At the edge of the developer ecosystem: Observations in & around Vercel Ship (NET, DOCN, SHOP, OKTA, CRM)

Research evidence excerpt

At the edge of the developer ecosystem: Observations in & around Vercel Ship (NET, DOCN, SHOP, OKTA, CRM)

Goldman Sachs Technology: Software

enterprise agents. In this evolution, Vercel aims to take outsized developer market

share with innovative and opinionated AI tooling.

2. Puts and takes to token spend: We had several conversations with engineers

observing or managing token spend. The key dynamic centers around mapping

specific tasks to specific levels of intelligence, and a multi-quarter process in which

enterprises will become more fine grained on which tools and models are used for

which tasks. This month, we heard that sovereign hosting of models, for example in

Google Vertex, can cost ~3x more than public cloud, and that it is hard to focus on

quality and cost at the same time. For apples-to-apples use cases at a point in time,

engineers believe they can realize meaningful cost downs with a few weeks of focus;

we heard this from both a large financial organization, which quantified the potential

for 80% cost downs; and an AI-native app company Legora, which is in the process

of realizing a ~20pp gross margin uplift from cost optimization work. Vercel is also

targeting token and compute efficiency with Fluid Compute, a serverless

architecture that reduces costs and cold starts by decoupling active CPU usage from

idle wait time. Rather than paying for provisioned capacity on a VPS, customers are

billed only for active execution (~$0.15 per CPU hour), driving 3–4x lower costs vs.

traditional always-on infrastructure, a dynamic that is increasingly important as

agents run longer-duration, stateful workloads. As is the case with infrastructure and

observability tools, some years are growth years and some are optimization years,