Baseten as hidden compute layer

This kind of partnership turns Baseten from a direct sales infrastructure vendor into hidden plumbing inside another company’s workflow. In practice, a Benchling customer can run heavy scientific models through Benchling Inference, with GPU capacity supplied by Baseten under the hood, so the scientist stays inside Benchling while Baseten still captures usage, workload volume, and a path into biotech accounts that already trust Benchling for regulated R&D work.1 2 3 4 5

• Benchling launched Benchling Inference in May 2026 as a Baseten powered product for bursty scientific workloads, positioning it as scalable GPU capacity for training and inference without the customer managing infrastructure. Benchling said it had already been using Baseten internally for its own Model Hub before productizing that stack for customers.3
• The distribution advantage is Benchling’s installed base. Benchling says it powers more than 1,300 biotech organizations, and prior research estimated about 1,200 customers and roughly $210M ARR by mid 2024. That gives Baseten exposure to a large set of biopharma teams through one platform integration instead of winning each lab one by one.2 4 5
• The product fit is unusually concrete for biotech. Benchling describes scientists needing to run large prediction batches when lab data arrives in waves, then scaling back to zero when the work is done. That maps well to Baseten’s value in elastic inference capacity, especially for regulated teams that want the model inside an existing scientific system rather than a separate AI ops tool.3 1 2

More AI infrastructure companies will try to reach customers this way, by becoming the compute layer inside vertical software rather than asking every end user to buy a standalone inference platform. For Baseten, the next step is turning Benchling from a single partnership into a repeatable channel model across other software platforms where AI demand is large, bursty, and embedded in daily work.1 2 3