Pennings, A.J. (2026, Sep 19) Will “Ontology Velocity” Pick the AI Economy’s Winners? apennings.com https://apennings.com/artificial-intelligence/will-ontology-velocity-pick-the-ai-economys-winners/

Introduction

Antonio Linares, presents a multi-year investment blueprint centered on three primary holdings. In the YouTube video below he examines Spotify, Duolingo, and Hims & Hers (HIMS) from data derived from an analysis of their Q2 2026 earnings and performance.

In this post, I examine his core investment thesis that built upon a framework Linares termed “ontology velocity,” which seeks to identify which companies will capture the majority of the economic value generated in the expanding artificial intelligence economy.[1]

By looking past short-term market sentiment and previous fears that generative AI would disrupt software platforms, Linares argues that these specific businesses are constructing defensive, self-reinforcing systems that the market fundamentally misunderstands.

Central to this framework is the assertion that proprietary data alone is insufficient to build a lasting competitive moat. The real asset is a closed-loop operational system that continually observes user actions, executes targeted interventions, and measures real-world outcomes. This attribution loop produces continuous, high-volume labeled data, which is the only mechanism through which artificial intelligence models genuinely learn and improve. Without this verification mechanism, raw data remains static and incapable of generating lasting technological differentiation.

Simultaneously, the baseline cost of machine intelligence is undergoing rapid deflation due to intense competition among foundational open-weight models and expanding global compute capacity. As raw token generation commoditizes and trends toward marginal compute cost, the extra expense required to produce one more unit of computing output, such as an additional GPU-hour or a single AI token. Also important is the standard generic tokens lose their pricing power.

Consequently, economic surplus shifts entirely to the owners of the closed-loop infrastructure who can produce non-replicable tokens rooted in unique, proprietary interaction data that outside competitors cannot buy or replicate.

This dynamic gives rise to what he calls “ontology velocity,” defined as the rate at which an organization captures behavioral feedback, verifies real-world results, and updates its underlying models to deliver superior user outcomes.

Because market participants inevitably default to the single best and fastest-improving interface, even a fractional lead in speed compounds into a winner-take-all advantage. This rapid learning loop ultimately drives customer lifetime value, translating into long-term free cash flow per share growth that public markets systematically fail to price in advance.

Spotify serves as the prime example of this model at scale through its proprietary “taste ontology.”[2] Processing roughly 3.4 trillion behavioral events per day across hundreds of millions of monthly active users and paying subscribers, Spotify’s taste model personalizes media consumption with an accuracy no external competitor can match. This closed loop directly drove record gross margins and lifted active listening days within months of rolling out its latest AI-driven autoplay recommendation system, demonstrating how scale converts directly into unassailable user engagement.

Beyond user-facing recommendations, Spotify exemplifies how operational velocity is optimized internally through automated developer infrastructure. Tools like “Honk” allow software engineers to delegate bug fixes and feature builds to autonomous agents via mobile messaging, while “Chirp” dynamically routes inference workloads to the most cost-effective models in real time while tracking token expenditure per developer. By treating token usage as an optimized cost of goods sold (COGS), Spotify has kept headcount flat for years while doubling revenue per employee.

Duolingo applies the same foundational blueprint to language and cognitive learning, driven by what management terms its internal “green machine.” By running hundreds of small, continuous product experiments, Duolingo drove current user retention to all-time highs in Q2 2026, setting the stage for re-accelerating daily active user growth. Customer captivity is an important barrier to entry against competing firms.[2]

Concurrently, the platform aggressively pushed down its internal AI video call costs from $0.30 to $0.01 per session by shifting workloads onto open-weight models, demonstrating how dominant platforms exploit cheap intelligence to expand their operating margins.

