Smart stuff
Compile |Gong Yuwen
Edit |Cheng Qian
Zhidong reported on March 5 that yesterday, the American AI software startup Baseten announced that it had received a $40 million Series B investment, and according to a direct source familiar with the matter, Baseten was valued at more than $200 million.
The round was led by world-renowned venture capital firms IVP and Spark Capital, with participation from existing backers such as established US venture capital firm Greylock, newly established venture capital firm Conviction Partners, and angel investor Lachy Groom.
After seeing the explosion of tools like chatbot ChatGPT, Baseten focused its business on AI inference, which saw revenue skyrocket over the last year to more than millions of dollars and attract several large companies and tens of thousands of developers.
At present, inference is gradually dominating the AI market, and many companies have shifted their focus on large models from spending billions of dollars to train AI models to practical applications of AI. Nvidia CEO Jensen Huang also noted that reasoning accounted for 40% of its computing revenue last year.
According to the blog, the startup will continue to focus on the inference business, helping customers deploy trained AI models into applications.
Baseten co-founders Amir Haghighat, Tuhin Srivast**a, Philip Howes and Pankaj Gupta. (from left to right).
Founded in 2019 and headquartered in San Francisco, USA, Baseten was originally founded to focus on machine learning. In the three years since its founding, Baseten's business software has made it easier for businesses to build AI-powered applications. Tuhin Srivast**a, CEO of Baseten, co-founded the company with Amir Haghighat, Philip Howes and Pankaj Gupta but they didn't get much help at first. Srivastava told Forbes that the income they earned in exchange for their efforts was "basically zero".
Against the backdrop of rapid growth in large models and mass consumer adoption of products such as chatbot ChatGPT and Bunsheng Diagram Midjourney, Baseten began to explore the most performant, scalable, and reliable ways for machine learning workloads.
Over the past 12 months, we've been in the midst of a wave of open source models becoming really good. Srivastava said this in an interview. 2023 is a key time for Baseten to seize the opportunity. "2023 has made us realize that reasoning is actually the biggest market here, so we basically cut everything and refocus on reasoning," Srivastava said. ”
Specifically, he was referring to the huge repercussions generated by ChatGPT around OpenAI last year. "What makes ChatGPT so great is that it doesn't feel clumsy. It turns out that this is because there are more than a dozen or dozens of people studying reasoning inside Openal", which makes ChatGPT more responsive and eliminates the risk of possible downtime.
Baseten helps customers publish their trained open source or custom AI models into applications for public use. Customers can deploy models to Baseten's cloud without the need to build the required AI infrastructure.
Baseten's software automatically adjusts the amount of computing resources a company pays for based on the amount of traffic the AI model has at a specific time. It can also help customers get scarce GPUs if they need them. Guaranteed by a full-process security approach (SoC, HIPAA), Baseten's autoscaling and cold start capabilities can effectively control costs, and native workflows can serve large models in production, so users don't need to worry about version management, rollout, and observability.
The company powers core machine learning workloads for enterprises and developers, such as Descript, Picnic Health, Writer, Patreon, Loop, and Robust Intelligence. In 2023, Baseten has scaled the inference load hundreds of times "without a minute of downtime," Srivastava mentioned. In the future, the company will further introduce new features such as multi-cloud support and integration with best-in-class runtimes such as Tensorrt.
Baseten has observed the development trend of the market from the rapid development of the large model behind ChatGPT, focusing on the inference market, and its business has grown rapidly, and its revenue has jumped to one million dollars in one fell swoop. Somesh Dash and Shr**An Narayen, partners at venture capital firm IVP, told Forbes that they were desperate to invest after seeing the company grow so quickly without a sales and marketing team.
Srivastava said part of the financing amount will be used to build a sales and marketing team. He expects Baseten to nearly double its current size of 25 by the end of the year. Some of the new funds will also be used to purchase more GPUs and upgrade and optimize the company's software.
About 20 large companies and tens of thousands of developers are already paying for Baseten's product, which is billed based on the computing power used, he said. For example, Patreon, a membership-based platform where consumers pay creators for access to exclusive** and other content, uses Baseten's products to manage the deployment of OpenAI's open-source speech recognition model, Whisper, which can automatically generate subtitles.
However, the current boom in AI inference could also pose risks later in the future. Baseten's products rely on customers opting to use a custom open source model, although managing the infrastructure requires additional work.
If closed-source models such as OpenAI's GPT, Anthropic's Claude or Google's Gemini dominate the market, Baseten will largely lose its relevance. But Srivastava believes that in the future, developers will use both open source and closed source models. He said that while closed-source** services may have better performance, factors such as cost, speed and data privacy may attract startups to open source services.
Another risk comes from the increasing competition among AI infrastructure providers. Will Reed, a general partner at Spark Capital, a well-known U.S. venture capital firm who joined the board of directors in the fundraising, told Forbes that based on his observations, Baseten is winning an "incredibly disproportionate share" of clients, even though it's not the cheapest option.
Among Baseten's competitors is Descript, a startup backed by Reed that provides all the machine learning capabilities for ** and podcast editing apps. Srivastava bets that customers will choose Baseten over rival Descript because of its good performance – and no downtime for the AI applications it supports in 2023.
As mentioned earlier, reasoning accounted for 40% of its calculated revenue last year. Nvidia CEO Jensen Huang recently said in a recent interview with Wired that if so much computing power is dedicated to reasoning, it means that "AI has finally succeeded."
Investors in the market are increasingly backing companies focused on AI inference: Modal Labs, which runs on the cloud, has received funding from Redpoint and Amplify Partners; Fireworks AI, which builds an AI production platform for developers, is supported by Benchmark and Sequoia; Last year, Replicate raised $40 million from investors, including Andreessen Horowitz, a company that aims to simplify and personalize open-source AI models for users.
At the same time, unicorn startups such as Together AI, which focuses on generative AI platforms, and Anyscale, which focuses on distributed machine learning, are also entering this space. Srivastava expects Amazon or Microsoft to be probably the next business to get into the space, "actually, you have to think about hyperscalers as well, because I think they do see that it's a very competitive market." ”
More and more players in the large model industry have begun to pay attention to the application and implementation of AI, and more abundant industry application cases have emerged, while the demand for AI inference has expanded rapidly.
In this context, the enterprises represented by BASETEN, based on the previous product and technology accumulation, have seen a significant trend of market development and won market opportunities. It can be seen that the phased transformation and development of the large model industry will still bring new opportunities for the development of start-ups.
*：forbes