In recent weeks, two related developments have threatened to reshape the rapid boom in artificial intelligence: Companies using the most advanced AI tools have begun to balk at the cost, and new open-source models from China have become nearly as powerful as the proprietary technology of major U.S. labs.
For Vipul Ved Prakash, co-founder and CEO of Together AI, a start-up specializing in open source AI, this confluence presents an opportunity.
Company margins “are being eroded by the cost of using closed-loop models like those offered by OpenAI and Anthropic,” Mr. Prakash said.
Executives have begun pushing back against “token maxxing” — tokens are the units of data used by AI models — while looking for solutions to further expand the technology’s use. As companies adopt open source models that can be freely used and modified, Together AI’s business has grown significantly.
The venture capital community has taken note.
On Wednesday, Together AI announced an $800 million funding round at a valuation of $8.3 billion, bringing its total funding to $1.3 billion. The new round was led by Prosperity7 Ventures, the venture arm of Saudi Aramco, Saudi Arabia’s state oil company, with participation from Nvidia, Vista Equity Partners, General Catalyst and others.
Mr. Prakash, a serial entrepreneur and former Apple executive, founded the San Francisco-based company Together AI in 2022 along with four academics. The company’s chief scientist, Tri Dao, wrote FlashAttention, a groundbreaking algorithm for improving model speed that is now a core part of Together AI’s product.
Demand for Together AI’s products is increasing rapidly. Over the past year, the company has seen a 10,000x increase in the number of tokens it processes per month. In the most recent quarter, annual revenue rose to $1.2 billion.
Together AI’s growth is evidence of the increasing demand for open source models, many of which are made in China. In June, a Chinese startup, Z.ai, released an AI model with similar capabilities to Anthropic’s powerful Fable and Mythos models.
Some experts are concerned about the adoption of Chinese open source models by US companies. They raise concerns that the models are being built by companies linked to the Chinese government and that the developers have illegally used American technology.
Despite these concerns, the relatively low price of Chinese models is tempting. According to a recent report from JP Morgan, usage costs per token can be one-fiftieth compared to their American counterparts.
Open source processing on OpenRouter, a marketplace for AI models, rose from 34 percent in January to 65 percent in June, according to a Citi analysis reported by Reuters. Microsoft, which helped fuel the AI boom with its early investment in OpenAI, is considering hosting a version of DeepSeek, a Chinese open-source model, in its Copilot program.
Together AI helps companies gain access to cheaper open source models for their AI products. The company rents and buys AI computing chips and uses specialized software to run AI models for customers faster and more efficiently – a process called “inference.”
The startup then offers hundreds of open source models on these chips that can be tailored to a company’s needs, eliminating the need for a company to hire a team of engineers to do it themselves.
Ashwin Sreenivas, co-founder of AI startup Decagon, said using Together AI has brought “tremendous” savings to his company. Decagon, which makes an AI customer service product, uses a mix of closed models from leading AI companies such as OpenAI, as well as open source models from Together AI. Mr. Sreenivas estimated that moving tasks from closed models to Together AI cost a fifth to a seventh as much.
Demand for open source models is increasing as Anthropic and OpenAI prepare for large public offerings. “As they go public, many other companies are thinking about a future without them,” said Mr. Prakash, CEO of Together AI.
Last week, The New York Times reported that OpenAI was considering delaying its offering, in part due to financial difficulties.
Together, AI plans to spend the majority of its newly raised capital on research and development. The company is also using some of its funds to build more computing capacity. To achieve this, its investors are proving helpful, including by introducing Together AI to “the right people who might be thinking about building more capacity,” said Abhishek Shukla, chief executive of Prosperity7.
Competition between companies hosting open source models is hot. Last month, Together AI’s rival Baseten raised $1.5 billion at a valuation of $13 billion. Another competitor, Cerebras, held an IPO in May and now has a market value of around $50 billion.
Mr. Shukla believes Together AI will try to follow suit. The company is “heading to the public markets,” he said.
https://www.nytimes.com/2026/07/01/business/dealbook/together-ai-funding.html
