Together AI raised an $800 million Series C at an $8.3 billion valuation on July 1, 2026, and the thesis behind the check is the most important part: that the future of AI infrastructure is running open-weight models cheaply, not renting the most expensive closed APIs. Led by Aramco Ventures with Nvidia, Vista Equity, General Catalyst and others, the round more than doubles Together's valuation in about 16 months and funds a roughly 50-fold expansion of its compute. The company's pitch to every startup watching frontier-model bills eat their margin is blunt: run open models on our optimized stack and pay a fraction as much.
- Together AI raised $800M at an $8.3B post-money valuation, led by Aramco Ventures, up from a $305M Series B at $3.3B about 16 months earlier.
- Its business is running open models (DeepSeek, Nemotron, MiniMax, Kimi) on custom inference software it says cuts costs by up to 80% versus closed systems.
- Annualized bookings crossed $1.15 billion last quarter, with customers including Cursor, Cognition and Decagon.
- It has secured commitments for over 500 megawatts of compute to fund about 50x capacity growth over five years.
What does Together AI actually sell?
Infrastructure for running and training open-weight models cheaply. Rather than build its own frontier model, Together acquires or rents GPUs and layers on inference-optimization software that speeds up popular open models like DeepSeek, Nemotron, MiniMax and Kimi. The pitch is that these open models now deliver comparable or better performance than closed systems for many tasks, and Together can serve them at a fraction of the cost, by its numbers cutting the price of running popular models by up to 80%. Where pure GPU-rental players like RunPod hand you raw capacity, Together bundles the capacity with the optimization, selling served tokens rather than bare hardware.
RelatedGeneral Intuition Raises $320M to Turn Games Into Agents
Why are investors paying $8.3 billion for it?
Because the growth and the timing are both strong. Together's valuation jumped from $3.3 billion at its Series B to $8.3 billion in about 16 months, and annualized bookings crossed $1.15 billion last quarter across thousands of paying customers, including high-profile AI companies like Cursor and Cognition. The round is led by Aramco Ventures, an energy-backed investor with the balance sheet for the capital-intensive compute buildout Together needs, and Nvidia's participation is a tell about who supplies the chips. The macro backdrop is a market pouring money into AI infrastructure as GPU demand outruns supply, which is why neocloud valuations are soaring across the board.
| Company | Together AI | Baseten | RunPod | Groq |
|---|---|---|---|---|
| Core offering | Open-model train + inference | Model inference | Raw GPU rental | Custom inference chips |
| Latest valuation | $8.3B | $13B | Smaller | Large |
| Latest raise | $800M Series C | $1.5B | $100M Series A | Prior rounds |
| Differentiator | Inference optimization software | Deploy + scale tooling | Low-cost capacity | Speed via custom silicon |
How does this fit the wider funding wave?
It is one of several megarounds landing in the same window. Baseten raised $1.5 billion at a $13 billion valuation the prior month, Upscale AI took $500 million, and TensorWave pulled $350 million, all inside the neocloud and AI-inference lane. The common thread is that the money is flowing into the operating layer of AI, the companies that make models cheaper to run, rather than into consumer apps. Together's edge in that crowd is its software: bundling inference optimization with compute is a stickier, higher-margin position than renting bare GPUs, and its named customers suggest the approach is winning real production workloads, not just demos.
- ~2023Together AI founded. By Vipul Ved Prakash with Percy Liang and Ce Zhang, focused on open models.
- ~Early 2025$305M Series B at $3.3B. The base the new round more than doubles.
- Q2 2026Bookings cross $1.15B annualized. Thousands of paying customers, including Cursor and Cognition.
- Jul 1 2026$800M Series C at $8.3B. Led by Aramco Ventures; total funding to $1.3B.
- 2026 to 2031~50x compute expansion. Backed by 500MW-plus of committed power.
- Open vs closed quality. The whole bet rests on open models staying competitive. Watch DeepSeek, Kimi and Nemotron keep pace.
- Power, not chips. With 500MW committed, energy and datacenter buildout become the real constraint. Watch delivery timelines.
- Margin durability. Neocloud pricing can compress fast. The inference-optimization moat has to hold to justify $8.3B.
- Customer concentration. Marquee names like Cursor are great until one leaves. Watch how diversified the $1.15B in bookings is.
Our take
Together's raise is a clean bet on the most defensible idea in AI infrastructure right now: that the money is in running models, not owning them. By staying model-agnostic and pouring its effort into making open weights cheap and fast to serve, Together sidesteps the brutal, capital-incinerating race to train a frontier model and instead sells the shovels, with better margins than the bare-metal renters because it owns the optimization layer. The risks are real and specific: the thesis dies if open models stop keeping up with closed ones, and the 50x expansion is now a power-and-datacenter execution problem as much as a software one. But $1.15 billion in annualized bookings and customers like Cursor say this is a business, not a story. In a year when frontier-API costs are eating startups alive, Together is selling the escape hatch, and $800 million says the market believes a lot of companies want out.
- OfficialTogether AI Series C announcement , the official release
- FundingTechCrunch, the raise , valuation, investors and context
- ReferenceTogether AI blog , product and inference-engine details
Original analysis by GenZTech. Figures current as of July 2026. Source: techcrunch.com