Together AI closed $800 million Series C at $8.3 billion valuation, backed by Aramco Ventures and others.
Together AI Inc., an operator of a cloud platform optimized for running open-source artificial intelligence models, has raised $800 million in Series C funding led by Aramco Ventures. Nvidia Corp., Vista Equity Partners, General Catalyst and several other institutional investors also contributed to the round. The funding values Together AI at $8.3 billion.
The company's platform includes a serverless inference service that enables developers to run open-source AI models without needing to configure graphics cards or network equipment. Together AI claims its serverless environments deliver approximately twice the performance of the fastest alternative offering.
Beyond serverless inference, Together AI provides three additional services. Two employ dedicated infrastructure for customers requiring greater reliability guarantees and customization options. A third offering, Batch Inference, prioritizes cost-efficiency over speed and delivers up to 50% price reduction for models that do not require real-time responsiveness to user prompts.
The platform runs on Nvidia chips and a proprietary software engine called ATLAS. At its core is a machine learning technique called speculative decoding, which accelerates customer workloads by integrating an AI model with a second, lightweight neural network. When a user enters a prompt, the lightweight algorithm rapidly generates a draft response that the main model then validates, corrects if necessary, and delivers to the user—a process significantly faster than generating output from the main model alone.
Typically, fixed-configuration lightweight algorithms lose accuracy over time. According to Together AI, its ATLAS technology addresses this limitation by automatically adapting the lightweight model to evolving user requirements. The company claims its software can accelerate certain inference workloads by up to 400%.
The platform also enables customers to fine-tune open-source models using training clusters containing thousands of graphics cards. Developers can manage these clusters via Kubernetes, which is relatively straightforward, or Slurm, which offers greater customization. Together AI's training clusters include software that automatically detects and remediates graphics card failures that could otherwise introduce errors into the training workflow.
The company disclosed that its annual bookings exceeded $1.15 billion in the second quarter. Its platform is used by several thousand organizations, including LG Inc.'s AI research lab, Cohere Inc., and the Mozilla Foundation.
Together AI plans to deploy the newly raised capital toward expanding infrastructure capacity. The company aims to grow its public cloud capacity by a factor of 50 over the next five years and intends to enhance both its training and inference features.