Together AI, the San Francisco company that has turned running open-source models into a billion-dollar business, said on July 1 that it raised an $800 million Series C at an $8.3 billion post-money valuation — more than doubling the price tag investors put on it just sixteen months ago. The round was led by Aramco Ventures, the corporate venture arm of the Saudi oil giant, with participation from Vista Equity Partners, General Catalyst, Emergence Capital, NVIDIA, March Capital, Pegatron and SentinelOne's S Ventures.
The raise lands at a moment when the economics of building with AI have grown uncomfortable. Companies that wire their products to closed frontier models increasingly find that token pricing can swallow their entire gross margin. Together AI has positioned itself as the infrastructure layer for the alternative: helping firms train and serve workloads on open models such as DeepSeek, Nemotron, MiniMax and Kimi at a fraction of the cost of closed systems, and, the company claims, at comparable or better performance.
A billion-dollar business on open models
The numbers behind the round explain the investor enthusiasm. Together AI said its annual bookings crossed $1.15 billion last quarter, up sharply as open-source model usage across the industry tripled over the past twelve months, according to data from AI gateway OpenRouter. The company now counts thousands of paying customers, including some of the most closely watched names in AI — coding startup Cursor, autonomous-software firm Cognition and customer-service company Decagon among them.
Those customers report cost savings of six to sixty times versus closed-model pricing for equal or better output. Decagon, the company said, cut its inference costs sixfold after moving to the platform. The pitch is landing with the broader market too: nearly three-quarters of organizations expect to increase their use of open-source AI, according to McKinsey research cited by the company.
"Intelligence is becoming a foundational resource for the modern economy, every bit as essential as electricity, bandwidth or capital," said Vipul Ved Prakash, co-founder and CEO of Together AI. "Our mission is to ensure that intelligence is abundant, not expensive. The future of AI won't be owned by a few companies. It will be built by millions of developers and businesses, and open-source models are making that possible."
Building for a 50x footprint
Founded in 2022 by Prakash — who sold his previous startup, social-search company Topsy, to Apple in 2013 — along with Stanford professor Percy Liang and CTO Ce Zhang, Together AI has raised aggressively from the start. It closed a $102.5 million Series A led by Kleiner Perkins in 2023 and a $305 million Series B at a $3.3 billion valuation in early 2025. The Information reported in March that the company had been seeking as much as $1 billion at a $7.5 billion valuation; if accurate, Together took somewhat less money at a higher price.
The company says it will use the capital to expand its inference products and dramatically scale its infrastructure footprint, which it expects to grow roughly 50-fold over the next five years. Its edge, it argues, is software: a proprietary engine called ATLAS uses adaptive speculative decoding — a lightweight model drafts responses that a larger model verifies — to speed some inference workloads by up to 400 percent, squeezing more output from the same NVIDIA GPUs.
"Building AI infrastructure over the next decade will be the biggest infrastructure project in human history," said Abhishek Shukla, Managing Director of Prosperity7 Ventures US, part of Aramco Ventures. "Together has built the platform that makes open source models genuinely usable at enterprise scale, and the team's ambition matches the scale of the opportunity in front of them."
The neocloud land grab
Together AI is not alone in chasing the open-model wave, and the competitive field is thickening. Fireworks AI and Baseten are pursuing the same enterprise inference customers with similar promises of speed and cost efficiency, while NVIDIA — an investor here — increasingly sells managed inference directly through its own cloud partners, a reminder that Together's most important supplier is also a potential rival. Beneath them sits a broader neocloud boom: Upscale AI recently reached $500 million in total funding at a $2 billion valuation, and AMD-focused TensorWave raised a $350 million Series B at $1.55 billion, both within the past month.
What distinguishes Together's bet is the wager that the model layer, not just the metal, is where durable value accrues — that owning the research, the serving optimizations and the customer relationships will matter more than renting raw GPUs. The Aramco-led round, paired with commitments of well over 500 megawatts of compute to be capitalized separately, suggests deep-pocketed backers agree.
What to watch next
The open question is whether $1.15 billion in bookings converts into durable, high-margin revenue as GPU supply loosens and inference prices fall industry-wide. Watch for how quickly Together brings its promised half-gigawatt of capacity online, whether marquee customers like Cursor and Cognition deepen or diversify their spending, and whether NVIDIA's dual role as investor and competitor becomes a source of friction. If open-source models keep closing the quality gap with closed frontier systems, Together's timing could look prescient; if the frontier labs cut prices to defend share, the margin math that powers this valuation gets far harder.
"Intelligence is becoming a foundational resource for the modern economy, every bit as essential as electricity, bandwidth or capital. Our mission is to ensure that intelligence is abundant, not expensive."— Vipul Ved Prakash, Co-founder and CEO, Together AI