A month ago, generative AI startup Typeface raised $100 million at a valuation of $1 billion. Around the same time, French startup Mistral AI received $113 million funding without a working product in sight. Separately, media reports revealed open-source platform Hugging Face raised $200 million at a $4 billion valuation. Such multi-million dollar deals reveal a rare spot of joy for venture capitalists facing a dwindling startup market. While interest in startups building AI products has grown among VCs in the light of OpenAI’s ChatGPT, some see a bubble in the making.
“Generative AI is in a bubble. And this time the wrong assumption is that generative AI is either the product or the innovation,” said Vin Vashishta, founder of data consulting firm V Squared
No wonder VCs have their ears to the ground and stick to the basics. “We think of AI startups in five layers: hardware infra, software infrastructure, foundational models, data assets, and end-user applications. Our primary interest is in hardware infra and data assets - we don’t feel the other 3 layers are particularly conducive to small-cheque, seed-stage investing,” Sheetal Bahl, partner at Merak Ventures said.
AI startups and failures
There is genuine interest coupled with wariness, Bahl explained stating, “There’s a lot of froth, and misrepresentation. Like one couldn’t make a cup of chai without using blockchain two years ago, it seems one can’t do that without Gen AI now. So yes, there is a huge bubble in the making and in two years many people will look back and lament their impetuousness once more.”
(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)
While any startup venture is plagued with failures, in the case of Gen AI-focused startups, the failure rate will be even higher as a large number of founders and investors are participating in the AI ecosystem, said Swathi Dhamodaran, Associate at Trifecta Growth Equity.
“Startups that haven’t proven themselves in various ways will be vetted out,” she said. Some may run out of capital, and others may not find the right product market fit, she added. In evaluating AI startups, Damodaran “analyses the quality and amount of time spent by the founding team on problem statement, the strength of their network to find experienced talent on a broad front.”
On the product and distribution front, there are other details thoroughly looked into, including whether there is any proprietary data, the ease of acquiring data, model accuracy, how fluent and necessary the use-case is and the market for it.
“While it is early to talk about metrics like customer acquisition and sales cycle before the company gathers at least 50 customers, it is indicative of the urgency of the customer to find a solution and their propensity to pay and implement the solution, “ she added.
“It still remains to be seen where the real-world value will be in the generative AI startup space once their novelty is gone,” said Christian Cantrell, former VP, Product at Stability AI, and now founder of a generative AI startup. “It’s easy to do a demo but it’s something else entirely to build a product around it.”
In Cantrell’s experience, VCs are displaying more caution than usual because of these reasons.
“Nobody wants to get into a crypto situation again, so they are being probably more conservative than usual because it’s still unclear where the money will accrue eventually,” he added.
But in the catchall that generative AI has come to indicate, there are some that will inevitably rake in a lot. “Parties like NVIDIA or AWS Sagemaker are the ones guaranteed to make money, because they are the hardware part in the equation,” Cantrell said.
There will also be some exceptions on the research side. “If a handful of noted researchers are working on foundation models like GPT (like Anthropic AI), then they will probably raise a few million dollars easily because everyone wants to desperately replicate what OpenAI has done,” he explained.
And what of the spate of product companies that have mushroomed? Cantrell believes that, at the moment, generative AI is the most useful in the growing enterprise-productivity space.
“Generative AI tools are very good at analysing and interpreting, so their usage is more defined there. When it comes to creativity, its less clear where generative AI tools like image generators, etc. will find actual value, because creative professionals usually have a very specific vision in mind that they want to realise. Having said that, there’s a lot that depends on how teams want to incorporate these tools. But the few investments that have been made recently in AI product startups, are very good in my opinion and very well deserved,” he noted.