Open-Weight AI Models Gain Significant Market Share and Spending
Vercel data reveals a rapid shift toward open-weight models despite rising geopolitical regulatory uncertainties.
1Rapid Adoption of Open-Weight Models
Open-weight models have seen a dramatic surge in usage, processing over 70 percent of tokens via Vercel's AI Gateway by September. Their share of total spending quadrupled in just three months, rising from one in twenty dollars in June to nearly one in four by the end of the third quarter.
2Gateway Spending and Price Dynamics
While total gateway spending in September was 13 times higher than in December 2025, the average price per token dropped by 47 percent. This significant cost reduction, combined with nearly one in three teams switching their primary AI provider, indicates that intense market competition is effectively limiting traditional vendor lock-in strategies.
3Model Popularity Versus Revenue Impact
TypeSafe AI's Jev model achieved rapid adoption, reaching 35 percent of paid teams within just 14 days of its release. Despite this popularity, its unique architecture resulted in minimal financial impact, accounting for only 1 percent of total tokens processed and a mere 0.15 percent of the month’s total AI spending.
4Dominance of US-Based Proprietary Models
Anthropic and OpenAI continue to capture the largest portions of AI spending. Claude Opus 5.5 leads with 11.03 percent of expenditure, while GPT-6 Astra follows at 4.86 percent. These proprietary models remain the primary revenue drivers for the gateway, maintaining a strong foothold despite the growing popularity of alternative open-weight options.
5Rising Influence of Chinese Providers
Chinese AI providers are gaining traction, with DeepSeek V4.1 Flash ranking third in spending at 4.19 percent. Notably, over two-thirds of September spending on Chinese models went to US-based providers hosting them, a sharp increase from less than one-tenth in December 2025, highlighting the complex nature of current AI supply chains.
6Efficiency Gaps in Model Economics
Different models demonstrate vastly different token-to-spend ratios. DeepSeek captured 58 percent of tokens for only 6.6 percent of spending, while Moonshot AI’s Kimi captured 3.6 percent of tokens but 13.2 percent of spending. Z.ai’s GLM-5.3 model occupied 7.2 percent of tokens and 4.5 percent of total spending during the same period.
7Geopolitical Risks and Export Controls
China recently proposed adding 40 technology items, including 23 AI-related ones, to its export control list. Models like Moonshot’s Kimi K3 and Alibaba’s Qwen 3.8-Max meet parameter thresholds that could be affected. This creates significant uncertainty for US businesses that rely on these models, as licensing agreements could be curtailed at any time.
8Regulatory Outlook for US Businesses
Chris McGuire of the Council on Foreign Relations suggests the US must adopt a comprehensive regulatory framework for AI technology used overseas. Integrating Chinese models into domestic regulations is proposed to mitigate risks, as reliance on foreign-hosted models grows increasingly precarious amidst shifting trade barriers and potential US-led open-weight model bans.