Together AI raised $800M in a Series C at an $8.3B valuation to rent compute and serve open-source models at scale, with Aramco leading and NVIDIA joining: proof that running open AI is now a profitable business decision, not a plan B.
For years, saying "let's go with an open-source model" in a meeting sounded like the excuse of a startup with no budget: the unspoken assumption was that the truly good stuff lived behind a closed API and that open was the stopgap until the money showed up. That narrative just took an $800 million hit. Together AI —a neocloud that rents out NVIDIA GPU clusters and serves open models at scale— announced a Series C round of that size that more than doubled its valuation to $8.3 billion. And the one leading the round isn't a crypto fund chasing hype: it's Aramco Ventures, the investment arm of the largest oil company on the planet, joined by Vista Equity Partners, General Catalyst, NVIDIA, March Capital and Pegatron. When that kind of money moves into open-source inference infrastructure, we're no longer talking about ideology. We're talking about business.
What exactly Together AI sells (and why anyone cares)
Together AI doesn't build the flagship model everyone talks about; it does something less glamorous and far more profitable: it rents out compute power —NVIDIA GPU clusters— and on top of it serves open models like DeepSeek, MiniMax or Kimi, fast and cheap, to companies that need to train and run AI at scale without paying the toll of closed models. It's the difference between buying the car and renting the highway. And the highway is filling up with traffic: its annual bookings crossed $1.15 billion at the close of the last quarter, and industry usage of open-source models tripled in barely twelve months. Tripling in a year isn't a passing fad, it's a structural shift in where demand is headed.
Why Aramco and NVIDIA sign the same check
The investor list tells a story all by itself. NVIDIA invests because every open model run at scale means more GPUs sold: it's in their interest for the open ecosystem to flourish. Aramco and Vista invest because they see an infrastructure market with real margins and triple-digit growth, the kind of boring-but-huge business that serious capital funds love. And behind all of them sits the same cold calculation their customers make —Cursor, Cognition, Decagon, names already operating at scale—: when your product makes millions of model calls a day, the cost difference between a premium closed model and a well-served open one stops being an accounting detail and becomes the line that decides whether your business is profitable or not. The round doesn't validate any one specific model; it validates that there's an entire lane of companies that simply don't want, and don't need, to pay the price of closed.
The "plan B" that turned out to be the business plan
For a long time open-source carried a second-class stigma: powerful for tinkering, risky for production. What these $800 million certify is that the quality gap closed enough for the conversation to shift from "is it good enough?" to "why would I pay ten times more for something equivalent?". The result is a market that's no longer binary. It's not closed versus open as if it were a holy war; it's an engineering and finance decision made task by task. For a sensitive case you might pick the most capable closed model; for the routine 80% of the volume, a cheaply served open model saves you a fortune without the user noticing the difference. Market maturity is measured precisely by that: by being able to choose without friction.
What it means for those building with AI
The lesson is easy to state and hard to apply: don't marry a single model provider, because the one that's unbeatable on quality today can be unsustainable on cost tomorrow, and vice versa. Anyone who built their product locked to a single closed API discovers, when they scale, that their margin depends on pricing decisions they don't control. That's why the design that ages well is the one that treats the model as an interchangeable part, not a foundation. At NeuralOS we start from exactly there: we're model-agnostic by conviction, not as a slogan. You build your apps, your automations and your agents on the platform, and the model underneath is a decision you can revisit whenever the market —or your invoice— calls for it. No smoke: the Together AI round isn't news about a faraway company, it's confirmation that betting on the freedom to choose is, on top of being sensible, the side the real money is moving toward.