Together AI Raises $800M to Make Open-Source AI Production-Ready

30-Second Summary

Together AI raised $800 million at an $8.3 billion valuation on 1 July 2026, led by Aramco Ventures with backing from Nvidia, Vista Equity, General Catalyst, and others. The company runs the leading platform for deploying open-source AI models at scale, reporting more than $1 billion in annual bookings and triple the open-source model usage compared to a year ago. The raise is the largest ever for an AI inference infrastructure company and arrives as open-source models close the capability gap with closed systems while costing dramatically less.

At a Glance

• Topic: AI Infrastructure
• Company: Together AI
• Date: 1 July 2026
• Announcement: $800 million Series C at $8.3 billion valuation
• What Changed: Together AI's valuation more than doubled from $3 billion in February 2025 to $8.3 billion, with $1 billion-plus in annual bookings confirming commercial traction
• Why It Matters: The round validates open-source AI inference as a credible enterprise infrastructure category, not just a cost-cutting workaround
• Who Should Care: Business operators, finance leaders, and heads of product managing AI spend and build decisions at companies with 10 to 200 employees

Key Facts

• Round size: $800 million
• Post-money valuation: $8.3 billion
• Lead investor: Aramco Ventures
• Co-investors: Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia, March Capital, Pegatron, SentinelOne S Ventures
• Annual bookings: more than $1 billion
• Open-source model usage growth: tripled on the platform over the past 12 months
• Infrastructure growth target: roughly 50-fold expansion over the next five years
• Previous round: $305 million Series B in February 2025 at approximately $3 billion valuation
• Cost advantage cited: open-source models on Together run 60 to 90 percent cheaper than comparable closed models

What Happened

Together AI, founded to make frontier AI accessible through open-source model inference, closed its Series C on 1 July 2026 with $800 million raised at an $8.3 billion valuation. The round was anchored by Aramco Ventures, the investment arm of Saudi Arabia's national oil company, alongside Nvidia, Vista Equity Partners, General Catalyst, Emergence Capital, March Capital, Pegatron, and SentinelOne's S Ventures.

The company provides infrastructure for running open-source AI models, including DeepSeek, Nvidia's Nemotron series, MiniMax, and Kimi, through a single API. Its customers include developers, startups, and enterprises that want the performance of frontier AI at dramatically lower cost than closed systems from OpenAI or Anthropic.

Together reported $1 billion-plus in annual bookings and noted that usage of open-source models on its platform has tripled over the past year. The company plans to use the new capital to expand product features, grow its commercial footprint, and scale its infrastructure roughly 50-fold over the next five years.

The round is the largest ever raised by an AI inference infrastructure company and comes as enterprise demand for open-source AI accelerates. More than 30 percent of US enterprise API tokens now flow through open-source or Chinese AI models, up from roughly 11 percent a year ago, driven largely by cost pressure from escalating closed-model prices.

Why It Matters

Open-source AI is now enterprise-grade infrastructure, not a workaround. When Aramco Ventures, Nvidia, and Vista Equity commit $800 million to an inference platform, they are making a structural bet, not a speculative one. Together AI is already generating $1 billion in annual bookings, which means enterprises are not experimenting with open-source models. They are running production workloads on them.

The cost gap is the defining business story. Open-source models available on Together's platform currently cost 60 to 90 percent less than comparable closed models for many business tasks. For companies running high volumes of AI-assisted work, that gap is not a minor optimisation. It is the difference between AI that scales affordably and AI that becomes a growing liability on the P&L.

Nvidia's involvement signals model quality parity. Nvidia is simultaneously supplying the chips that power closed-model providers and co-investing in the infrastructure designed to commoditise them. That position only makes sense if Nvidia believes open-source models will reach quality parity across enough use cases to sustain a distinct market segment. Their check is a capability vote.

Infrastructure investment unlocks competitive advantage for fast movers. Together plans to grow its capacity roughly 50-fold. Operators that lock in infrastructure relationships, build on standardised APIs, and develop model-switching capability now will be insulated from pricing shifts across any single provider, whether open or closed.

