⚠ Open Models & Fast FollowersModerate threat

The model advantage assumes Google's AI is meaningfully better than what others can build, and the danger is that this edge has largely evaporated. The frontier is now crowded with well-funded rivals whose models match or beat Google's, and open models are improving so fast that capable AI is becoming something anyone can obtain rather than a proprietary advantage. Meta released its frontier-class Llama family as open weights¹. The measured gap collapsed within a year: on the Arena leaderboard the best closed model outscored the best open one by 8.04% in January 2024 and by only 1.70% in February 2025². The open models closest to the frontier now come from Moonshot, Z.ai and Alibaba, and in September 2026 Moonshot's Kimi K3 Max trailed the leader, Anthropic's Claude Opus 5 Max, by 29 rating points³. When the intelligence itself is widely available, being good at models stops being a moat.

This is dangerous because the model was supposed to be where Google's data scale turned into an un-catchable quality lead. If rivals reach parity by training on the same public web, and open models put strong AI in everyone's hands, then the flywheel's payoff at the intelligence layer shrinks, and the differentiation must come from somewhere else — data freshness, distribution, integration — rather than from simply having the smartest model.

The comfort is that models are only one ingredient: turning a good model into a great product still requires the data, the distribution, the infrastructure, and the integration that Google has in abundance and a model-maker alone does not. And Google remains at or near the frontier itself, so it is not being left behind — it is merely no longer alone at the front, which is a different and lesser problem.

Moderate, all told. The commoditization of capable models genuinely erodes the intelligence-layer advantage that data scale was meant to produce, and it is the thread of the flywheel where rivals have most clearly drawn level — but models are not products, and Google's surrounding advantages in data, distribution, and scale still convert a merely-competitive model into a superior experience. The edge in raw intelligence narrows; the edge in everything around it endures.