San Francisco just crowned another AI heavyweight. Baseten raised a $150 million Series D at a $2.15 billion valuation, led by BOND with Jay Simons taking a board seat. Fresh money flowed in from CapitalG, Premji Invest, and Kevin and Elizabeth Weil of Scribble Ventures, while Conviction, 01 Advisors, IVP, Spark Capital, and Greylock Partners came back for more. Numbers grab headlines, but it’s the architecture underneath that makes this story worth reading twice.
Baseten isn’t building shiny AI toys. They are the backbone of the builders, the steel beams under the glass towers. Founded in 2019 by Tuhin Srivastava, Amir Haghighat, Philip Howes, and Pankaj Gupta, the company emerged from a shared frustration: getting machine learning models out of notebooks and into production without weeks of patch work engineering. Srivastava brought scars from co-founding Shape, Haghighat carried data platform chops from Clover Health, Howes supplied the mathematician’s rigor, and Gupta delivered systems muscle. Together they turned those scars into infrastructure.
The results speak louder than pitch decks. Revenue has multiplied more than tenfold in the past year. More than 100 customers already lean on Baseten. Abridge processes over a million clinical notes every week on their stack. OpenEvidence serves billions of AI model calls across hospitals. Patreon, Writer, Descript, Clay, and Zed keep workloads humming on the same rails. Customers report inference costs cut by 40 percent while performance climbs. That’s not efficiency jargon, it’s the math that keeps businesses alive when margins get tight.
The tech reads like a manual for weaponizing GPUs. Baseten provisions thousands of accelerators across ten cloud platforms in minutes, autoscaling down to zero when idle, with a 99.99 percent uptime guarantee. From NVIDIA T4s to Blackwell architecture, they squeeze silicon for every last cycle. Optimizations like FP8 quantization, speculative decoding, operator fusion, and topology-aware parallelism shave costs and latency. Developers get the Truss framework for packaging, Model APIs for instant access, Training for customization, and Chains SDK for stitching workflows into compound AI systems. This is inference stripped of friction.
The new capital scales more than servers. Headcount has tripled year over year, now topping 100 employees. The hiring machine is rolling, engineers, forward-deployed experts, and customer-facing talent to keep pace with demand. Investment pours into research, infrastructure for larger workloads, and industry-specific solutions in healthcare and finance. The bigger play? To become the index of the AI economy, much like Stripe became the index of the internet economy.
The takeaway is simple: in every boom, the real power belongs to those selling leverage. Baseten is packaging the modern picks and shovels, and the line outside their store keeps getting longer.