In the telehealth and wellness sector, Hims & Hers represents the early-stage deployment of this closed-loop architecture applied to complex biological outcomes. By expanding its subscriber base toward three million individuals and capturing longitudinal health data across multiple treatment categories, Hims increases its average revenue per user as each new medical vertical layers on additional patient biomarkers. This proprietary infrastructure unites manufacturing, digital consultations, pharmacy fulfillment, and clinical feedback into a cohesive system that standard software vendors cannot duplicate.

The ultimate objective for Hims is transitioning from a descriptive healthcare platform into an autonomous, predictive system. Initial deployments of AI clinical coaches within complex categories like weight management have already tripled patient communication while reducing non-clinical workloads by 50% and curbing cohort cancellations. As longitudinal health data compounds over time, the system moves toward predicting individual treatment efficacy prior to administration, creating an enduring operational moat analogous to the consumer-centric flywheels of Amazon and Costco.

Linares concludes that public financial markets consistently err by evaluating these businesses as static legacy apps and valuing them on backward-looking metrics such as units sold or short-term gross margin fluctuations. More important are dynamic optimization engines. By recognizing that ontology velocity dictates long-term market leadership and structural margin expansion, long-term investors can identify mispriced platform assets before the broader market recognizes their compounding earnings power.

The prevailing market narrative that foundational models will effortlessly disintermediate consumer and vertical software relies on a flawed economic assumption. Which is that intelligence itself will remain an expensive, proprietary asset. As raw compute commoditizes and open-weight models compress token production costs toward their marginal compute floors, value begins to evacuate from the model layer and accrues to the operational systems orchestrating closed-loop attribution.

The enduring competitive moats of Spotify, Duolingo, and Hims & Hers are not built on static datasets or off-the-shelf APIs, but on continuous, real-world behavioral verification cycles that generate high-fidelity labeled data at machine speed. By evaluating these platforms as dynamic optimization engines rather than legacy application wrappers, investors can look past superficial multiple compressions and recognize the compounding free cash flow generated by organizations capable of maintaining superior ontology velocity.

References

Arthur, W. B. (1989). Competing Technologies, Increasing Returns, and Lock-In by Historical Events. The Economic Journal, 99(394), 116–131.
Duolingo, Inc. (2026). Form 10-Q for the Quarterly Period Ended June 30, 2026. US Securities and Exchange Commission.
Hims & Hers Health, Inc. (2026). Form 10-Q for the Quarterly Period Ended June 30, 2026 Accessed ended. US Securities and Exchange Commission.
Karp, A. (2024). Palantir AIP and the Principles of Operational Ontologies. Palantir Technologies.
Spotify Technology S.A. (2026). Form 6-K for the Second Quarter 2026 Earnings Report. U.S. Securities and Exchange Commission.
Varian, H. R. (2018). Artificial Intelligence, Economics, and Data. National Bureau of Economic Research (NBER Working Paper No. 24839).
von Ahn, L. (2026). Product Velocity and Current User Retention in Scaled EdTech Platforms. Duolingo Shareholder Letter.

Notes

[1] I often teach market structure in my economics classes. I have traditionallt emphasized digital/media economics. AI represents interesting challenges to media economics that have had near-zero marginal costs with substantial cost per compute, even if they are decreasing with scale.
[2] These are important moves towards competitive advantage.

© ALL RIGHTS RESERVED

Not to be considered financial advice. AI is often used, and results are thoroughly interrogated. Links are used for some citations. Views are my own and do not express the stances of my employers, past or present.

---

Anthony J. Pennings, PhD is a Professor at the Department of Technology and Society, State University of New York, Korea and a Research Professor for Stony Brook University. He teaches AI and broadband policy. From 2002-2012 he taught digital economics and information systems management at New York University. He also taught in the Digital Media MBA at St. Edwards University in Austin, Texas, where he lives when not in Korea.

Related

Category: artificial intelligence, digital media economics, digital media management, technologies of meaning, technology management
Tags: Duolingo > Hims & Hers (HIMS) > ontology velocity > potify > Spotify

Comments

Comments are closed.