Surging interest in artificial intelligence (AI) has generated huge gains for tech stocks across the world in 2025. But with the stock market hitting new highs, there are growing concerns that we might be in the midst of an AI bubble. We put four possible flaws in the AI investment case to Terence Tsai, one of the Fidelity portfolio managers who has invested strongly in the boom and therefore one of the people looking hardest for evidence of when it ends.
Not dotcom yet
The numbers on AI investment have become astronomic over the past year, stretching into trillions of dollars. Where does that stop, and do the valuations of those big tech companies investing in AI continue to grow?
Judging by previous investment booms and busts, one canary in the coal mine is what the large tech companies - and their managers - think about the trend. So far, Tsai says, management at hyperscalers like Alphabet or Facebook show no sign of abandoning ship.
“For investors, historically we're known to be the smart money, but I think the real smart money are the people running these large hyperscalers,” says Tsai.
“In the 2000s, the management of these companies were trying to get out - they were going public, they were trying to sell their shares. [But] over the last 12 months, there have been over a trillion dollars in buybacks for the US market. So if there is a bubble, their actions certainly do not support that argument.”
Making AI profitable
A second batch of worries concerns the monetisation of AI. Tech firms are spending hundreds of billions of dollars on advanced chips and datacentres, but is there evidence of how will they make money off the new technology? Tsai says one positive trend is that large platforms have already seen revenues from their cloud businesses accelerate hard on the back of the broader changes in industry structure.
“Before the ChatGPT moment, most of these hyperscalers’ cloud revenues were growing at single digits, and now they're all growing at 30 per cent, 40 per cent,” he says.
Still, Tsai says investors should monitor the monetisation path to make sure the thesis is well on track.
Supply chokepoints
Chip makers are clearly one of the biggest beneficiaries of the AI boom, but critics have made a number of arguments about current industry valuations: that companies have artificially inflated their numbers with changes in depreciation, or that sales and investments are circular and too centered around Nvidia.
Tsai, originally a semiconductor sector analyst by trade, has been searching the industry for signs of weakness. Right now, they are hard to find.
“The downcycle normally starts when you have a lot of inventory in the system,” he says. “At the moment we have no inventory. Whatever is produced by TSMC goes straight to Nvidia. [And then] they go straight to the datacentres to flip the switch. We’re going from hand to mouth right now.”
That would suggest the top of the cycle is still a long way off.
“If April this year was the bottom of the last cycle, we are in maybe the early innings of a new [semiconductor] upcycle,” Tsai says.
Software slump
While the market sees hardware and infrastructure as “picks and shovels” powering the AI gold rush, there are also growing concerns that this disruptive technology is going to hurt other companies, particularly software and IT services firms, whose businesses can be replicated with AI tools.
“I don't think they are going to be big losers across the board,” says Tsai, referring to software and IT services companies. “This is where active stock picking matters. Identifying the companies that are probably going to survive and come out stronger at the end of the tunnel would be key.”
You can read more on the software industry’s issues - and opportunities - with AI from Tsai’s colleague Ashish Kochar here.
China breakthroughs
Tsai has also been bullish on China since the DeepSeek moment in January 2025. Despite limited access to the best AI chips, Chinese tech companies have been able to work within the constraints and develop models approaching the capabilities of leading systems from the US, but with much less capex.
“China focuses on a different philosophy on AI,” says Tsai. “It's about mass adoption. How can I get the cost low so that more people would use it?”
Because of the lower costs, “the proliferation of AI in terms of adoption in China has grown much more quickly than in the US,” says Tsai.
China’s vast pool of talent is crucial in advancing its own AI ecosystem.
AI is “all about talent and China has a lot of that talent,” says Tsai. Still, hampered by chip supply issues, China trails the US in the global AI race.
“China lacks compute, [and] the US lacks power,” says Tsai. “It'll be interesting to see which one is able to [best] work around its constraints.”
