma$7, so effective quality sensitivity $A = Ma$8 rises with reinforcement; consequently the Gini coefficient is strictly increasing in $A = Ma$9 on $B$0, and the median-to-mean ratio $B$1 declines toward zero as $B$2. At $B$3 with heterogeneous qualities, no interior fixed point exists—a necessary condition for condensation—though the paper concedes that formally establishing condensation as the long-run outcome would require ruling out limit cycles and other non-stationary attractors, which it does not attempt.

Simulations corroborate these predictions. In the illustrative setting ($B$4, $B$5), moving from $B$6 to $B$7 raises the top 1% attention share from 4.8% to 62.7%, the Gini coefficient from 0.31 to 0.87, and drives the median/mean ratio from 0.78 to 0.04. The pattern is reported as robust across alternative quality distributions, reallocation rates, builder counts, and horizons.

Calibration to the iOS App Store

To assess quantitative plausibility, the paper calibrates the simulation to the U.S. iOS App Store using 2025 figures: $B$8 active publishers, $B$9 annual downloads as an attention proxy, widened quality dispersion ($k > 0$0), and substantial outside-option weight ($k > 0$1). With reinforcement $k > 0$2, the model reproduces the key empirical regularities: top 1% download share of 68.7% versus roughly 70% observed, Gini coefficient of 0.91 against an empirical value above 0.90, and 22.8% of apps below 100 downloads against roughly 25% empirically.

Three interpretive points follow. First, quality heterogeneity alone is insufficient—at $k > 0$3 the top 1% captures only 9.2% of downloads and no apps fall below 100 downloads—so observed concentration requires reinforcement mechanisms such as network effects, recommendation algorithms, and brand entrenchment. Second, the best-fitting reinforcement is sub-linear ($k > 0$4), below the $k > 0$5 condensation threshold, suggesting real markets exhibit strong but not maximal reinforcement, leaving room for multiple winners alongside extreme inequality. Third, the calibration treats annual downloads as a proxy for attention and uses stylized dynamics; the match is order-of-magnitude rather than exact, as the paper itself acknowledges given the model's deliberate simplicity.

Implications for the "billions of companies" claim

The discussion section draws a distinction between nominal artifacts, which may grow without bound as $k > 0$6 falls, and economically viable firms, which remain constrained by finite attention. Competition reorients from production toward discovery, retention, trust, and distribution; switching costs and incumbent inertia (the outside option $k > 0$7) become first-order determinants of outcomes. Strategically, the model implies that relative differentiation, early traction (which compounds under reinforcement), avoidance of highly substitutable categories, and complementarity-oriented positioning matter more than the act of building itself.

The strongest analytical move in the paper is its response to the objection that AI-mediated discovery could expand aggregate attention. Because equilibrium attention per builder is pinned at $k > 0$8 regardless of $k > 0$9, any expansion of attention induces offsetting entry, and per-builder attention declines over time whenever entry costs fall:

$q_i$0

which is negative precisely under the premise of AI-assisted building. Attention augmentation therefore changes market scale but not per-builder outcomes. The paper notes, more speculatively, that autonomous AI consumers could grow $q_i$1 itself, but shows even this would be absorbed by entry, while raising distinct questions about agent oligopsony and algorithmic herding that lie outside the framework.

Limitations and open questions

The paper concedes several limitations at the points where they bear on results. Aggregate attention is treated as exogenous and fixed in the baseline; behavioral or technological expansion of attention is acknowledged but not modeled beyond the dynamic extension. Complementarities that allow entrants to expand rather than divide demand are abstracted away, so the dilution result may overstate saturation in strongly complementary ecosystems. The reinforcement process is introduced in reduced form, and the long-run behavior of the dynamics at the $q_i$2 boundary—including whether condensation actually obtains rather than limit cycles—is left unproven. The calibration relies on industry-sourced statistics and a downloads-as-attention proxy. Interpretive claims about entrepreneurship and the "billions of companies" narrative are self-described conjectures resting on institutional assumptions the model does not capture. Open questions include endogenous, agent-mediated attention; formal derivation of limiting distributions under reinforcement; empirical testing on platforms where entry is cheap and attention measurable; and welfare analysis incorporating search costs, consumer surplus, and platform design.

Conclusion

The paper formalizes a structural tension underlying current narratives about AI-enabled entrepreneurship: production scales elastically while attention does not. Its proven core results—monotone attention dilution, the zero-profit identity $q_i$3 independent of aggregate attention, excess entry relative to the social optimum, and monotone concentration in reinforcement strength—are simple but consequential, and the calibrated simulation shows they are quantitatively consistent with the largest existing digital marketplace. The central implication is that falling build costs intensify competition for a fixed attention pool rather than broadening realized success, with winner-take-most outcomes emerging from the interaction of heterogeneity and cumulative advantage. The framework's value lies less in novel theory than in making explicit why democratised production decouples participation from economic viability.