Vercel says DeepSeek is taking a bigger slice of AI token traffic, while Anthropic remains the top destination for spending inside its AI Gateway. The company shared the observation in a post that pointed to a broader article about how production AI usage is shifting across model providers.

The update offers a snapshot of how developers are using large language models in real applications, rather than in benchmark tests or public model rankings. According to Vercel, its AI Gateway handles tens of trillions of tokens every month as traffic moves between apps and AI providers. That volume gives the company a view into which models are being used at scale and how that usage is changing over time.

The main takeaway from the latest readout is that DeepSeek is capturing more of the token volume flowing through the gateway. Token volume is a measure of how much text, or model work, is being processed. A higher share of tokens generally suggests a model is being used more frequently or for longer prompts and responses across customer applications.

At the same time, Anthropic continues to dominate spend. That suggests that even if another provider is moving more of the raw token traffic, Anthropic models are still generating the largest portion of customer outlays on Vercel’s platform. The split between token share and spend can happen when one provider is used for high-volume, lower-cost tasks and another is chosen for premium workloads or more expensive model calls.

Vercel’s comments highlight the difference between usage and revenue signals in the AI market. Token share can show where developers are routing work, while spend can reveal which models are generating the most economic value. Together, the two measures may point to a market that is becoming more competitive and more segmented by use case.

The company did not provide detailed figures in the social post itself, but framed the data as part of a monthly look at AI Gateway traffic. Vercel said the service routes production application traffic to AI labs, which means the numbers reflect real-world deployment patterns rather than casual experimentation.

DeepSeek, a Chinese AI model provider that has attracted attention for its performance and pricing, appears to be making inroads on usage. Anthropic, meanwhile, continues to hold a strong position on the spending side, underscoring the appeal of its models among customers willing to pay for them.

The report also shows how quickly the competitive landscape in AI can shift as developers test different models for different tasks. Cost, reliability, speed, and output quality can all influence which provider gets chosen for a given workflow. As those choices change, usage share and spend share may not move in lockstep.

For now, Vercel’s latest readout suggests DeepSeek is gaining ground in the race for volume, while Anthropic remains the leader where it counts most for revenue. That combination points to a market that is still fluid, with developers spreading workloads across multiple AI labs rather than settling on a single dominant provider.