VCs who bet US$300 billion on AI say DeepSeek will fuel industry
The Chinese company’s success will galvanise US startups and put them under pressure to quickly catch up with its advances
VENTURE capitalists (VCs) have sunk more than US$300 billion into US artificial intelligence (AI) startups in the past five years – a massive outlay that was called into question this week by the release of the new cheap and efficient model from Chinese AI company DeepSeek.
DeepSeek has already roiled the public markets, with investors sending the Nasdaq down 3 per cent on Monday (Jan 27) – the equivalent of a US$1 trillion haircut. And because so many Silicon Valley VCs have gone all-in on AI, as indicated by PitchBook data, private markets could be even more exposed.
But venture capitalists at firms including Battery Ventures, Andreessen Horowitz and Menlo Ventures are unfazed.
Motivation
Instead, they are saying that the rise of a formidable Chinese rival could serve as motivation for US tech companies and lead to faster AI adoption – with many emphasising the opportunity represented by potential breakthroughs in efficiency promised by the new model.
Investors fleeing tech stocks are “unduly panicking”, said Y Combinator’s Garry Tan. “Training costs coming down and capabilities going up will create demand, not reduce it.”
In particular, startups making cloud-based software for companies could reap big gains from any decrease in the cost of AI computing, he added.
BT in your inbox
Start and end each day with the latest news stories and analyses delivered straight to your inbox.
The effect of DeepSeek will be more startups than ever “going from zero to US$10 million per year in recurring revenue”, he wrote in a post on X. “A thousand flowers will bloom!”
Global AI researchers will try to quickly replicate DeepSeek’s efficiency gains, said Kyle Harrison, a general partner at Contrary Ventures.
Open-source models and scientific progress “pushes everything forward”, he said. “You can bet that every company from OpenAI to Mistral is going to take those approaches and leverage them on their own products. So that’s good.”
Some AI companies may be overvalued, Harrison added, but Monday’s public market wipeout “is completely irrational and borderline stupid”.
Several investors said that DeepSeek’s success would galvanise US startups. “Competition that makes everyone better,” said Pelion Venture Partners partner Tyler Hogge.
While he expressed concern about the role the Chinese government may have in the company, he believes the technology will mean “there are many more businesses that can be built because of lower costs”.
Besides a rallying cry, DeepSeek also offers a warning to US tech companies, said Venky Ganesan, a partner at Menlo Ventures.
“It’s a true wake-up call for the US in terms of the race to win AI,” said Ganesan, whose firm has invested in Anthropic. He compared the advance to the Russian satellite that spurred the US to race ahead with its space programme. “DeepSeek is the Sputnik moment for AI,” he noted.
Of the billions that VCs have poured into AI startups, much has gone to companies such as Anthropic, OpenAI and Elon Musk’s xAI, which are building so-called “foundation models”.
These serve as the technological bedrock for other startups, and have been fabulously expensive to build. The companies now find themselves on the defensive, under pressure to quickly catch up with the advances seemingly made by DeepSeek.
More than US$40 billion in 2023 and 2024 went to investments in foundation model businesses, noted PitchBook.
The latest news could offer vindication to investors who chose to mostly sit out those large funding rounds, and instead focused on startups that use AI for more specific applications.
For example, legal company Harvey, compliance startup Vanta and health businesses such as cancer-detection startup Freenome all rely on AI, and would stand to benefit if costs decline.
Dharmesh Thakker at Battery Ventures has largely eschewed bets on the biggest companies. In late 2022, he made an analogy with the early Internet era, comparing foundation models to the fibre-optic companies of the late 1990s, and saying those did not emerge “as the big winners”.
Economic value
Instead, it was Amazon, with its hugely popular shopping site, Facebook and Instagram – with their viral consumer services – that benefited and grew into hugely valuable companies.
This week, Thakker said that while he had some reservations over whether DeepSeek’s costs were as low as widely portrayed, its results were important.
The efficiency gains show that the notion of building bigger compute for AI “may be great for marketing, but ultimately doesn’t drive economic value beyond a certain point”, he wrote in an e-mail.
Ultimately, the usefulness of AI tools will make or break the companies that build them, said Andreessen Horowitz general partner Vijay Pande.
The products are “how you can differentiate yourself, and that’s always going to be in the apps”, rather than the infrastructure. He called the emergence of DeepSeek “an inspiration” and also “a shot across the bow”.
Copyright SPH Media. All rights reserved.
