The Company

Founding (2022)

Together.AI was founded in 2022 by Vipul Ved Prakash, Ce Zhang, Chris Re, Percy Liang, and Tri Dao — a group whose mix of academic research credibility (Stanford, including FlashAttention's Tri Dao) and entrepreneurial experience set the company's strategic posture from day one.

The founding timing matters. 2022 was when proprietary LLMs were starting to dominate (GPT-4 era), and the team bet that the open-source ecosystem (Llama, later Mixtral, Qwen, DeepSeek) would matter enough commercially to support a platform business.

Founders

• Vipul Ved Prakash — CEO. Background in entrepreneurship and ML / data infrastructure (Topsy founder, sold to Apple). Drives the commercial strategy.
• Ce Zhang — CTO. ETH Zurich professor. Distributed systems and ML systems research.
• Chris Ré — Stanford professor; co-founder. Deep ML systems and weak-supervision research lineage.
• Tri Dao — Stanford / Princeton; co-founder. Best known as the author of FlashAttention — one of the most-used pieces of modern transformer infrastructure.
• Percy Liang — Stanford CS professor. NLP and ML researcher.

The academic credibility of the team is a competitive asset. Customers and partners take Together's technical claims more seriously because the founders are visible researchers.

Original thesis

The original thesis combined several beliefs:

1. Open-source models will be competitive with closed-source for many use cases.
2. Inference is a much bigger market than training in dollar terms over time.
3. Customers want a managed inference platform with per-token pricing, not raw GPU rental.
4. A platform with strong open-source-model curation, inference optimization, and customer support could capture meaningful share.
5. The platform should support training and fine-tuning alongside inference — full lifecycle.

Together has executed against each of these. The company's product is per-token inference on curated open-source models plus fine-tuning plus dedicated training infrastructure.

Research lineage

The Together team has continued shipping research, not just product:

• RedPajama: An open-source reproduction of the LLaMA training dataset. Major credibility-builder in 2023.
• StripedHyena: Architecture research on alternative attention mechanisms.
• Sequoia (speculative decoding): Inference acceleration work.
• Together Research: The internal research arm publishing papers regularly.
• FlashAttention continues to be improved and integrated.

This research output is unusual for a startup at Together's stage. Most companies focus on shipping product; Together ships both. The research builds credibility and recruits talent.

Funding

Together has raised substantial funding through 2023-2025:

• Seed and Series A rounds in 2022-2023.
• Series B in 2024 at a multi-billion valuation.
• Additional financing in 2025 supporting infrastructure investments.

Investors include strategic AI-focused funds and major institutional investors. The capital base supports both the platform engineering and the infrastructure investments to grow inference capacity.

Scale today

• Inference API serving meaningful production traffic from many customers.
• Dedicated training clusters available for customers who need them.
• GPU footprint via owned hardware and partnerships at substantial scale.
• Customer base spanning AI startups, enterprises, and platform companies.
• Revenue not officially disclosed; industry estimates suggest meaningful nine-figure annualized run rate.

Together is the largest pure-play managed-inference platform competing in the open-source-model space.

Takeaway

Together.AI is positioned differently from any other company in this guide. It's a software platform on top of GPU infrastructure rather than a GPU cloud per se. The research credibility, the open-source focus, and the per-token commercial model define its strategic posture. The next chapter unpacks the product surface in detail.