Chinese AI Firms Build Habit Moats While US Rivals Chase Benchmarks

Two campaigns, two philosophies

During Chinese New Year 2026, Alibaba spent over $400 million subsidizing customer transactions—restaurant meals, movie tickets, flight bookings—on one condition: users had to complete them through its AI assistant Qwen. Two weeks later, OpenAI, Google, Meta, and Amazon each paid up to $10 million for 30-second Super Bowl ads promoting their AI capabilities.

The contrast reveals fundamentally different strategic approaches. US firms are locked in capability wars, racing to build superior models and better benchmarks. Chinese companies are pursuing what researchers call "habit moats"—embedding AI so deeply into daily routines that switching becomes psychologically costly, even when competitors offer marginally better technology.

According to research published in Harvard Business Review, the US capability-first strategy is showing cracks. OpenAI's enterprise market share fell from roughly 50% in 2023 to 27% in 2025, while Anthropic climbed from 12% to 40%. Every benchmark advantage proves temporary, and social media analysis from Super Bowl week found negative sentiment toward AI ads outpaced positive reactions by more than 2.5 to 1.

Why it matters

The habit moat strategy addresses a problem capability competition cannot solve: when all major AI assistants perform competently, marginal improvements become invisible to ordinary users. The difference between 90% and 93% accuracy matters to researchers, not to consumers deciding which app to open. Companies that win user defaults through behavioral integration may build more defensible positions than those competing on technical superiority alone.

How habit moats work

Qwen launched in January 2026 as what observers called the world's first comprehensive AI agent super-app, integrating over 400 capabilities across Alibaba's ecosystem—shopping, payments, food delivery, travel, and navigation. The app gained 10 million downloads in its first week. By mid-May, 300 million users had completed AI-driven shopping experiences through Qwen.

The strategic insight: Qwen doesn't ask users to visit it for AI-specific tasks. It intercepts tasks users were already performing and provides a faster path to completion. Booking a movie collapses from seven or eight discrete steps into a single sentence: "Book two tickets for tonight's showing, seats in the middle."

This mirrors Tencent's 2014 playbook with WeChat red envelopes—a feature that trained hundreds of millions to link bank accounts and normalize mobile payments. Within three years, WeChat Pay captured 40% of China's mobile payment market. Tencent spent almost nothing on subsidies; it simply inserted itself as the path of least resistance.

The Western opportunity

Researchers Yuanyuan Gina Cui, Patrick van Esch, and Jan Kietzmann outline four moves for building habit moats:

Hunt for habit cues, not feature gaps. Starbucks' Deep Brew AI doesn't try to be the smartest—it tracks ordering patterns and pre-selects usual drinks, timing push notifications to morning commutes. By early 2026, Starbucks unveiled an AI companion that lets users describe moods rather than browse menus.

Subsidize behavior, not subscriptions. OpenAI's Instant Checkout feature, launched in September 2025 with free trial access, was abandoned in March 2026 after minimal adoption. Qwen subsidized actual purchases, reducing the cost of actions users would take anyway.

Build for ambient utility. GitHub Copilot succeeds because it occupies the exact place developers are already typing. Microsoft 365 Copilot, despite 450 million potential users, achieved only 3.3% paid penetration because it functions as a feature to invoke rather than a default path.

Compete on the second transaction. Most companies optimize first-transaction metrics—signups, conversions, installs. Habit moats are built on whether customers reach for you again, unprompted.

While Western consumer habits are fragmented across Amazon, Google, Apple, and vertical players, the researchers note this fragmentation represents opportunity. The first company to achieve cross-domain behavioral integration in the US could capture disproportionate value precisely because the problem is harder to solve.

These findings were first reported by Yuanyuan Gina Cui, Patrick van Esch, and Jan Kietzmann in Harvard Business Review.