Open-weight AI models hit 62% of Vercel AI Gateway traffic
Vercel says open-weight AI models are rapidly taking over real production usage on its AI Gateway. On Aug. 22, CEO Guillermo Rauch reported that open-weight AI models accounted for 62% of all tokens processed by the Vercel AI Gateway in August, up from 28.4% on June 24 and 11% in April.
The AI Gateway is a routing and traffic-management layer for AI-powered apps, so the figures reflect enterprise workloads in production rather than benchmark or lab tests. Vercel links the shift mainly to cost: open-weight AI models can run at roughly one-tenth the price of closed-source models, so enterprises are using them more often when premium models aren’t required.
A key “spending paradox” remains. Despite 62% of token volume, open-weight AI models do not capture 62% of the revenue. The closed models from Anthropic reportedly take about 61%–65% of total gateway expenditure, meaning systems like Claude process fewer tokens but generate most of the money.
On model usage mix, DeepSeek has risen to the top (or near the top) of Vercel’s token-volume leaderboard and has overtaken Google in processing share. Large firms highlighted for cost-reduction approaches include AT&T and Coinbase, aligning with the workload-routing pattern Vercel is observing.
For traders, this is a signal of changing AI inference economics (open-weight vs closed models), but it is not a direct crypto market catalyst.
Neutral
This news is about AI infrastructure routing and pricing dynamics, not crypto protocol changes or token-specific adoption. The key takeaway is that open-weight AI models are winning on usage volume (62% of tokens) because they are far cheaper to run, while closed models still capture most of the spend (Anthropic reportedly 61%–65%). That is an industry cost-structure shift rather than a direct driver of blockchain demand.
In crypto terms, it could indirectly affect market sentiment around “AI compute” narratives (e.g., if traders extrapolate enterprise adoption of certain AI tooling). But the article doesn’t mention any crypto assets, on-chain activity, or a clear mechanism that would change token flows or network security.
Historically, when tech vendors release enterprise deployment metrics, crypto usually reacts only mildly unless there’s a direct linkage to token incentives, compute-trading products, or regulatory/on-chain milestones. Here, the linkage is mostly thematic (cost optimization), so the most likely market impact is limited and short-lived.