Baseten

Baseten was founded at the end of 2019 by Tuhin Srivastava, Amir Haghighat, Philip (Phil) Howes, and Pankaj Gupta, who had worked together previously and shared frustration with how difficult it was to get machine learning models out of notebooks and into production. The company's original product (launched in public beta in May 2021) was a general-purpose toolkit for data science and ML teams to deploy models, wire up backend infrastructure, and build UIs around predictions — an 'ML application builder' aimed at letting technically-savvy data scientists become full-stack shippers without needing to master Docker, Kubernetes, or frontend engineering. Early backing came from Greylock Partners (via then-partner Sarah Guo, who co-led the seed round and later led the Series A), South Park Commons, Lachy Groom, and Caffeinated Capital, alongside angel investors including Greg Brockman, Dylan Field, and Mustafa Suleyman.

As the generative-AI wave accelerated after ChatGPT's November 2022 launch, Baseten repositioned from a general ML tooling/app-builder platform toward a dedicated AI inference infrastructure company. The company built out its 'Inference Stack' — combining applied model-performance research (custom kernels, quantization, speculative decoding), globally distributed multi-cloud infrastructure (spanning 18+ cloud providers), and observability/model-management tooling — to serve open-source, custom, and fine-tuned models at production scale with 99.99% uptime targets. Its open-source Truss framework became a widely adopted standard for packaging and serving models. The company later expanded into post-training and reinforcement learning (Model Labs, Baseten Loops) and forward-deployed engineering, embedding its own engineers with customers to tune deployments for specific latency and cost targets.

Baseten's growth accelerated sharply through 2025–2026 as inference spending outpaced model training in the broader AI infrastructure market. The company raised a $40M Series B (March 2024, led by IVP and Spark), a $75M Series C (February 2025, co-led by IVP and Spark, joined by new investors Adam Bain and Dick Costolo via 01A), a $150M Series D (September 2025, led by BOND, with BOND partner Jay Simons joining the board), a $300M Series E (January 2026, at a $5B valuation, led by IVP and CapitalG with a reported $150M anchor from NVIDIA), and a $1.5B Series F (June 2026, at a $13B valuation — a split-priced round with some investors valuing the company at $11B — led by Altimeter Capital, Conviction, and Spark Capital, co-led by Sands Capital and Wellington Management). By its own account, Baseten's revenue grew roughly 20x and its inference volume grew roughly 40x in the twelve months leading into the Series F, following ~100x inference-volume growth in the prior year; Greylock's portfolio page states Baseten processes over one billion inference calls per day.

Business model

B2B SaaS / infrastructure-as-a-service; usage-based (compute and inference-volume) pricing combined with dedicated-capacity and enterprise contracts

Usage-based billing for GPU compute and inference calls, plus dedicated/reserved-capacity contracts, embedded forward-deployed engineering, and post-training/RL services for enterprise customers