---
id: "concept-capability-competition"
type: "concept"
source_timestamps: ["§ The U.S. Strategy: Compete on Capability"]
tags: ["us-ai-strategy", "benchmarks", "feature-wars"]
related: ["concept-habit-moat", "claim-capability-depreciation", "contrarian-marginal-improvements-invisible"]
definition: "A strategy focused on winning market share through superior technical metrics, larger models, and advanced features, assuming that the best technology will naturally attract and retain users."
sources: ["attention"]
sourceVaultSlug: "hbr-seg-attention"
originDay: 4
articleStem: "hbr-tier2-07-chinese-ai-firms-habits"
sourceUrl: "https://hbr.org/2026/06/lessons-from-chinese-ai-firms-on-owning-customers-habits"
sourceTitle: "Lessons from Chinese AI Firms on Owning Customers’ Habits"
---
# Capability Competition

## Capability Competition

The dominant AI strategy in the U.S., which focuses on creating competitive advantages through **superior training data, larger models, better benchmarks, and more advanced features**. The underlying assumption: superior capability inevitably leads to user adoption, which leads to market dominance. This logic stems from previous platform wars (search engines, mobile OS) where the best product generally won.

However, in the current AI landscape, competing on capability produces a **hyper-competitive environment where leads are temporary**. Companies pour billions into marginal improvements (e.g., raising accuracy from **90% to 93%**) only to see rivals match them within weeks. This strategy treats the AI interface as a [[concept-destination-experience]] — a distinct place consumers must consciously decide to visit to research or transact.

The authors argue this is a flawed long-term strategy because advantages depreciate on a **six-week cycle**, turning R&D into a **holding cost** rather than a durable moat (see [[claim-capability-depreciation]]). The contrarian corollary is that once models cross a "good enough" threshold, further marginal gains become **invisible to ordinary consumers** ([[contrarian-marginal-improvements-invisible]]).

The strategic alternative the authors advocate is the [[concept-habit-moat]].

> Related quotes: [[quote-today-leader-tomorrow-scrambler]] and [[quote-capability-demo-habit-default]].

**Enrichment / external grounding:** External reporting confirms that U.S. labs ([[entity-openai-d7]], [[entity-anthropic-d7]], [[entity-google-d7]]) emphasize model improvements, benchmarks, and feature launches — but the specific "arms race" and "advantage-depreciation" framing is interpretive rather than directly documented. A counter-view holds that capability and habit are **complementary**: in high-stakes enterprise settings (legal, medical, financial), marginal reliability/safety gains are *not* invisible and are necessary before users will trust AI enough to routinize it.