Vercel usage report shows why AI adoption and AI spending are different measures

What the report measures, and what it does not

Vercel published its October AI Gateway Production Index on October 15, 2026, a monthly report built from anonymized aggregate traffic routed through its AI Gateway through September. The headline finding concerns Jev, a decision model from TypeSafe AI that Vercel reports reached about 35% of paid teams in its first two weeks, the fastest adoption of any model in the gateway's history. TechCrunch's Russell Brandom reported the same figures on October 16 and called the data a limited but real view of usage. The lesson for a business owner is not which model is popular. The report measures adoption, request volume and expenditure separately, and they point in different directions.

The report distinguishes three metrics that are easy to blur. Reach counts how many paid teams try a model. Token volume counts how much work a model processes. Spend share estimates how much of the gateway's money a model takes, using published list prices rather than actual bills.

Jev shows the gap clearly. Vercel reports that it reached more than a third of paid teams and received the most requests of any model in September, yet accounted for a fraction of one percent of gateway spend over the same period. Vercel and TechCrunch attribute this to Jev's design for choosing among a few predetermined outputs, which they report lets it run cheaper and faster per request; broad trial therefore does not translate into a large bill.

The aggregate picture separates the same way. Open-weight models processed 71.5% of gateway tokens but took 23% of estimated spend, and that share has more than quadrupled since June. September's gateway spend was roughly 13 times December's while the average price per token fell 47%. Among paid teams with at least 10 million paid tokens in both compared months, nearly one in three changed their most-used lab in September, an index record; Vercel measures most-used lab by paid tokens. Usage is growing, prices are moving, and provider loyalty is not fixed.

All of these figures are vendor-reported observations inside one gateway. The sample excludes traffic that goes to other providers directly, and TechCrunch notes that this risks undercounting major labs that sell their own APIs. Spend shares rest on published list prices rather than actual bills. These sources do not establish an independent audit of the underlying traffic. The report describes what happened on Vercel's platform; it does not measure worldwide adoption, model quality, anyone's invoice, or what a competitor's customers do.

What this means for your business

If you are choosing AI for repetitive document or intake work, the report is a reminder that popularity is the wrong shortcut. The model that the most teams try is not necessarily the model that fits your task, and the model that processes the most requests may not be the one that costs the most. The right question is which measure matches your decision: how often the model is used, how much work it handles, and what each task actually costs you.

A concrete example Detailed In Design could build as custom work is an intake and document classification workflow. Inbound requests and attachments would be captured into your customer records, sorted into categories your team approves in advance, and routed through interchangeable model adapters so no single provider is locked in. Every item would carry a quality, latency and cost record per task, an ordinary rules baseline would run alongside the model for comparison, and anything the system is unsure about would land in a review queue for a person before it moves. The expected benefit is clearer model decisions, more consistent routing and fewer manual handoffs, evaluated against your own acceptance criteria. No numeric savings are promised.

That proposal is Detailed In Design custom software. It is not a Vercel product, not a packaged integration, and not a partnership with Vercel or TypeSafe AI. It depends on authorized access to your systems, a privacy and provider review, your approved routing categories, representative labeled examples, and agreed cost limits. Misrouting and fallback behavior would be tested before deployment, uncertain cases would go to a person, and the system would not price, pay or act toward customers on its own. For some tasks a small deterministic workflow may serve better than a model at all. Detailed In Design builds custom AI, software and workflow automation with human review gates, described at detailedindesign.com.

Sources

Charles, A., & Arora, H. (2026, October 15). Open weight takes 71.5% of tokens and 23% of spend. Vercel. https://vercel.com/blog/ai-gateway-production-index-october-2026

Brandom, R. (2026, October 16). Jev is Vercel's fastest-adopted AI model ever. TechCrunch. https://techcrunch.com/2026/10/16/jev-is-vercels-fastest-adopted-ai-model-ever/