Key facts

Founded
2019
HQ
San Francisco, United States
Employees
—
Website
Funding stage
Series F · US$1.5bn
Further facts
  • Capital: Baseten raised a US$1.5bn Series F announced on 22 June 2026, led by Altimeter Capital, Conviction and Spark Capital with Sands Capital, Wellington Management, IVP, Greylock and Battery Ventures participating; it closed in two tranches at valuations of US$11bn and US$13bn.
  • Funding pace: The Series F was its fourth round in about 16 months, after a US$75m Series C in February 2025, a US$150m Series D in September 2025 and a US$300m Series E in January 2026.
  • Origins: Baseten was founded in San Francisco in 2019 by Tuhin Srivastava, Amir Haghighat, Philip Howes and Pankaj Gupta.
  • Safety: On 9 October 2026 it announced Project Beacon with Goodfire AI, which reads model activations during generation to flag prompt injection, out-of-policy actions and data exposure inline.
  • Offices: Baseten Labs, Inc. is based in San Francisco and hires in New York, Toronto, Montreal, Seattle and London.

Focus areas

LLM inference GPU cloud Model APIs Reinforcement learning Voice AI On-premises deployment

About

Baseten is a San Francisco company, operated by Baseten Labs, Inc., that runs AI models in production for companies building AI products. Customers deploy open-source, fine-tuned or custom models on its Inference Stack, either as dedicated deployments or through Model APIs for open models such as Qwen, DeepSeek, GLM and gpt-oss, and can run them in Baseten's cloud, single-tenant, in their own VPC or in a hybrid set-up. It also offers training through its Loops SDK for reinforcement learning, embeddings, transcription and real-time audio, image generation and forward deployed engineers. Customers include Cursor, Notion, Lovable, Harvey, HubSpot, OpenEvidence and Abridge. In June 2026 Baseten raised a US$1.5bn Series F, its second tranche priced at a US$13bn valuation.

How they use AI

Baseten's product is the infrastructure that serves other companies' AI models. The Baseten Inference Stack combines model-level runtime optimisation with infrastructure that autoscales across GPU capacity, aiming at low latency and high throughput with a 99.99% uptime target. Teams package models and multi-model workflows with Chains, use BEI for high-throughput embeddings, and run speech, real-time audio and ComfyUI image pipelines. The Loops SDK supports reinforcement-learning training so customers can improve models and then serve them on the same platform. Forward deployed engineers help customers tune performance, and deployments can be single-tenant, self-hosted in a customer VPC or hybrid for data control.

Products and services

  • Baseten Inference Stack — Runtime and infrastructure for serving AI models at low latency and high throughput
  • Dedicated deployments — Autoscaling dedicated inference for open-source, fine-tuned and custom models
  • Model APIs — Hosted APIs for open models including Qwen, DeepSeek, GLM and gpt-oss
  • Loops SDK — Reinforcement-learning training for models served on Baseten
  • Chains and BEI — Multi-model workflow framework and high-throughput embeddings inference
  • Self-hosted and hybrid — Single-tenant, customer-VPC and hybrid deployment options

Notable customers / partners

Key people

  • Tuhin Srivastava — Co-Founder and Chief Executive Officer
  • Amir Haghighat — Co-Founder and Chief Technology Officer

Investors

  • Altimeter
  • Conviction
  • Spark Capital
  • Sands Capital
  • Wellington Management
  • IVP
  • Greylock
  • Battery Ventures