Open-weight artificial intelligence models handled 71.5 percent of tokens passing through Vercel's AI Gateway in September and captured nearly one in four dollars spent, according to new usage data published by Vercel this week. The figures reveal a dramatic shift in how businesses purchase AI services, with open-weight models more than quadrupling their portion of gateway spending between June and September. August marked the first time these models processed the majority of token volume flowing through the platform.
The data shows gateway spending in September reached roughly 13 times the level recorded in December 2025, even as the average cost per token fell approximately 47 percent over the same period. Open-weight models' share of expenditure climbed from one in twenty gateway dollars during June to nearly one in four by September. Almost one in three teams switched their most frequently used AI provider in September, suggesting competitive pressure may be preventing vendor lock-in. TypeSafe AI's Jev became the most popular model in September, reaching 35 percent of paid teams using the gateway in just 14 days, though its usage represented only 1 percent of the month's tokens and 0.15 percent of spending. US-based models from Anthropic and OpenAI still attracted the largest portions of spending, with Claude Opus 5.5 capturing 11.03 percent and GPT-6 Astra taking 4.86 percent, while Chinese provider DeepSeek V4.1 Flash ranked third at 4.19 percent.
According to Vercel, 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. Moonshot AI, which makes Kimi, captured 3.6 percent of gateway tokens and 13.2 percent of spend in September, while DeepSeek took 58 percent of tokens for just 6.6 percent of spend, and Z.ai, maker of GLM-5.3, took 7.2 percent of tokens and 4.5 percent of spend. The report notes that efforts to ban or limit open-weight models, something the current US administration has considered, would affect American businesses serving those models and their customers.
The data comes as trade barriers affecting AI models may reshape market dynamics. China on Friday proposed adding 40 technology items to its export control list, 23 of them related to AI, with the Associated Press noting that Moonshot's Kimi K3 and Alibaba's Qwen 3.8-Max are among the models meeting the parameter threshold set by Chinese authorities. Chris McGuire, a senior fellow for China and emerging technologies at the Council on Foreign Relations, said in a social media post that this "introduces significant uncertainty for US businesses that host or rely on Chinese models." He noted that because Chinese AI labs are moving toward licensing the use of open-source models to generate revenue, the Chinese government stating it can curtail those licensing agreements at any time should make US businesses more hesitant to rely on those models. McGuire said the US needs to adopt a comprehensive regulatory framework governing the use of US AI technology overseas and include Chinese AI models used or hosted in the United States in its domestic AI regulations. The growing reliance on open-weight and Chinese models through American intermediaries means any policy shift—whether export controls from Beijing or domestic restrictions from Washington—could disrupt the AI infrastructure that now processes the majority of gateway traffic. Companies racing toward cheaper inference may find themselves caught between competing regulatory regimes, with no clear path to compliance that doesn't sacrifice either cost efficiency or geopolitical risk.