Firms Break Venture Rules for AI.

"Venture investors are diving into generative AI -- and breaking some of their own rules in the process.

Generative AI startups globally raised $17.8 billion as of Sept. 19, up from the $3.9 billion invested in all of 2022, according to data provider Dealroom.co. Venture firms are raising massive funds with the intention of investing more capital in artificial intelligence.

Many investors believe the technology will one day unlock a vast new market. While that may be true, one shouldn't lose sight of some of the issues in the business model.

Venture capital frowns upon the use of equity dollars for capital expenditures. Yet many startups spend expensive equity capital on computing infrastructure. Another rule of thumb startups in this sector break -- avoiding overreliance on fast-moving incumbents. Many generative AI startups do just that, with OpenAI as the main supplier of large language models, such as GPT-4, used by many other companies.

The news that OpenAI ousted its chief executive, Sam Altman, rattled investors and founders in the sector, re-emphasizing the risk of dependence on a mega tech provider.

Venture capital in generative AI now often goes to securing scarce computing infrastructure. "In the last nine months, a lot of venture capital went to lock in supply contracts," said Max Gazor, general partner at CRV. His firm has made several bets in early-stage startups in generative AI.

"Right now it's a land grab. Everyone is operating to stay in business and we'll figure out the economics later," said Gazor. Venture firms reportedly signed deals with compute providers on behalf of their startups, for example.

Rounds for generative AI startups tend to be much larger than for other software companies due to the capital intensity of the businesses. The median Series A round for generative AI startups was $16 million, more than double the size of all Series A rounds in 2023, according to Dealroom.

Longer term, generative AI startups will need to figure out where to get continued funding to scale their businesses at a time when many later-stage private investment firms have pulled back from the venture market. Financing risk is significant, Gazor said.

Startups would need for chip supply to increase and become more reliable, something that they have limited control over.

"The reality is that these large language models do consume significant amounts of electricity and will remain an important infrastructure cost that should depress gross margin," said Tomasz Tunguz, general partner at Theory Ventures, which has made a couple bets in generative AI. He said he believes generative AI companies will have gross margins below those of traditional cloud software.

Will the market be so huge that it would compensate for the large upfront and continuing capital expenditures?

For now the market for the technology remains limited. Generative AI spending by enterprises will amount to about $2.5 billion this year, according to estimates in a recent report from Menlo Ventures. While enterprises are eager to explore the technology, they are concerned about the uncertain return on investment, the report found.

Not only is the market still limited, incumbents dominate, due to their "scale, distribution, brand, and engineering resources," the Menlo report said. That's the other big challenge to the VC model when it comes to generative AI deals.

Many startups in the sector rely on OpenAI. "It is the engine that's powering most of the production-grade AI today," Gazor said.

OpenAI's board said on Friday that Altman will leave the company. "The board no longer has confidence in his ability to continue leading OpenAI," the company said. Over the weekend, investors and many employees pushed to bring Altman back, while Altman considered launching another company in the AI space.

Gazor said the events only reinforce the problem in the generative-AI market. "We need many strong platforms to emerge to create a resilient ecosystem. Otherwise events like this create major turbulence and disruption," Gazor said.

Reliance on another company has generally been viewed as a bad idea in venture, since startups were burned when mega companies they depended on changed terms and prices, or released new features or competitive products.

"What's so scary about the AI space is that the incumbent is actually a net new player who has this breakneck speed of product development and evolution of talent that you don't typically see," said Caitlin Bolnick Rellas, general partner at CRV, speaking about OpenAI. "A lot more chess has to be done in terms of evaluating AI opportunities because the incumbent is so nimble," she added.

OpenAI has more than two million developers using its platform, said a representative for the company, who responded to questions prior to Altman's departure.

"It's important to us to equip developers with the tools they need to build great products and businesses," the representative said, in response to a question about the company potentially becoming a threat to its own customers. The representative didn't respond to a question about the impact of Altman's departure on OpenAI's ability to deliver.

Moves by larger players in generative AI caused issues for startups so far. The release of GitHub Copilot, developed in collaboration with OpenAI, squeezed competitive AI productivity tools for developers, while Bloomberg's launch of BloombergGPT for financial data negatively affected fintechs, Bolnick Rellas said.

"One core question in investing right now is to identify and invest in areas that are not likely to get run over by the foundation models," said Sheila Gulati, managing director at venture firm Tola Capital.

There are reasons why large tech companies may limit how much they step on the toes of customers -- for one, they need a thriving ecosystem of businesses to make calls to their models. Competition among developers of the large LLMs could also help smaller players." [1]

1.   Firms Break Venture Rules for AI. Chernova, Yuliya. Wall Street Journal, Eastern edition; New York, N.Y.. 21 Nov 2023: B.2.