The raise accelerates a cost-pressure cycle for closed-model providers. As Together's infrastructure scales and open-source adoption grows, the cost gap will likely widen further, putting pressure on OpenAI and Anthropic to reduce prices or differentiate more sharply on capability. Operators on fixed-cost AI contracts signed in the past 12 months may find themselves overpaying sooner than expected.

The David and Goliath View

This raise is the structural confirmation of something that has been building for 18 months. Open-source AI is not catching up to closed models. For the majority of business tasks that operators actually run, it has already arrived. The cost numbers at 60 to 90 percent lower are not benchmark estimates. They are what companies are actually paying when they switch.

For a 10 to 200 person business, this is a real operational decision, not a tech trend to monitor. Every AI workflow you are running today was likely designed around closed-model availability and pricing. Most of those workflows did not need GPT-4 or Claude Opus in 2023 and they do not need GPT-5.6 Sol or Claude Opus 4.8 now. The question worth spending an hour on this week is: which of your AI tasks actually need the premium, and which are just running there because you have not checked recently.

The governance question is the one most operators skip. Open-source models on third-party inference platforms like Together require the same data handling review you would give any SaaS tool. The models are not inherently less secure, but the infrastructure agreements are different and need to be reviewed deliberately. Get that right and the cost argument becomes straightforward.

Where This Fits in the AI Stack

Together AI sits at the infrastructure layer between foundation model developers and the applications and workflows businesses build. Rather than building its own models, Together focuses on hosting, optimising, and serving open-source models with enterprise reliability and speed.

This layer is sometimes called the inference tier. Winning at inference means running models faster, cheaper, and more reliably than running them yourself. As open-source models improve, the inference tier becomes more strategically important because the cost savings from model choice only materialise if the infrastructure can serve them at production quality and scale.

The companies investing alongside Aramco in this round, particularly Nvidia and Vista Equity, suggest the market sees inference infrastructure as a durable category rather than a transitional one.

Questions Operators Are Asking

Do I need to move my AI workflows to Together AI specifically? Not necessarily. The story here is that the open-source inference category has reached sufficient scale and quality to be a serious evaluation option. Together is one platform in this space. The decision is whether to evaluate open-source alternatives at all, and if so, Together provides a reasonable starting point with a broad model catalogue.

How do I know which of my AI tasks can handle open-source models? Start with tasks where you are using AI for drafting, summarising, classifying, or formatting rather than complex multi-step reasoning or high-stakes decisions. These tasks are typically well within the capability range of current open-source models. Run a parallel test for two weeks and compare output quality against your current provider.

Is it safe to put business data through Together AI's infrastructure? That depends on your data classification and industry requirements. Together AI offers enterprise agreements and data handling terms similar to other SaaS infrastructure providers. For regulated industries or data that cannot leave specific jurisdictions, you need to review their data processing agreements carefully before proceeding.

What happens if Together AI changes its pricing or goes under? This is a reasonable concern for any single-provider dependency. The practical mitigation is to build your AI integrations on standardised API patterns that make provider switching feasible within a sprint. The infrastructure investment in this round suggests Together is not a near-term solvency risk, but the point holds for any provider.

Does this mean I should stop using OpenAI or Anthropic entirely? No. Premium closed models retain genuine advantages for complex reasoning, multi-step agent tasks, and high-stakes outputs. The smarter approach is deliberate tiering: open-source for volume tasks, closed premium models for the work where accuracy and nuance genuinely justify the cost.

Citable Summary

Together AI raised $800 million at an $8.3 billion valuation on 1 July 2026, led by Aramco Ventures with co-investment from Nvidia, Vista Equity Partners, General Catalyst, and others. The company reports $1 billion-plus in annual bookings and triple the open-source model usage on its platform compared to a year ago. Open-source models available through Together's platform cost 60 to 90 percent less than comparable closed alternatives from OpenAI and Anthropic, making this raise a structural signal that cheap, production-grade open-source AI has arrived as a serious enterprise infrastructure category.