According to the latest Brex Benchmark, the software landscape is being reshaped by startups that provide the underlying infrastructure for artificial intelligence (AI) rather than the AI products themselves. Fourteen of the twenty‑five fastest‑growing vendors this summer sell AI compute resources, databases, or related services that enable developers to build AI agents and applications.
Infrastructure takes the lead
The number‑one spot belongs to Together AI, a company that rents access to open‑source AI models. Behind it are firms that supply the GPUs, caches, sandbox environments, and database engines that AI agents need to function. In previous Benchmark cycles, mobile‑app platforms and cloud services such as Amazon Web Services dominated the list; this year the focus has shifted to the “plumbing” of the AI economy.
How startups are cutting costs
Brex data shows that moving from proprietary “Big Lab” APIs to open‑source models can reduce compute expenses by up to 80 percent overnight. An Nvidia H100 GPU now rents for roughly $3.40 per hour, down from about $8 in 2023, and when workloads are batched efficiently the cost per million tokens drops to pennies. By contrast, closed‑source APIs can charge $5 per million input tokens and $25‑$30 per million output tokens.
Database innovation fuels AI agents
Database providers such as Supabase and Neon are seeing explosive adoption. Supabase’s usage grew seventeen‑fold between the 2022 and 2025 startup cohorts, with nearly one in ten 2025 entrants paying for the service in their first year. AI agents now create more than 80 % of new databases for Neon, and over 60 % of Supabase deployments are launched by AI tools. The new generation of cloud databases separates compute from durable storage, allowing instant provisioning and near‑zero idle costs—features that align perfectly with the rapid, high‑frequency demands of AI agents.
What the spending data reveals
Entry‑level bills for these services are modest. The median monthly spend for Supabase’s Pro plan is $25, while Railway’s median bill has halved even as its paying customer base grew eightfold. The revenue model is volume‑based: millions of tiny databases are created and billed like utilities, scaling with each AI‑generated app. Startups now purchase both compute and database resources within their first twelve months, often spending only tens of thousands of dollars in total.
Implications for the tech ecosystem
The trend suggests that the next wave of successful software companies will be those that build on top of this AI infrastructure rather than trying to compete directly in the crowded AI‑product market. By providing affordable, on‑demand compute and storage, these vendors enable smaller teams to launch sophisticated AI agents without the massive capital outlays previously required.
All analysis in the report is based on anonymized, aggregated credit‑card and bill‑pay data from Brex’s finance platform, ensuring privacy while offering a rare glimpse into the spending habits of emerging tech firms.
Original reporting: KTVZ (Central Oregon) — read the source article.
