Can AI startups outlast the dot-com bubble? Investors aren’t so sure.

ChatGPT fever peaked last year, investors said this week at Collision, a tech startup and investor conference in Toronto, and in the aftermath, and with a sense of déjà vu, investors are becoming more aware of what’s in the cards.

“Everybody thought the dot-coms were convenient. ‘Oh, I can shop online,’ and look what happened,” Wesley Chan, co-founder and managing partner at FPV Ventures, said in Toronto. “I’m anti-AI,” he said of his investment strategy.

Of the 1,623 startups exhibiting at Collision this year (the most in the history of any Collision event), 20% are developing AI products, according to organizers, though a spokesperson said that figure doesn’t include the majority of startups that currently have an “AI component” to their business.

Investors said this week that only a handful of companies will be able to survive and thrive in the AI ​​boom.

Some said they are increasingly looking for startups with business models with long-term viability, products that solve enterprise business problems, and access to private or proprietary stores of data to train AI models.

The dot-com boom of the late 1990s was “disruptive” because every venture capital firm needed to invest in the sector, leading to inflation of costs like jobs and office space, said Mike Schroepfer, former chief technology officer at Facebook parent Meta Platforms and a partner at Gigascale Capital.

Investors say a similar dynamic is happening now with the AI ​​boom.

Schroepfer said in Toronto that AI startups should now be under strict scrutiny because the rush to fund them has created so much noise. Moreover, training large language models on the scale of OpenAI requires millions of dollars in computing and AI chips, so it’s not an area where startups can be competitive.

“The question remains: ‘Is there a market where I have the data to train a model and I can capture a very unique data moat?'” he says. “And then as customers use it, it gives me new data, so I’m building a data flywheel.”

Since the AI ​​boom began, investors have poured record amounts of money into AI companies such as Mistral AI, which raised $650 million earlier this month at a valuation of about $6 billion. Tech company Amazon invested $2.75 billion in Anthropic in March, bringing its total investment in AI companies to $4 billion. AI computing startup CoreWeave raised $7.5 billion in private debt financing in May.

These are just a few of the biggest investments: Investors poured $21.8 billion into generative AI-related investments last year, five times more than the previous year, according to research firm CB Insights. The average round size for these investments was $51 million, well above the industry average of $8 million, according to CB Insights research.

Judging by the types of startups exhibiting at Collision, some investors say the market has reached a tipping point, with a few companies like OpenAI and Anthropic dominating building large-scale language models, startups like Databricks and Scale AI providing data capabilities, and others tinkering with image generation and customer service, while others compete to stand out.

News Corp, owner of The Wall Street Journal, has a content licensing partnership with OpenAI.

Alex Manns, founder and CEO of AI-powered travel and transportation platform Flyr, said at Collision that he’s come across a number of AI startups that are developing products that do the same things as existing AI models.

“There are a lot of companies out there that look like vertical software-as-a-service, but they’re just a beautiful interface on top of a big language model,” Manns said. Others offer products to help companies analyze invoices, but there’s no reason to use them if an AI chatbot could do the job just as well, Manns said.

It’s the same reason why Matt Wood, vice president of AI products for Amazon’s cloud division, said some AI startups are being outdone by the pace of technological advances: Taking input data and feeding it to an AI model like OpenAI’s GPT-4 doesn’t make a big difference, he said in Toronto.

One solution, Wood said, is for startups to build “autonomous agents” — virtual AI workers that can carry out specific tasks on behalf of humans. He said the technology is developing so quickly that AI agents’ names may soon be as valuable as web domain names.

Wood added that the cloud giant will invest $230 million in generative AI startups this year. “We’re really going after what’s working,” he said. “There are some breakthrough areas right now where we have great capabilities, and we’re seeing an increased percentage of our investment going into those.”

Joseph Domani, a partner at Thomson Reuters Ventures, said his team met with about 50 startups at Collision this week. He said the most interesting companies weren’t those with general AI capabilities, but those that offered the ability to use AI to search across databases and help companies use multiple AI models.

He said Thomson Reuters’ venture arm is likely to invest in some of these companies. Last year, the arm invested in a startup that creates AI summaries for health insurance claims.

“When people say they’re investing in AI, they’re basically just investing in software at this point, which doesn’t make much sense,” Dormani said. “It’s all about the product.”

Write to Belle Lin at belle.lin@wsj.com