Generative AI companies continue to raise huge amounts of capital to fuel their commercial — and, in some cases, open source — ambitions.
See Together, a startup creating open source generative AI and AI model development infrastructure, which today announced that it closed a $102.5 million Series A funding round led by Kleiner Perkins with participation from Nvidia and Emergence Capital. The tranche — over five times the size of the company’s previous round — will be put toward expanding Together’s cloud platform that lets developers build on open and custom AI models, co-founder and CEO Vipul Ved Prakash says.
“Startups and enterprises alike are looking to build a generative AI strategy for their business that is free from lock-in to a single vendor,” Prakash wrote in a blog post published on Together’s website this morning. “Open source AI provides a strong foundation for these applications with increasingly powerful generative models being released almost weekly … We believe generative AI is a platform technology, a new operating system for applications, and will have a long-range impact on human society. The AI ecosystem will consist of proprietary models and open models, and it’s incredibly important that this future has choice and options.”
Vipul co-founded Together, launched in June 2022, with Ce Zhang, Chris Re and Percy Liang. Prakash previously founded social media search platform Topsy, which was acquired in 2013 by Apple, where he later became a senior director.
With Together, Prakash, Zhang, Re and Liang aim to create open source models and services that help organizations incorporate AI into their applications. To that end, Together has built a cloud platform for running, training and fine-tuning models that the co-founders claim offers scalable compute at lower prices than the dominant vendors (e.g., Google Cloud, AWS, Azure).
As Together explains in a blog post: “We’re achieving and offering significant reductions in cost of interactive inference workloads on large models. We optimize down the stack, with thousands of GPUs located in multiple secure facilities, software for virtualization, scheduling and model optimizations that significantly bring down operating costs.”
At present, Together operates a cloud spanning datacenters in the U.S. and EU, including servers operated by partners Crusoe Cloud and Vultr, that delivers around 20 exaflops of compute in total — scaling in clusters from 16 GPUs to 2,048 GPUs. Customers include NexusFlow, Voyage AI and Cartesia, some of whom also leverage Together’s APIs for serving models.
Pika Labs, which coincidentally raised $55 million this week, also built a model — a text-to-video model — on Together’s GPU clusters. Pika’s recently trained new iterations of the model from scratch with the clusters, which it uses to run its models to generate millions of videos per month for its early access users.
“By creating custom infrastructure, we can offer significantly better economics,” Vipul continued. “The Together platform allows developers to quickly and easily integrate leading open source models or create their own models through pre-training or fine-tuning. Our customers choose to bring their generative AI workloads to Together owing to our industry-leading performance and reliability, while still having comfort that they own the result of their investment in AI and are always free to run their model on any platform.”
To complement the cloud service, Together offers what it calls Custom Models, a consulting offering that allows customers to bring their own data to the Together cloud platform and have Together’s team work to design, build and test a model.
Together also heavily invests in open source AI research. One of the company’s first projects, RedPajama, sought to foster a set of open source generative AI models including “chat” models along the lines of OpenAI’s ChatGPT. Together’s also released a fine-tuned version of Meta’s Llama 2 text-generating model; GPT-JT, a fork of the open source text-analyzing model GPT-J-6B (from the open research group EleutherAI); and OpenChatKit, an attempt at a ChatGPT equivalent.
Companies — and investors — are hungry for generative AI of all sorts. According to IDC, generative AI investments will rise from $16 billion this year to a whopping $143 billion in 2027. Generative AI firms are monopolizing venture capital investments, having attracted about 40% of all startup rounds — $11.9 billion — in the first six months of 2023.
But generative AI is far from a sure bet. Stability AI, for example, once the darling of VC firms including Lightspeed Venture Partners, O’Shaughnessy Ventures and Coatue, is said to be exploring a sale as profits fail to materialize and the company’s financial position worsens.
Comment