Together AI

Together AI operates what it calls an "AI Native Cloud": a full-stack platform spanning inference, model shaping (fine-tuning) and accelerated compute for training and pre-training, built around open-source models. Product lines listed on its site include serverless inference, batch inference (scaling to 30 billion tokens per model), provisioned throughput with reserved capacity and a 99% uptime SLA, dedicated model inference, dedicated container inference for generative media (video, audio, image) workloads, accelerated compute clusters, code sandboxes for AI apps and agents, managed object storage and parallel filesystems with zero egress fees, and fine-tuning services.

Externally the company is characterized as an AI "neocloud" that rents Nvidia GPU clusters and other AI-specific infrastructure. At its February 2025 Series B the company described an AI Acceleration Cloud supporting more than 200 open-source models across chat, image, audio, vision, code and embeddings, powered by a proprietary inference engine and research techniques such as FlashAttention-3 kernels and quantization, and claimed 2-3x faster inference than hyperscaler alternatives. Infrastructure expansion at that time included 200 MW of secured power capacity, deployments of NVIDIA Blackwell GPUs across multiple North American data centers, and a partnership with Hypertec to co-build a cluster of 36,000 NVIDIA GB200 NVL72 GPUs.

The company maintains an in-house research organization, Together Research, publishing work on kernels (FlashAttention-4, Together Kernel Collection, ParallelKernelBench), architectures (Mamba-3, Parcae), inference optimization (speculative decoding, cache-aware prefill-decode disaggregation, consistency diffusion language models) and agents (ThunderAgent, CoderForge, DSGym), often co-authored with researchers at Princeton, Stanford, CMU and Berkeley.

Founding story

The company was founded in 2022. Co-founder and CEO Vipul Ved Prakash previously founded social media search platform Topsy, which he sold to Apple in 2013 for a reported $200+ million. His co-founders include Stanford professor Percy Liang and ETH Zürich / University of Chicago associate professor Ce Zhang; Chris Ré and Tri Dao (creator of FlashAttention, who serves as Chief Scientist) are also listed as founders.

Business model

Together AI sells cloud infrastructure and platform services for AI workloads, including on-demand and serverless inference, committed/provisioned throughput, dedicated deployments, GPU clusters, storage and fine-tuning. It is available directly and through the AWS Marketplace, and offers both self-serve sign-up and enterprise sales contact.

Usage- and capacity-based cloud pricing: token-based pricing for serverless and provisioned inference, committed reserved throughput, dedicated infrastructure deployments and GPU cluster rental, with no long-term commitment on serverless and zero egress fees on managed storage. TechCrunch reported annual bookings of over $1.15 billion as of the company's last quarter before July 2026.

Traction

Over 450,000 AI developers and companies were reported on the platform as of February 2025, with more than 200 open-source models supported. By July 2026 the company reported thousands of paying customers and annual bookings above $1.15 billion, and had raised roughly $1.2 billion across Series A, B and C rounds at an $8.3 billion valuation.

Latest developments

In July 2026 the company announced an $800 million Series C at an $8.3 billion valuation led by Aramco Ventures, with Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia, March Capital, Pegatron and SentinelOne's S Ventures participating. Recent site announcements include a partnership with Y Combinator to deliver a dedicated YC GPU cluster, on-demand B200 availability on Together GPU Clusters, and serving of MiniMax-M3. Research output includes FlashAttention-4, Mamba-3, Aurora, ThunderAgent and CoderForge-Preview, with a stated presence at ICML 2026.