Together AI Raises $800 Million at $8.3 Billion to Run Open-Source Models at Enterprise Scale

Together AI raised $800 million in a Series C. Aramco Ventures led the round, with Nvidia, General Catalyst, Emergence Capital, and Vista Equity Partners also participating. Annual bookings crossed $1.15 billion last quarter - all from a company that has never trained a single foundation model.

Together AI Does Not Build Models. It Runs Everyone Else's.

Founded in 2022, Together AI rents Nvidia GPU clusters to developers and enterprises who want to run open-source models like DeepSeek, Nemotron, MiniMax, and Kimi - without paying the per-token rates that OpenAI, Anthropic, and Google charge for closed systems. Customers report cost savings of 6x to 60x compared to closed-model pricing for equal or better performance. Decagon, one of its enterprise customers, cut inference costs sixfold after switching.

More than one million developers use Together AI. Paying customers include Cursor - now owned by SpaceX since its $60 billion acquisition in June - plus Cognition and Decagon. Developers building production AI apps use Together as their compute layer; the models they run on top of it change whenever a better open-source option ships.

Open-source model usage tripled in the 12 months leading up to this round, according to OpenRouter data cited in the press release. Together AI's bet - that enterprises would rather cut inference costs sixfold than stay loyal to any single vendor - appears to be correct at $1.15 billion in annual bookings.

$800 Million, 16 Months After a $305 Million Series B

Together AI's last round was a $305 million Series B at a $3.3 billion valuation about 16 months ago. Valuation jumped from $3.3 billion to $8.3 billion in that window. Aramco Ventures leading this round - rather than a traditional Silicon Valley VC - reflects how Together AI now looks from the outside: more like energy or compute infrastructure than an early-stage startup.

Together AI secured 500 MW of investor-backed compute commitments alongside the funding. CEO Vipul Ved Prakash framed the company's mission around access: intelligence as a resource like electricity or bandwidth, not a service locked behind a vendor relationship. The company expects its infrastructure footprint to grow 50-fold over five years.

Developers Now Have a Funded Alternative to OpenAI for Inference

McKinsey data cited by Together AI shows nearly three-quarters of organizations expect to increase open-source AI usage. That is the demand curve Together AI is positioning against. Together AI has not disclosed any plans for IPO or acquisition - at $8.3 billion and growing, it does not need to.

GitHub Copilot's metered billing shock last month showed how quickly closed-model inference costs can spiral once usage scales. Together AI's pitch to those developers is direct: run equivalent workloads on open models, on Together's infrastructure, for a fraction of the cost. Decagon's sixfold savings is the clearest data point in the press release - but the 6x to 60x range suggests the variance across workloads is wide.

McKinsey data cited by Together AI shows nearly three-quarters of organizations expect to increase open-source AI usage. That is the demand curve Together AI is positioning against. Together AI has not disclosed any plans for IPO or acquisition - at $8.3 billion and growing, it does not need to.