Nvidia Recently Discussed Acquiring Software Startup OctoAI for $165 Million
The Takeaway
• Deal adds to a string of small Nvidia acquisitions
• OctoAI’s service helps developers run open-source AI models faster
• It raised $132 million from investors including Tiger, Madrona
Nvidia has been in advanced talks to buy OctoAI, a Seattle-based startup that sells software to make artificial intelligence models run more efficiently, according to a message sent to shareholders. The purchase would add to a string of small AI startup acquisitions by the AI chip heavyweight, which is trying to strengthen its software and cloud-computing services while U.S. antitrust officials probe its activities.
Nvidia has proposed paying about $165 million for the company, before accounting for the company’s debt and other expenses, according to a document OctoAI sent to shareholders.
The price suggests that shareholders—especially employees—won’t get much of a payout. The company raised $132 million from investors including Tiger Global Management, Madrona Venture Group and Amplify Partners, ultimately fetching a valuation of about $900 million in 2021.
OctoAI CEO and co-founder Luis Ceze didn’t respond to a request for comment. A spokesperson for Nvidia declined to comment.
Nvidia has been buying small AI and cloud startups to make it less expensive and easier for developers to run AI models using Nvidia chips. One of the biggest criticisms of Nvidia’s chips is that because they are expensive and in high-demand, it’s difficult for companies to scale AI at a reasonable price.
Founded five years ago, OctoAI targets software developers who work on machine learning and has taken aim at one of the toughest problems associated with building applications that use AI: getting ML models to perform well across different types of processors. OctoAI does this by translating the models into code that different processors can handle. OctoAI says its software can help organizations launch machine-learning applications more quickly while also cutting the computing costs of running the models.
For engineers, OctoAI automatically tweaks models to run well on specific chips. It claims that its approach works better than the customized software major manufacturers develop for that purpose. One of Nvidia’s strengths is Cuda, its proprietary software for developing apps that use the company’s vaunted AI chips.
Formerly known as OctoML, the startup spun out of the University of Washington in 2019 by the creators of an open source project called Apache TVM. It originally focused on helping developers deploy machine learning models, but last year shifted its focus to selling a service that allows developers to tweak open-source or custom AI models quickly, regardless of what type of chips power the models. It generated revenue in the low-single-digit millions last year, company leaders told some employees, according to a person who heard the remarks.
OctoAI has worked closely with Amazon Web Services, Advanced Micro Devices and Qualcomm to help developers use those companies’ AI chips, which compete with Nvidia’s. That’s one reason the U.S. Department of Justice, which is examining other recent Nvidia acquisitions and other allegations against the company, might decide to examine the Nvidia-Octo deal if it closes.
OctoAI and Nvidia have a relationship too. This year Nvidia launched a software offering, known as Nvidia Inference Microservices, which is aimed at solving some of the problems OctoAI was focused on. Nvidia and OctoAI announced a collaboration earlier this year related to the NIMs service. Before that, the two companies worked closely together—for instance, Nvidia has previously given OctoAI early access to its cutting-edge chips so the startup could test how to run AI models on the chips effectively.
Nvidia earlier this year bought two small Israeli startups, Run:AI and Deci, which also aim to lower the cost of developing or running AI models that generate text, images and code.
Despite investor enthusiasm for startups involved in conversational artificial intelligence, some founders have struggled to stay afloat as competition rises. Major AI software startups such as Adept, Inflection and Character recently agreed to quasi-acquisitions with larger cloud companies.
Some arrangements, such as Google’s deal for chatbot startup Character, investors received more than double the money put in. But the acquisitions can leave little for employees.
In OctoAI’s case, common stock holders are only getting two cents for every share they own, according to merger documents. Ex-employees who exercised their vested options at higher price when they left the company would be looking at losses.