Together AI — the “neocloud” that rents out Nvidia GPU clusters and runs open-weight models for other companies — has raised an $800 million Series C at an $8.3 billion valuation, the company announced. The lead investor is not a Silicon Valley megafund. It is Aramco Ventures, the venture arm of Saudi Arabia’s state oil giant, with Nvidia, Vista Equity Partners, General Catalyst and Salesforce Ventures among those alongside it. The price tag is more than double the $3.3 billion valuation Together carried at its $305 million Series B roughly 16 months ago.
What Together sells is the plumbing for the open side of AI. Instead of renting a closed API from OpenAI or Anthropic, customers use Together to train, fine-tune and run models they can download and self-host — DeepSeek, Nvidia’s Nemotron, MiniMax, Moonshot’s Kimi. The company reports annual bookings above $1.15 billion and names Cursor, Cognition and Decagon among thousands of paying customers. The pitch is blunt: comparable performance to the closed labs at a fraction of the cost.
The timing is not an accident. CNBC reported Tuesday that US companies are increasingly routing work to cheaper open models — many from Chinese labs — as OpenAI and Anthropic prices climb, sending “good enough” tasks to the cheapest capable model rather than the best one. That is precisely the demand Together monetizes. And it cuts against the closed camp, which spent this same stretch locking up scarce compute and power: Anthropic just signed a 20-year, $19 billion power lease, and OpenAI gated GPT-5.6 to the government before anyone else.
Our take: The headline is the valuation; the tell is the lead investor. When a sovereign oil giant’s venture arm leads a round in an open-model cloud — not a frontier lab — it is betting the money in AI’s next phase sits in running models cheaply at scale, not in training the single biggest one. Open weights turn the model itself into a commodity, and margin drifts to whoever serves it best. Together is selling picks and shovels to everyone who would rather not pay closed-API rates — including the labs building their own inference chips to escape the same bill. The catch: neocloud is a capital-heavy, thin-margin business squeezed between the hyperscalers above it and GPU pricing below it. A billion in bookings is not the same as a billion in durable margin.
What to watch
- Bookings vs. margin. $1.15 billion in bookings is a real top line, but the economics only work if Together converts it to durable gross margin against AWS, Azure and CoreWeave. That number, not the valuation, is the one to track.
- The Aramco signal. Gulf oil and sovereign money is flowing into AI infrastructure, and now into the open layer specifically. Watch which other open-side companies the same checkbooks back next — it maps where they think the returns are.
- Open staying “good enough.” The whole thesis rests on open models being close enough to justify the savings. If a closed lab reopens a wide quality gap, the cheap-and-open trade cools fast — and so does Together’s tailwind.