Venture capitalists at Bessemer Venture Partners, on an offsite in Montana earlier this year, decided to drop $1 billion on AI startups.
Who hasn’t spent a bit more than they bargained for while on vacation?
“Literally the No. 1 topic that every single partner at the firm was talking about was AI in their respective verticals and categories,” Bessemer partner Talia Goldberg told Tech Brew. “It was pretty striking because at Bessemer, it’s pretty rare that we all agree on anything.”
Bessemer hasn’t actually spent the sum yet, but it announced this week that it’s committed to doing so out of its currently available funds. The venture firm—which has existed for more than a century and whose portfolio companies have included Pinterest, Shopify, and LinkedIn—used the announcement to telegraph to AI-native startup founders how serious it is about the technology’s potential.
Even as the overall VC market tanks amid economic uncertainty that’s hit the tech industry especially hard, firms like Bessemer continue to find money for startups based around the latest advances in language- and image-generating AI. Sound Ventures announced a similar $240 million fund dedicated to AI earlier this month, Salesforce’s venture arm unveiled a $250 million AI fund, and Lux Capital announced a $1.15 billion fund with AI mentioned as one focus area.
Goldberg said Bessemer partners were encouraged by the business impact they’ve seen from the technology in terms of cost savings and in areas like customer service and support. Early usage of the tech has so far proved promising, she said.
“We have never seen such rapid adoption with such clear impact, either like business impact or operational impact on the companies, ever,” Goldberg said.
Burning millions: At the same time, however, training large language models (LLMs) and other types of generative systems currently has quite a price tag. A recent report from The Information found that OpenAI lost $540 million last year as it developed ChatGPT.
Goldberg said she sees those economics improving as hardware costs decrease, training data becomes more available, and development processes get more efficient.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
“We’re already seeing that cost curve improve,” Goldberg said. “The processing power needed to train AI models is increasing. But the tasks that can be performed with fewer resources are also increasing. So that’s reducing the cost of training.”
Speaking at startup A.team’s generative AI summit event in Manhattan this week, Lux Capital partner Grace Isford said that some of her firm’s AI-focused portfolio companies, which include generative AI players like Hugging Face and Runway, are nearing profitability.
“There is a way to run a profitable business in this space, but it does depend on how much compute you’re running and what your contracts are structured like, and what your business model really is,” Isford said.
Legacy versus startup: There has also been a question around whether established companies or startups might have a leg up when it comes to implementing generative AI.
On the one hand, big, established companies that have explored LLMs have the investment money to train these systems and the accumulated data and expertise to give them proprietary knowledge.
“I believe that incumbents and existing enterprises have a really big advantage here,” Madison Hawkinson, an investor at Costanoa Ventures, said at the A.team event. “Anyone who owns a problem and a solution and builds that reinforcement learning feedback loop via tasks or humans or data collection to do so, I think has a really big advantage.”
On the other hand, Goldberg said Bessemer is expecting generative AI to create entirely new categories of business in which there won’t yet be legacy competitors, mentioning image-generation companies like Midjourney and copywriting startups like Jasper as examples.
“I get really excited about these net-new categories that just previously weren’t possible before, and are now possible, these AI-native categories,” Goldberg said. “And I think those are types of categories that are really, really well-suited to startups, in part because there aren’t existing incumbents that you have to compete with.”