Open-weight AI models quadruple gateway spend, now 1 in 4 dollars

Open-weight models have quadrupled their share of spending through Vercel's AI Gateway in just three months, now accounting for nearly one in four gateway dollars, according to data released by the company. In August, they processed the majority of AI Gateway token volume for the first time, and by September they handled 71.5% of all tokens. The money followed: from one in twenty gateway dollars in June to nearly one in four in September, a fourfold jump that signals a structural shift in how enterprises are buying AI compute.

Vercel's AI Gateway, like OpenRouter, acts as an intermediary that lets customers purchase tokens from different AI providers through a common interface. While it works with popular harnesses like Claude Code and Codex and provides access to both proprietary and open-weight models, it doesn't capture usage through other gateways or from customers who access AI models directly from vendors or via on-premises hosting. Still, the data offers a rare, real-world window into enterprise AI adoption, and the trend is unmistakable: open-weight models are winning on cost and flexibility, and the money is following.

The shift is being driven in large part by Chinese AI providers. DeepSeek V4.1 Flash ranked third in spending in September at 4.19%, while Moonshot AI, maker of Kimi, captured 3.6% of gateway tokens and 13.2% of spend. Z.ai, maker of GLM-5.3, took 7.2% of tokens and 4.5% of spend. Notably, more than two-thirds of September spending on Chinese models went to US providers serving those models, up from less than a tenth in December 2025. That means American cloud and hosting companies are increasingly the delivery mechanism for Chinese open-weight models, creating a complex web of dependencies.

The most popular model in September was TypeSafe AI's Jev, which reached 35% of paid teams using Vercel's AI Gateway in just 14 days. Yet its architecture is so efficient that it generated only 1% of the month's tokens and 0.15% of spending. This highlights a growing divergence between adoption and revenue, a dynamic that could reshape how model providers monetize their work. Meanwhile, US-based models from Anthropic and OpenAI still attract the largest share of spending, with Claude Opus 5.5 at 11.03% and GPT-6 Astra at 4.86%, but their dominance is eroding.

The regulatory backdrop is adding urgency. The current US administration has considered limits on open-weight models, and China on Friday proposed adding 40 technology items to its export control list, 23 of them related to AI. The Associated Press notes that Moonshot's Kimi K3 and Alibaba's Qwen 3.8-Max are among the models that meet the parameter threshold set by Chinese authorities, though the official message did not identify affected models. Chris McGuire, a senior fellow for China and emerging technologies at the Council on Foreign Relations, warned in a social media post: "This introduces significant uncertainty for US businesses that host or rely on Chinese models. Particularly given that Chinese AI labs are moving toward licensing the use of open-source models to generate revenue, the Chinese government stating that it has the ability to curtail those licensing agreements at any time should make US businesses more hesitant to rely on those models going forward."

For executives, the implications are immediate. Gateway spending in September was about 13x higher than in December 2025, but the average price per token fell 47%, indicating intense price competition as open-weight models commoditize inference. At the same time, almost one in three teams changed their most-used AI provider in September, suggesting that vendor lock-in is weakening. The combination of falling prices, shifting provider loyalty, and regulatory uncertainty means that AI procurement strategies built around a single vendor or a single model family are increasingly risky.

The strategic takeaway is clear: the AI supply chain is fragmenting, and open-weight models - especially those from Chinese labs - are becoming a central, if volatile, part of the mix. US businesses that host or rely on these models must weigh the cost advantages against the risk of sudden export controls or licensing changes. McGuire argues the US needs a comprehensive regulatory framework governing the use of US AI technology overseas and should include Chinese AI models used or hosted in the United States in its domestic AI regulations. Until then, the smartest move for CFOs and CTOs is to diversify providers, negotiate flexible contracts, and keep a close eye on both Washington and Beijing.