Together AI raises $800M at a reported $8.3B valuation as open-model inference demand compounds

Together AI has raised an $800 million Series C, and second sources put the round at a roughly $8.3 billion valuation, more than double the $3.3 billion it carried after its Series B in early 2025. The raise is one of the largest yet for a neocloud, the class of providers that rent Nvidia GPU clusters to serve AI models, and it appears aimed at a single bet: that demand for cheaper open-model inference keeps compounding faster than the frontier-token premium can hold.

What the round funds

The equity round was led by Aramco Ventures, NVIDIA, Vista Equity Partners, General Catalyst, and Emergence Capital, with participation from Schneider Electric, Pegatron, Salesforce Ventures, and others. Alongside the $800 million in equity, Together says more than 500 MW of compute capacity will be capitalized independently by its new investors. That split is worth noting: neocloud deals increasingly separate the equity story from the infrastructure financing, because the GPUs and the power to run them are a capital problem of their own.

The business metrics behind the round are substantial. TechCrunch reports annual bookings above $1.15 billion as of Together’s last quarter, and the company describes itself as one of the largest producers of AI tokens in the world. Its customer list now includes Cognition, Decagon, Eleven Labs, Cursor, and Suno. CEO Vipul Ved Prakash framed the thesis plainly: “Intelligence should be abundant, not expensive.”

For a team already serving Llama, Qwen, or DeepSeek in production, the practical read is capacity. More GPU financing means more headroom on the serverless and dedicated-endpoint tiers developers lean on, and a stronger balance sheet behind the provider they are trusting with production traffic. Capacity and reliability, not model quality, are the things a neocloud is actually selling.

Where this lands in the market

The round is a clean signal of where the money is flowing. Together cites cost advantages of 6x to 20x versus closed frontier models, with one customer, Decagon, reportedly cutting inference costs sixfold. Those numbers only matter if open models stay good enough to run the workload, and by TechCrunch’s account they increasingly do: usage of open-source models across the industry has roughly tripled in the past year, citing data from OpenRouter.

Together is not alone in raising against that trend. In June 2026, Upscale AI raised $500 million at a $2 billion valuation and TensorWave raised $350 million at a $1.55 billion valuation. Capital is pooling in the open-model serving layer specifically, which suggests investors expect the split between frontier models and cost-optimized open models to widen rather than collapse. The workhorse layer is being funded like a durable category, not a temporary arbitrage.

What’s worth watching

The obvious risk is that the bet rests on two things staying true at once: open models remaining close enough to frontier quality to carry real workloads, and neocloud pricing staying low enough to justify the switch. Either could soften. If a closed vendor cuts token prices aggressively, or if open-model quality plateaus while frontier models pull ahead, the cost argument narrows.

The 500 MW compute commitment is its own watch item. Capitalizing capacity is not the same as delivering it, and power, data-center buildout, and GPU supply have tripped up faster-moving infrastructure plays before. Over the next 90 days, the signals worth tracking are whether that capacity comes online on schedule, whether bookings growth holds as more neoclouds compete on price, and whether margins survive a market where three well-funded providers are all promising the cheapest open-model tokens. Stackmaven’s follow-up coverage will land on or around September 30.

Sources cited

1. Announcing our $800M Series C (Together AI) www.together.ai
2. Neocloud Together AI raises $800M, leaps to $8.3B valuation (TechCrunch) techcrunch.com