Aramco Ventures' $800 million lead check into Together AI is a blunt bet on one idea: companies want open AI models badly enough to pay for the infrastructure around them.
Together AI raised $800 million on July 1, and the round gives the San Francisco company an $8.3 billion post-money valuation, according to a Business Wire release and TechCrunch's report on the financing. That is a real number. Aramco Ventures led the Series C through Prosperity7 Ventures US, with Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia, March Capital, Pegatron and SentinelOne's S Ventures also joining the round. Total funding now stands at $1.3 billion.
The speed of the markup is the story. Together AI was valued at $3.3 billion after a $305 million Series B led by General Catalyst and Prosperity7 in February 2025, according to earlier funding reports. Seventeen months later, investors are paying more than twice that valuation for a company whose central pitch is not that it owns the next closed model, but that it can make open-weight models useful at scale.
That is the bet. Together AI runs training and inference infrastructure for models including DeepSeek, Nemotron, MiniMax and Kimi, giving companies a way to avoid building their entire AI stack around OpenAI or Anthropic. If you're buying AI for a business, that matters because price and control eventually beat brand awe. You don't want every application tied to one vendor's model, terms and margin structure.
TechCrunch called Together AI a neocloud in its July 1 report, and the company says annual bookings now exceed $1 billion. Usage of open models on its platform has tripled over the past year. That is not small. A lot of AI infrastructure startups talk about demand, but fewer can point to billion-dollar bookings while leaning on open-weight models rather than a proprietary model they built themselves.
The Real Bet Is Compute #
Aramco Ventures leading the round isn't a stray venture check from an oil giant looking fashionable. This is strategy, not fashion. Saudi Arabia has spent years trying to diversify its economy under Vision 2030, and AI infrastructure has become one of the kingdom's preferred routes into technology ownership. Prosperity7's Abhishek Shukla said in the Business Wire release that AI infrastructure over the next decade would be the biggest infrastructure project in human history, and he framed Together AI as a platform that makes open models usable for enterprises.
Strip away the grand language and the practical point remains. This money buys capacity. Together AI says it plans to scale its infrastructure roughly 50 times over the next five years. The bill will be large. Data centers, GPUs, networking gear and power contracts are where an $800 million round goes when the product is compute rather than a lightweight software dashboard.
Nvidia's participation is worth reading plainly. It is both a supplier and an investor. That is not new. Nvidia has repeatedly taken equity positions across the AI infrastructure market it helps supply, and here it benefits whether a customer uses a closed frontier model or an open-weight one running on Together AI's platform. The chipmaker does not need to win the model argument to win the compute argument.
Vipul Ved Prakash, Together AI's co-founder and chief executive, framed the round around intelligence becoming as essential to the economy as electricity, bandwidth or capital. Fine. The sharper part of his argument is about open ecosystems making technology cheaper and faster to adopt. That is a direct challenge to the closed-model labs, and frankly, it should be. If open models are good enough for a growing share of enterprise workloads, the premium pricing power of closed providers starts to look less permanent.
Open Models Now Have A Buyer #
None of this means OpenAI and Anthropic are suddenly weak. They still have distribution, brand trust and product polish that infrastructure companies can't copy overnight. But the market is no longer waiting for open models to become respectable. DeepSeek forced that conversation into the open, and companies like Together AI are turning it into a purchasing decision.
Price changes habits. If a company can get acceptable performance from DeepSeek, MiniMax or Kimi through a platform that handles inference, scaling and deployment, it has a reason to test the open route. You don't need every workload to use the most famous model in the market. Customer support, internal search, coding agents, document processing, you name it, plenty of enterprise AI work is judged by cost and latency - and increasingly by reliability - before philosophical loyalty to a model provider.
The risk is that infrastructure startups are expensive animals to feed. Together AI's 50-times capacity target sounds impressive, but it also puts the company in a capital race with hyperscalers, specialist clouds and the AI labs themselves. Bookings are not profit. A billion dollars of committed demand still has to survive hardware costs and power constraints - and the brutal reality that inference prices keep falling.
Still, Aramco Ventures did not lead an $800 million round because open-source AI makes for a nice slogan. It backed a company sitting at the point where open models and Nvidia hardware meet enterprise procurement. That is where the next fight is. Together AI now has the money to make open-weight models feel less like an experiment and more like infrastructure.
Also read: America's AI Boom Is Starving the Chip Factories It Depends On • Tencent's Hy3 Bets on Smaller Agent Models Instead of Bigger Ones • Bitcoin ETFs Bled $4.67 Billion in Q2 as AI Stocks Stole the Money