Nvidia-Backed Startup Together AI Raises $102.5 Million

Open-source cloud AI startup Together is the latest developer to draw a hefty payday as more and more investors pour money into artificial intelligence, announcing on Wednesday that the company raised $102.5 million in Series A investments—including from technology giant NVIDIA.

“I think one of the things that excited investors is our focus on research,” a Together spokesperson told Decrypt. “We’ve been building out a real leading research team that is doing deep research into how you optimize both the training and inference for these large generative models. And that’s proven out with some of the publications we’ve launched, [and] in the inference engine we launched recently, which is about three times faster than just about anything else out there.”

American venture capital firm Kleiner Perkins led the $102.5 million investment with participation from NVIDIA and Emergence Capital. Together said Kleiner Perkins partner Bucky Moore will join the Together board as a stipulation of their investment.

“[NVIDIA] is a partner and a leading company in this space,” Together co-founder and CEO Vipul Ved Prakash told Bloomberg. “They are the platform today for AI.”

New Nvidia Supercomputing Chips to ‘Fuel the Acceleration’ of AI

Others joining the Series A investment include NEA, Prosperity 7, Greycroft, 137 Ventures, Lux Capital, Definition Capital, and Long Journey Ventures. In May, Together announced that the company raised $20 million in seed funding, according to a report by Venture Beat.

“We care a lot about price performance, so the services we provide for our inference servers are six times cheaper than Open AI’s, and our training services are four times cheaper than AWS,” Prakash claimed.

The investment, the Together spokesperson said, would go towards improving its cloud platform for AI applications, investing in research and development, contributing to open-source AI, and enhancing its infrastructure.

The spokesperson outlined the company’s privacy-focused approach, saying developers have significant control over how data is shared and used on the platform with the flexibility to decide on data storage options.

“By default, we don’t use any data from our customers or their users for training of models,” the spokesperson said. “And [customers] able to control whether or not they turn on or off the data storage.”

During his interview with Bloomberg, Ved Prakash said Nvidia was the only option for Together’s computing needs.

Nvidia Created an AI to Better Understand COVID-19 and Accelerate Medical Discoveries

During its annual Supercomputing 2023 event in November, Nvidia unveiled its upgraded Grace Hopper Superchips, named for the pioneering computer scientist and U.S. Navy rear admiral. Looking to push AI development forward, Nvidia revealed work on developing quantum processing units (QPUs).

Earlier this year, Nvidia reported a 101% increase in quarterly revenue compared to the same time in 2022, bringing in total revenue of $13.51 billion in its second quarter ending July 30.

Together AI, the spokesperson said, believes strongly in open-source as a key part of the future of generative AI, adding that instead of a single dominant model emerging, a variety of proprietary and open-source models.

“You have more depth of control when fine-tuning an open-source model,” they said. “And most of these customers are finding they can achieve higher accuracy in their task by fine-tuning an open source model than just using a large proprietary model like GPT-4.”

The spokesperson emphasized the significance of using smaller open-source models, which he claimed would lead to several benefits, including accuracy, ease of use, cost-effectiveness, and enhanced performance.

“So, for all those reasons, we think open source will be a major thrust of how most organizations implement generative AI,” he concluded.

Launched in June 2022, San Francisco-based Together develops open-source AI cloud software. The team comprises veterans from Apple, ETH Zurich, and academics from Stanford University, the University of California, Berkeley, and MIT.

Source