Artificial Intelligence · Profile from AI Research Companion
Together AI
The San Francisco "AI-native cloud" founded in June 2022 by Vipul Ved Prakash (founder of Topsy, which Apple bought in 2013), Ce Zhang (then of ETH Zurich), Chris Ré and Percy Liang (both of Stanford), with FlashAttention's author Tri Dao as chief scientist from July 2023. It began with a plan for a decentralised cloud of volunteered GPUs and open research (RedPajama, 2023), then became one of the largest independent platforms for running, fine-tuning and training open-weight models (Llama, DeepSeek, Qwen, Kimi, gpt-oss) on Nvidia GPUs that it mostly rents or leases from partners. Funding rose from a $20m seed (May 2023) to a $305m Series B at ~$3.3bn (February 2025) and an $800m Series C at $8.3bn led by Aramco Ventures (July 2026). The company said annual bookings passed ~$1.15bn in Q2 2026 (an undefined company metric, not revenue); outside estimates put annualised revenue near ~$1bn in early 2026 with a blended gross margin around ~45% and most revenue from GPU-cluster rental rather than per-token inference (Sacra, estimates, not company figures). Its strategic question is the one every inference cloud faces: whether research-led software (kernels, speculative decoding, serving engines) can keep margins above those of a GPU reseller as open-model token prices fall and hyperscalers, neoclouds and rivals such as Fireworks and Baseten fight for the same workloads.
At a glance
- Founded
- June 2022, San Francisco, by Vipul Ved Prakash, Ce Zhang, Chris Ré and Percy Liang (TechCrunch, May 2023); Tri Dao joined as chief scientist in July 2023 and is described as a co-founder in some later accounts
- Headquarters
- San Francisco, California
- Ownership / structure
- Private, venture-backed. Investors include Lux Capital (seed lead), Kleiner Perkins (Series A lead), Salesforce Ventures (2024 lead), General Catalyst and Prosperity7 (Series B leads), Aramco Ventures (Series C lead), Nvidia, Vista Equity Partners, Coatue, Emergence Capital, NEA, March Capital, Pegatron and others (company announcements). Total equity raised ~$1.33bn by July 2026 (reported tally)
- Leadership (2026)
- Vipul Ved Prakash (co-founder, CEO); Ce Zhang (co-founder, CTO); Tri Dao (chief scientist; also a Princeton professor); Chris Ré and Percy Liang (co-founders, academic roles at Stanford); Charles Zedlewski (chief product officer, per directories, verify); Alon Gavrielov (VP infrastructure strategy, ex-Cloudflare, from February 2026, reported)
- Products
- Serverless inference API for 200+ open models; dedicated inference endpoints; batch API (~50% of real-time rates); fine-tuning and, from 2026, a reinforcement-learning API; GPU clusters (H100, H200, B200, B300; instant and reserved); Together Code Sandbox and code interpreter (from the CodeSandbox acquisition, December 2024); data-preparation tools from Refuel.ai (May 2025); research artefacts including FlashAttention-2/3/4, the Together Kernel Collection and ATLAS speculative decoding
- Scale
- >450,000 developers and >$100m annualised revenue (company, February 2025); "more than 1 million developers" (company website, 2026); ~400 trillion tokens a month served (company, mid-2026); >500 MW of compute capacity commitments, financed largely by investors off Together's balance sheet (company/reported, July 2026)
- Valuation
- ~$1.25bn (March 2024); ~$3.3bn (February 2025); $8.3bn post-money (July 2026) (reported)
- Signature moves (incl. failures)
- Founding and the abandoned decentralised cloud (2022–23); RedPajama and open research as the brand (2023); Series A, Nvidia backing and the partner-sourced GPU cloud (2023–24); Blackwell build-out with Hypertec and the European expansion (2024–25); CodeSandbox, Refuel and the move up the stack (2024–25); the Series B, DeepSeek-R1 and the open-model price war (2025); the $8.3bn Series C and the IBM contract (2026); the HUMAIN 250 MW Saudi partnership (2026)
8 strategic moves analysed
Each one gets a full strategy file in the companion: context, the decision, the financing, what happened, and the lesson.
2022
Founding and the Decentralised Cloud Pivot
Founding / pivot
2023
RedPajama and Open Research as the Brand
Launch (open dataset, models and research)
2023
Series A Nvidia Backing and the Partner GPU Cloud
Funding round / compute deal
2024
Blackwell Build Out Hypertec and Europe
Compute deal
2024
CodeSandbox Refuel and the Platform Layer
Acquisition
2025
Series B DeepSeek R1 and the Inference Price War
Funding round / pricing
2026
HUMAIN Saudi 250MW Partnership
Partnership / compute deal
2026
Series C at 8 3bn and the IBM Contract
Funding round / compute deal