Together AI raises $800M Series C to scale open-source AI

Together AI announced on July 1, 2026 that it has closed an $800M Series C, one of the largest rounds ever for an AI infrastructure company betting on open models.

Who invested and where the money goes

Participants in the round include Aramco Ventures, NVIDIA, Vista Equity, General Catalyst, Emergence Capital, Schneider Electric, Pegatron, Salesforce Ventures, March Capital, DTCP Growth, Lux Capital, Geodesic, PSP Partners and others. On top of the equity, the new investors reserved more than 500 MW of compute capacity, separate from the main round, to support infrastructure growth in the coming years.

Key facts:

• Amount raised: $800M; closing date: July 1, 2026
• Lead investors: NVIDIA, Aramco Ventures, Vista Equity, General Catalyst
• Additionally: commitments for 500+ MW of compute capacity
• Platform customers: Cognition, Decagon, Eleven Labs, Cursor, Suno

Together AI was founded four years ago on the conviction that generative AI should be open and available to a broad community of developers, not concentrated in the hands of a few corporations. By mid-2026 the company had grown into a full AI platform with a customer base of thousands of organizations.

Why do closed models struggle under production load?

Together AI points to a basic economic contradiction: when AI moves from prototype to real product, inference costs with proprietary models often grow faster than budgets. Companies end up having to "ration intelligence" just as agents are writing code, handling customer requests, analyzing documents and automating entire workflows.

According to Together AI, open-source models such as DeepSeek, Nemotron, MiniMax, Kimi and GLM have closed the quality gap with proprietary alternatives, while still leaving developers free to fine-tune them for specific tasks. Platform customers save 6 to 20 times compared with closed LLMs. Decagon, for example, cut its inference costs sixfold after moving to Together AI.

"The economics of production AI are determined not only by models, but

by the whole stack: kernels, compilers, inference systems, hardware utilization and the software that ties it all together," said Vipul Ved Prakash, CEO of Together AI.

What the company is building

Together AI describes itself as a research company whose foundational work goes straight into production. Over recent quarters the team has shipped FlashAttention-4 for NVIDIA Blackwell, Together Megakernel and together.compile, kernel-level optimization tools for industrial workloads. It has also expanded its post-training APIs with support for tool calling, reasoning and multimodal models. According to management, Together AI has become one of the world's largest producers of AI tokens.

What this means

An $800M round with NVIDIA, Aramco Ventures and Schneider Electric signals that major industrial capital is betting on open-source infrastructure as the production standard for AI agents, and providers of proprietary models will feel that pressure.

Frequently asked questions

Who are Together AI's customers and how much do they save?

Customers include Cognition, Decagon, Eleven Labs, Cursor and Suno. Decagon cut inference costs sixfold after moving to the platform; on average, companies save 6 to 20 times compared with closed LLMs.

What do the 500 MW of capacity in the deal mean?

This is compute capacity that the new investors committed to provide separately from the equity capital, as a distinct tranche to support Together AI's expected infrastructure growth.