Together AI Raises $800M as Enterprises Ditch Closed Models

The enterprise AI market spent two years defaulting to closed frontier models — call the OpenAI or Anthropic API, pay the per-token toll, move on. On July 1, a funding round put a number on how quickly that default is breaking. Together AI raised an $800 million Series C at an $8.3 billion post-money valuation, more than doubling the San Francisco company's worth in roughly sixteen months. The lead investor is not a Silicon Valley venture shop but Aramco Ventures, the corporate arm of the Saudi oil giant — a tell about where the conviction, and the capital, for open-source infrastructure now sits.

The pitch: same intelligence, a fraction of the bill

Together AI's business is deceptively simple. It runs open-weight models — DeepSeek, MiniMax, Kimi, and the rest of the surging open catalog — as a managed cloud, giving companies a place to train and deploy AI workloads without paying closed-model prices. The claim it sells on is cost: Together says its customers cut inference costs by six to sixty times relative to running comparable work through closed APIs. Inference — the price of actually running a trained model in production, every query, every day — is where AI spending accumulates, and it is the line item enterprises are most desperate to compress.

That value proposition used to come with a quality tax. For much of the boom, open models trailed the closed frontier by enough that serious enterprise workloads stayed on OpenAI and Anthropic despite the cost. What changed is the open catalog itself. DeepSeek, the Chinese open labs, and a widening field of open-weight releases have closed the gap to the point where, for a large class of tasks, the open model is good enough — and once it is good enough, a 6-to-60x cost advantage stops being a trade-off and becomes an obvious choice.

The numbers behind the round

The growth Together is putting up explains why investors doubled the valuation. The company's annual bookings crossed $1.15 billion, and it says enterprise demand for open-source AI tripled over the period. That is not a speculative pre-revenue story; it is a business already booking north of a billion dollars a year, raising to fund the infrastructure that demand requires. The Series C brings Together's total funding to $1.3 billion, and the valuation arc is steep: the company last raised a $305 million Series B at a $3.3 billion valuation about sixteen months ago, so this round marks a jump from $3.3 billion to $8.3 billion in little more than a year.

The syndicate is as telling as the size. Alongside lead investor Aramco Ventures, the round drew Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia, March Capital, Pegatron, and SE Ventures. Nvidia's presence is the strategic thread — the chipmaker has a durable interest in a thriving market of alternatives to the closed labs, because every open-model inference cloud is another large, price-sensitive buyer of GPUs. A healthy open ecosystem sells more silicon, and Nvidia funds the companies that build it.

Why the money is coming from Riyadh

Aramco Ventures leading an open-source AI infrastructure round is not a coincidence of the cap table — it is the same logic driving Abu Dhabi's AI ambitions, applied one layer down the stack. The Gulf's strategic bet is on owning AI capacity and the infrastructure that runs it, and open models are the version of that bet you can actually control. A closed frontier API is a dependency on a US lab's pricing and access policy. An open-weight model running on infrastructure you fund is sovereign in a way a rented API never will be. For a state converting oil capital into technological leverage, backing the company that makes open models cheap to run is a strategically coherent move, not a financial afterthought.

The competitive squeeze it points at

Together's rise sharpens a question hanging over the entire closed-model business: how durable is pricing power when the open alternative is this cheap and this close in quality? OpenAI and Anthropic have built enormous revenue on the premise that their frontier models are worth a premium — and for the hardest reasoning tasks, they still are. But the enterprise workload distribution is not all hardest-case. A great deal of production AI is classification, extraction, summarization, and routing, where the marginal quality of the absolute frontier is invisible and the cost difference is not. Every one of those workloads is a candidate to migrate to an open model on a cloud like Together's.

That is the wedge, and $800 million is a bet that the wedge widens. If open models keep pace with the frontier — and the pace of open releases suggests they will — the closed labs increasingly compete on the shrinking set of tasks where their premium is defensible, while the broad middle of enterprise AI commoditizes onto open weights and cheap inference. Together AI has positioned itself as the pipe that middle flows through.

None of this dethrones the frontier labs, whose most demanding customers will keep paying for the best model available. But the market is bifurcating, and the open half is no longer the discount option for hobbyists. With a billion in bookings, a doubled valuation, and Gulf oil money underwriting the expansion, Together AI is betting that the future of enterprise AI is open by default and closed by exception — and enough capital just agreed to make it look like the safe bet.