Together AI Raises $1B at $7.5B Valuation to Expand AI Cloud Infrastructure
The AI boom has a certain rhythm to it right now. Data centers humming. GPUs running hot. Engineers pushing models to places that felt impossible two years ago. And somewhere inside that current, Together AI is building exactly what its name suggests. Infrastructure that brings the entire ecosystem together.
According to The Information, Together AI is reportedly raising $1B at a $7.5B valuation, a sharp move from the $3.3B valuation attached to the company after its $305M Series B in Feb 2025. In venture capital, numbers like that do not show up because a pitch deck looked pretty. They show up when the market starts to believe a company may be sitting directly in the path of where computing is headed next.
The company was founded in 2022 by Founder & CEO Vipul Ved Prakash, Founder & CTO Ce Zhang, Chris Ré, Tri Dao, and Percy Liang. This is not a founding team built around hype cycles. It is a group of researchers and engineers whose work has shaped modern AI systems from the inside. FlashAttention, distributed training systems, kernel-level optimization. The kind of technical foundation that makes investors lean forward before a single revenue slide appears.
Together AI calls its platform the AI Native Cloud. A full stack environment where developers can train, fine tune, and run models across 200+ open source models spanning chat, code, vision, audio, image, and embeddings. Infrastructure designed for the messy reality of AI workloads rather than the clean diagrams that show up in conference decks.
The adoption curve tells its own story. The platform now supports 1M+ developers and thousands of customers, with 10x year over year growth in annual contract revenue. Enterprise demand is showing up in real contracts as well, with 27 deals exceeding $1M and one deal pushing past $1B.
The technology layer is where this team really flexes. Tri Dao’s FlashAttention, now in its fourth generation, delivers up to 4x performance improvements at long sequence lengths. The Together Kernel Collection drives 24% faster training operations and up to 90% faster training performance on Blackwell infrastructure. Systems like ThunderAgent push agent workloads 3.6x faster, while ATLAS-2 accelerates inference by 1.5x.
Underneath it all sits serious compute. 200MW of secured power, large scale Nvidia GPU deployments, and infrastructure that can scale from 16 GPUs to more than 100,000. Partnerships like Hypertec Cloud, deploying 36,000+ Nvidia GB200 NVL72 GPUs, show how aggressively the company is building the physical layer of the AI economy.
Then there is the developer gravity. Together AI acquired CodeSandbox in Dec 2024, bringing a community of 4.5M+ monthly developers and native code interpretation directly into the platform. Infrastructure plus developer ecosystem is a powerful combination. And right now, Together AI is sitting directly where those two worlds meet.