Together AI
The inference cloud quietly hitting $1B ARR by running everyone else's open-weight models.
together.ai ↗Together AI is the picks-and-shovels winner of the open-weight AI wave. While the model labs fight over capability leadership, Together captures revenue every time a developer or enterprise deploys Llama, Mistral, DeepSeek, or Qwen instead of a closed API — and that category is growing explosively. The $1B annualized revenue number is real, the valuation is climbing accordingly, and the Aramco sovereign-capital involvement signals the company is being positioned as strategic AI infrastructure rather than just a VC-backed SaaS bet. For investors, Together is one of the cleanest exposures to open-weight AI adoption without needing to pick which specific model lab wins.
What's going for them.
- 01Annualized revenue tripled in six months — from $300M (Sep 2025) to ~$1B (early 2026). The fastest-growing AI infrastructure company outside of the hyperscaler cloud providers.
- 02The default inference provider for open-weight models — Llama, Mistral, DeepSeek, Qwen, Mixtral all run on Together AI for a long tail of developers and enterprises that don't want to deploy their own infrastructure.
- 03Aramco's Prosperity7 Ventures participating in the current $1B round signals sovereign-capital commitment to Together as part of the strategic Middle East AI infrastructure buildout — positioning the company for MENA expansion.
- 04Founder lineup is research-credentialed: Percy Liang (Stanford, one of the most-cited NLP researchers alive), Chris Re (Stanford, MacArthur Fellow), Ce Zhang (ETH Zurich). The depth is unusual for an infrastructure-first company.
- 05Custom training, fine-tuning, and inference — Together is one of the few AI infrastructure companies that genuinely operates across the full lifecycle, not just inference. That breadth creates customer stickiness the pure-play inference providers (Fireworks, Replicate) don't have.
What they built
Together AI runs a GPU cloud and AI inference platform specifically designed for running open-weight models. The product stack spans the Together Inference API (hosted access to Llama, Mistral, DeepSeek, Qwen, Mixtral, and dozens of other open-weight models), Together Fine-tuning (managed training for customer-specific models), Together GPU Clusters (dedicated infrastructure for customers training their own models), and Together Code Interpreter. The customer base ranges from individual developers on pay-as-you-go to enterprise customers with multi-million-dollar annual inference spend.
How they got here
Vipul Ved Prakash (serial ML entrepreneur), Ce Zhang (ETH Zurich), Percy Liang (Stanford), and Chris Re (Stanford MacArthur Fellow) founded Together in 2022 with a thesis that open-weight models would become commercially important, and the infrastructure layer to run them efficiently would be a large, independent business. The first 18 months were quieter — building out the training and inference infrastructure, winning early developers with aggressive pricing. When Meta’s Llama 2 released in July 2023, Together became one of the default inference partners, and the revenue curve inflected.
Growth accelerated through 2024 and 2025. The $305M Series B in February 2025 at $3.3B was quickly outpaced by usage — ARR climbed from $130M (end-2024) to $300M (September 2025) to ~$1B annualized by early 2026. As of March 2026, Together is in talks for a $1B round at $7.5B valuation, with Aramco’s Prosperity7 Ventures considering a significant strategic investment. The sovereign-capital involvement positions Together not just as an AI cloud company but as part of the strategic infrastructure layer for MENA-region AI buildout.
What’s ahead
Three things matter for Together’s next 12 months. First, gross margins: inference providers compete on cost-per-token, and Together’s ability to maintain gross margins while GPU availability expands is the key financial question. Second, enterprise expansion: the self-serve developer and startup segment got Together to $1B ARR; the next leg is enterprise-contract revenue with Fortune 500 customers, who need different deployment guarantees, security certifications, and SLAs. Third, training and fine-tuning mix: inference is the biggest revenue line today, but customer demand is shifting toward managed fine-tuning and training on customer-specific data — a higher-margin, stickier business.
Why it matters
Together AI is the reference case for AI infrastructure companies that built on top of the open-weight model ecosystem — and the clearest evidence that open-weight AI has become a multi-billion-dollar commercial category independent of any single model lab. For founders building in AI infrastructure, Together’s trajectory demonstrates that the open-weight thesis is real revenue, not just research chatter. For investors, Together is the cleanest way to be long open-source AI without having to pick between Meta, Mistral, DeepSeek, or any other specific model provider.
Founder interview coming soon.
We'll be sitting down with the founders and operators of the companies we profile — on fundraising, product decisions, and what they're building next. If you're part of the Together AI team and want to share a perspective, get in touch.
Thinking about fundraising or M&A?
Amafi Advisory works with AI companies on strategic, fundraising, M&A, and technical advisory. Even if you're just exploring.