Together AI company profile: valuation, funding & revenue

This Together AI company profile covers the company's valuation, revenue, funding rounds, investors and history. Together AI works in AI infrastructure, with a focus on AI cloud for open-source models. It was founded in 2022 and is headquartered in San Francisco, California, United States. Its latest valuation on record is US$8.3 billion (post-money) (July 2026). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

Together AI timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2022 — Together AI is founded in San Francisco by Vipul Ved Prakash (CEO), Ce Zhang, Percy Liang, Chris Ré and Tri Dao.
• 2025 — Series B: US$305M, valuing the company at US$3.3B.
• Jul 2026 — Series C: US$800M, valuing the company at US$8.3B. Investors: Aramco Ventures (lead).

Investors

Overview

Together AI runs a cloud platform designed for building and running AI applications with open-source and open-weight models. Developers and companies use it to run models such as Llama, DeepSeek, Qwen, Mistral and others through a simple API, fine-tune models on their own data, and rent GPU clusters for training. Together's pitch is that companies can get the performance of frontier AI at lower cost and with more control by using open models, without managing complex infrastructure themselves.

Together AI was founded in San Francisco in 2022 by Vipul Ved Prakash, the CEO, who previously co-founded Topsy (acquired by Apple) and Cloudmark, together with leading AI researchers including Ce Zhang, Percy Liang of Stanford, Chris Ré of Stanford and Tri Dao, creator of FlashAttention, a widely used technique for making AI models run faster. That research pedigree shapes the company's focus on making models faster and cheaper to run.

Products

• Inference API: pay-per-use access to hundreds of open models, optimised for speed and cost.
• Dedicated endpoints: reserved capacity for companies that need predictable performance.
• Fine-tuning: tools to customise models on a company's own data.
• GPU clusters: large clusters of Nvidia GPUs for training and research, rented by the hour or on longer contracts.
• Research and optimisation: software techniques such as FlashAttention and speculative decoding that make inference faster.

Business model

Together earns revenue from usage of its inference API, from dedicated capacity contracts and from GPU cluster rentals. Inference revenue grows with customers' usage of AI, while cluster rentals bring large contracts from AI labs and research teams. Third-party estimates in 2026 put its annualised revenue around US$1 billion; the company hasn't published audited figures, so we show revenue as not disclosed.

Funding

Together raised a US$305 million Series B at a US$3.3 billion valuation in 2025. In July 2026 it announced an US$800 million Series C at an US$8.3 billion post-money valuation, led by Aramco Ventures, the venture arm of Saudi Aramco. Earlier investors include General Catalyst, Prosperity7, Salesforce Ventures, Nvidia, Kleiner Perkins and Lux Capital. The company said the money would support its mission to make frontier AI accessible and expand its infrastructure.

Competition

Together competes with other AI inference platforms such as Fireworks AI and Baseten, GPU clouds such as CoreWeave, Lambda, Crusoe and Nscale, the hyperscale cloud providers, and closed-model providers that sell their own APIs. Its advantages are breadth of open models, performance research and a developer-friendly platform.

Together AI at a glance

Together AI is a San Francisco AI cloud company founded in 2022 by Vipul Ved Prakash and AI researchers including Ce Zhang, Percy Liang, Chris Ré and Tri Dao. It runs and fine-tunes open-source models and rents GPU clusters. It raised US$305 million at US$3.3 billion in 2025 and US$800 million at US$8.3 billion in July 2026, led by Aramco Ventures.

Why open models are growing

Open-weight models have improved rapidly, with releases from Meta, Mistral, DeepSeek, Alibaba's Qwen and others approaching the performance of closed models for many tasks at a fraction of the cost. Companies like open models because they can choose where to run them, customise them, avoid lock-in and protect sensitive data. But running models efficiently at scale is technically demanding. Platforms like Together let companies use open models without building that expertise, which is why demand has grown alongside the models themselves.

"Neoclouds" and the GPU economy

A group of specialist AI cloud providers, often called "neoclouds", has emerged to rent GPU capacity to AI companies. Unlike the hyperscalers, they focus entirely on AI workloads and can sometimes offer capacity faster or cheaper. Their economics depend on buying GPUs, often financed with debt, and keeping them busy at good prices. Together differs from pure GPU rental companies by adding software layers, inference optimisation and model hosting, which can earn higher margins than renting raw computing capacity.

Strategic investors from the Gulf

Aramco Ventures' lead role reflects Gulf countries' push to build AI capabilities and infrastructure. Saudi Arabia and the UAE have invested heavily in data centres and AI companies, seeking to diversify their economies. For AI infrastructure companies, Gulf investors bring capital and potential partnerships to build capacity in the region.

Questions for buyers of AI inference

• Which models do you need, and how do prices per million tokens compare?
• What latency and throughput can the platform guarantee?
• Can you deploy in specific regions for data residency?
• How easy is it to move workloads to another provider?

How inference pricing works

Most AI inference platforms charge per million tokens processed, with separate prices for input tokens (the prompt) and output tokens (the response). Larger models cost more per token, and output tokens usually cost several times more than input tokens because generating text takes more computation. Dedicated endpoints are priced by the hour for reserved GPUs, which suits companies with steady, predictable traffic. Because hardware gets cheaper and software optimisation improves, the price for a given level of model quality has fallen sharply year after year. That is good for customers but forces providers like Together to keep cutting their own costs through better software, cheaper power and higher utilisation of their GPUs.

Who uses Together AI

Together's customers range from individual developers experimenting with open models to AI-native startups building products on top of them, and large companies that want to run models privately. Some customers use it to serve chatbots and coding assistants; others fine-tune smaller models for specific tasks such as classification or document extraction, which can be cheaper and faster than calling a large general model. Research groups and AI labs rent its GPU clusters for training runs. This mix spreads Together's revenue across many types of demand, though large cluster contracts can make results lumpy from quarter to quarter.

Latest developments (October 2026)

Together AI raised an US$800 million Series C at an US$8.3 billion post-money valuation in July 2026, led by Aramco Ventures, according to Dealroom. Other investors include Vista, General Catalyst and Nvidia. The company, which rents GPU capacity and serves open-source models to developers, said annual bookings passed US$1.15 billion in the second quarter of 2026. See our AI cloud providers guide.

What to watch

Watch revenue disclosures, new data centre capacity, partnerships in the Gulf, the pace of open model releases and competition on inference pricing.

Takeaways

Developers and enterprises can use open models at lower cost without running infrastructure. Founders can see how research expertise translates into infrastructure advantage. Investors should watch margins as inference prices fall across the industry.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. Business Wire
2. TechCrunch