Together AI Raises $800M — Open-Source Inference Crosses the Billion-Dollar Line

Together AI Raises $800M — Open-Source Inference Crosses the Billion-Dollar Line

On July 1, 2026, Together AI closed an $800 million Series C at an $8.3 billion valuation — one of the largest AI infrastructure funding rounds of the year. The headline number is impressive, but the real story is in what it signals about where the AI industry is actually heading.

While OpenAI files for a trillion-dollar IPO and Anthropic eyes $965 billion valuations for their closed frontier models, the company quietly making open-source inference cheaper and more reliable just became worth eight times more than its previous round — and the money is flowing to infrastructure, not models.

The numbers behind the headline

Together AI doesn’t build foundation models. It builds the cloud infrastructure that lets enterprises run other companies’ open-weight models — DeepSeek, Nemotron, MiniMax, Kimi, and GLM — on NVIDIA GPU clusters through an OpenAI-compatible API. The bet is that as open-weight model quality converges with proprietary frontier models, the economic value in AI shifts away from whoever controls the model and toward whoever can serve it most cheaply at scale.

The numbers in this round suggest that bet is landing:

• Annual bookings crossed $1.15 billion in the most recent quarter — a 38x increase over two years
• Open-weight model usage tripled on Together’s platform over the past 12 months
• Customers report 6-20x cost reductions versus equivalent closed-model APIs
• The valuation jumped 2.5x from $3.3 billion (February 2025 Series B) to $8.3 billion

The round was led by Aramco Ventures — the venture arm of Saudi Arabia’s state oil company — with participation from NVIDIA, Vista Equity Partners, General Catalyst, Emergence Capital, Salesforce Ventures, SentinelOne’s S Ventures, and Schneider Electric’s SE Ventures. The investor list reads like a who’s who of cloud infrastructure and enterprise software.

Named customers include Cursor (the AI-powered code editor), Cognition (makers of Devin), and Decagon — an enterprise AI company that reportedly cut its inference costs sixfold after switching from closed to open-weight models on Together’s platform.

The economic argument for open-weight

CEO Vipul Ved Prakash (who sold his previous startup, Topsy, to Apple for $200+ million in 2013) has been clear about the thesis: the gap between open-weight and closed models is narrowing fast enough that the cost premium of proprietary APIs is no longer defensible for most workloads.

The company says the cost advantage climbs to 60x in configurations involving batch inference and heavily repeated workloads, where Together’s inference engine can reuse cached computation across similar queries. That’s not a marginal improvement — it’s a fundamentally different economic equation.

OpenRouter, another company cashing in on the same trend, independently corroborated the tripling of open-source model usage across the industry in the past year. They themselves more than doubled their valuation to $1.3 billion in a year, raising separate funding in May 2026.

The counter-narrative nobody wants to acknowledge

Here’s the uncomfortable truth this funding round highlights: while the public narrative is dominated by trillion-dollar IPO filings for closed-model companies, the actual enterprise spending is increasingly flowing to open-weight inference infrastructure.

Meanwhile, the closed-model narrative itself is showing cracks. At an internal town hall on July 2, Meta CEO Mark Zuckerberg admitted that the company’s AI reorganisation — designed to accelerate autonomous AI agent development — “hasn’t come to fruition.” AI agent progress over the past four months has been slower than expected, he told staff. Meta’s stock is actually down 11.7% year-to-date through July 2, trailing the S&P 500’s roughly 9% gain.

That’s not to say closed models are dead. GPT-5 and Claude still outperform open-weight alternatives on frontier reasoning tasks. But for the broad middle of production AI workloads — summarisation, classification, code assistance, customer support — the open-weight option is no longer a compromise. It’s the economic choice.

What this means

Together AI was co-founded by Vipul Ved Prakash alongside Stanford professor Percy Liang and ETH Zürich/University of Chicago associate professor Ce Zhang in 2022. Four years later, they’ve raised a total of roughly $1.2 billion across three rounds (a $102.5M Series A in 2023, $305M Series B in early 2025, and this $800M Series C) and built a platform that serves thousands of paying customers.

The broader trend is even more significant. Other neocloud providers are seeing the same capital influx: Upscale AI raised a Series A+ extension totalling $500 million at a $2 billion valuation last month, while AMD-focused TensorWave pulled in $350 million at $1.55 billion. The infrastructure layer is where the smart money is going.

The question every engineering and procurement team needs to ask now is increasingly difficult to avoid: is the cost premium of closed frontier models still justified for your workload? For a growing number of companies, the answer appears to be “no.”

Sources: TechCrunch, Reuters, TechTimes, BusinessWire