Open-weight models redistribute rents across the AI value chain
A Banca d'Italia paper shows that open-weight artificial intelligence models redistribute market rents rather than destroy incumbent profits. Author Marco Taboga found that infrastructure bottlenecks and enterprise switching costs shield established providers from Chinese competition.
From chipmaker rout to AI lab repricing
The study compares financial market reactions to China's DeepSeek R1 in January 2025 and Kimi K3 in July 2026.
While broad indices fell similarly, cross-sectional impacts diverged sharply.
DeepSeek R1 advertised training compute costs of just $5.58 million and API rates an order of magnitude below competitors, causing NVIDIA shares to drop 17.0 percent and the SOX semiconductor index 9.1 percent on fears of collapsing hardware demand.
In contrast, Moonshot AI's 2.8-trillion-parameter Kimi K3 arrived at frontier pricing of $3 input and $15 output per million tokens.
Its release triggered equity losses among competing Chinese laboratories rather than chipmakers, with Zhipu AI falling 35.1 percent and MiniMax Group declining 17.1 percent.
Complementary moats echo open-source history
Drawing parallels to open-source software in the 1990s, the paper demonstrates that free model weights expand total industry demand rather than destroying revenue.
OpenRouter weekly token volume expanded from 0.53 trillion in January 2025 to 58 trillion in July 2026, while NVIDIA quarterly revenues climbed from $44 billion to $82 billion.
Incumbents maintain defensible margins through complementary assets that raw model weights cannot provide, including enterprise trust, application harnesses, proprietary data moats, and dedicated serving infrastructure.
Weights do not make a moat
Taboga delivers a needed reality check by proving that benchmark parity does not equal enterprise utility.
Yet treating incumbent moats as durable overlooks how quickly software barriers erode once serving bottlenecks ease.
Relying on customer switching costs provides fragile defense against massive price differentials.