The enterprise race to deploy generative artificial intelligence has long suffered from winner-take-all orthodoxy. Boardrooms and procurement teams routinely operate under the flawed assumption that whichever vendor tops the latest synthetic benchmark leaderboard will inevitably crush its rivals and monopolize the market. An empirical study thoroughly dismantles this benchmark fallacy, demonstrating that the enterprise assistant ecosystem has already fractured into distinct architectural lanes running on incompatible unit economics.
Researchers Daniel McCarthy (University of Maryland), Maxime C. Cohen and Eddy Hage-Youssef (McGill University), and D. Daniel Sokol (USC) evaluated worldwide mobile application usage data through December 2025. Published in California Management Review as "Three Winning AI Strategies," their investigation confirms that business positioning and workflow integration reliably trump raw parameter muscle when determining long-term commercial survival.
Three Distinct Economic Positions
Drawing on Sensor Tower data, the researchers identified three distinct, non-overlapping market archetypes: a volume-driven scale player, an integrated ecosystem moat, and a specialized premium niche. OpenAI's ChatGPT commands 78.4% of worldwide mobile daily active users, securing unmatched consumer ubiquity while shouldering the brutal infrastructure cost of converting low-margin volume into sustainable cash flow. Google Gemini captures a 12.5% share of daily active users while generating nominal direct mobile revenue, deliberately serving as a loss-leader to defend Google Search and drive enterprise uptake across Google Workspace.
Anthropic's Claude claims a modest 0.5% share of mobile daily active users yet extracts an estimated revenue per active user roughly three times higher than ChatGPT and over 40 times higher than Gemini. As Daniel McCarthy pointed out:
"ChatGPT is selling to everyone. Google is giving Gemini away to protect its search business. Anthropic is charging a small group of professionals real money. All three are working for now, but they're working for completely different reasons."
Each model architecture is deliberately optimized around different economic constraints. Gemini functions to lock enterprise perimeters inside existing cloud suites, ChatGPT operates as the default consumer front-end for broad experimentation, and Claude monetizes specialized coding, legal, and analysis workflows where buyers willingly pay for depth over reach.
The Absence of Cannibalization
The prevailing narrative that enterprise AI is a zero-sum cage match where every frontier release obliterates competing models is contradicted by empirical usage data. Across 15 major model launches tracked between May 2023 and December 2025, competitor cross-effects averaged an imperceptible -0.1%.
Frontier model launches expand total category demand rather than cannibalizing existing enterprise workloads. For C-level executives, this data renders benchmark-chasing obsolete. Selecting enterprise AI vendors based purely on aggregate benchmark rankings risks severe overpayment and integration friction; CIOs must instead map internal use cases directly to provider economics—allocating mass queries to volume leaders, contextual collaboration to platform suites, and deep reasoning to premium specialists.