Lars Tvede: The AI bubble will come sooner or later, but it's not here yet

Article body

On October 21, the second day of the 2026 Sustainable Global Leaders Conference, Lars Tvede, a Danish financial investor and author of "The Economic Cycle That Can't Be Escaped," was interviewed by a 21st Century Business Herald reporter on the sidelines of the conference.

When asked about these two questions, Tvede said frankly: "If you look back at the entire history of general-purpose technologies, almost without exception they have all formed a bubble. So when the AI era arrives, if we don't even have one AI bubble, I would actually be very surprised." He said, "But I don't think we are in a bubble right now."

Using the peak of the 2000 internet bubble as a reference point, Tvede believes that we are still very far from an AI bubble.

The first reason he cited lies in valuations. "At that time there were iconic companies like Cisco. When valuations peaked in March 2000, its price-to-earnings ratio, that is, the ratio of its share price to earnings over the past 12 months, was 135 times; on forward earnings, the market consensus was about 200 times." Tvede said investors at the time were willing to buy at a valuation of 200 times forward earnings.

Today the company playing a similar role in the industrial ecosystem is Nvidia, and Tvede puts its forward price-to-earnings ratio at around 18 times. "Comparing Cisco during the internet bubble with Nvidia today, valuations back then were about seven times what they are now."

Tvede's criterion for assessing whether an industry has produced a bubble is whether there has been excessive construction of capacity. He pointed out that economics calls this "competitive overinvestment." During the internet bubble it was everywhere: people laid "dark fiber" far in excess of demand, and there was no traffic in the pipes.

He pointed out that bubbles usually begin with demand peaking. Tvede noted that what currently constrains AI development lies upstream: chip shortages, queues for grid connection, and lengthening lead times for bringing data centers online. In an earlier exclusive interview with this newspaper, he also mentioned that in the United States more than 2,300 gigawatts of data center and power projects are queued for grid connection, close to twice existing generating capacity.

Therefore, regarding market concerns about the valuations of South Korean high-bandwidth memory chip companies, Tvede is also unconcerned. "If you look at consensus earnings expectations for 2026, 2027, and 2028, they correspond to a price-to-earnings range of 3 to 8 times. That is very far from a bubble."

No need to overly worry about circular financing between shovel sellers and pick sellers

The third market concern is the systemic risk that circular financing may bring, namely chipmakers investing in model customers, who then turn around and buy chips, with revenue circulating within a closed system.

"A long time ago, gold was discovered in the Klondike in Canada. There was a saying during the gold rush, often repeated in marketing classes, that the people who made the most money were the ones selling shovels." Tvede believes that digging for gold requires shovels and also picks, and neither can be missing. The shovel seller makes $4 on each sale, while the pick seller loses $1 on each sale, so he stops selling. But if no one sells picks, no one digs for gold, and the shovel seller's business cannot be done either.

"So the shovel seller will say: how about this, I lend you the money first, so that you can sell picks, and then someone will go prospecting for gold."

This is similar to the current situation in the AI industry. Tvede estimates that chipmakers make $260 billion to $340 billion in profit on the AI chip side; the frontier model side loses about $50 billion to $60 billion.

"For the side with money, doing this makes complete sense. I must help you build the model, otherwise I can't sell anything. Only after you sell the model can my chips be sold." Tvede believes this process will continue for a while, and during that time some institutions will certainly run into risk, but overall doing this is "quite rational," and "on the whole I am not that worried."

At the same time, he cited a recent article in The Economist as supporting evidence. The article reviewed Nvidia's role in circular financing and concluded that it is not particularly dangerous because Nvidia has ample cash on hand.

Chipmakers' prices are currently "absurdly low"

Tvede has publicly pointed out many times that the AI industry is still in the early stage of a long cycle. As for where the most certain investment opportunities in AI will be in the future, Tvede said the answer will change over time, first benefiting those who build the technology, then the application layer of the technology, and finally consumers.

"At the beginning, the people making money were those building the technology itself, such as early Cisco, which made internet routers. Later it was companies at the application layer like Amazon and Alibaba. Finally it is consumers, who enjoy cheaper goods and services."

Applied to the present, Tvede's ordering is three layers: chips, applications, consumers. "In the next three to five years, I think chipmakers are very much worth watching. In my view, their prices are absurdly low." As value shifts more toward the application layer, the next to take over are users in biopharmaceuticals, media, finance, and insurance. "Companies that do not apply AI will pay the price for it, because they will lose competitiveness."

Finally, Tvede believes that over the long term wealth may no longer concentrate among a small number of people.

"The internet democratized access to information, and people could exchange information truly quickly, while AI democratizes access to skills." He pointed out that in the future, with just a smartphone, everyone will have the opportunity to access superintelligence at the level of top university professors in places like Shanghai or Boston. "Although not everyone will apply this capability, it distributes capability in an unprecedented way."

Tvede mentioned that the "AI+" initiative proposed in China's 15th Five-Year Plan includes education as an important proposition. He believes that China's huge base of technical talent gives it, among all countries, the best conditions to extend AI education down to middle schools and enable people to "create AI" rather than merely "use AI," which can put China at the forefront of AI competition.

(Source: 21st Century Business Herald) Middle of the body: embedded advertisement