The 6 phases of the AI investment boom: Where are we now?

When the capex is as enormous as it has been from the hyperscalers, it’s little wonder that any hiccups are met with an outsized market reaction. But should investors be worried about an AI bubble bursting?

According to Kerry Craig, Global Market Strategist at JP Morgan Asset Management, it’s important to remember that it is still very early in the development of this technology.

“We think there's a long way to go in terms of thinking about the opportunities to invest within this theme,” Craig says.

“We don't think it is a bubble. Are there pockets of things that probably highly value? There's definitely that. So, you have to be very discerning about how you approach some of the companies and the fact that you have this huge capex required to fund that build out means it has to come through in many different areas.”

The AI cycle

JP Morgan frames the AI investment cycle in six phases, drawing on its strategic investment advisory group's analysis of how transformative technologies develop historically. The first phase is the enthusiasm, where a new technology emerges that solves a fundamental economic problem, in this case productivity and ageing demographics. The second is the build-out: the capex cycle that allows the technology to develop and scale.

“That’s where we are at the moment. We’re in that build-out phase because we expect this to continue for a few more years.”

The third phase, or the reality check, is where competition intensifies, winners and losers diverge sharply and active management becomes essential. JP Morgan believes that phase is still some distance away.

“That's where people start to question who are going to be the winners and losers, and that's where you start to see a much more dispersed performance across companies related to this technology,” Craig says.

“It's very difficult to say who wins with everyone just competing to stay in the race, but that will become more obvious as we think about the increasing cost of that capital as bond yields go up.

“They can access capital at the moment fairly easily, but that will also be somewhat of a binding constraint in the future as some companies can't.”

The evidence is the capex pipeline itself: estimates for AI infrastructure spend have risen to $1.2 trillion by 2028, with JP Morgan’s own estimates putting total AI capex investment at $5.5 trillion out to 2030, funded across investment grade debt, equity issuance and private markets.

“The fact that there is many years of capex to go suggests we're still going to be in that build-out phase for a little while," Craig says.

“There will come a point where that $1.2 trillion bar suddenly becomes $1 trillion and then $800 billion and that will be a natural level of capex, which is still very high. But you'll need to also think about how companies can generate revenues to sustain that overall."

The cost of intelligence and the China variable

As Craig points out, questions about revenue generation are being exaggerated by lower-cost competition from Chinese open-weight models.

These models are “pretty sophisticated in their own right”, but much cheaper in terms of their tokens and the cost to perform tasks, creating a pricing floor that complicates the economics for premium model providers.

“It also creates more competition in terms of trying to grab market share because companies will differentiate between the models that are more sophisticated and have a place to accomplish those challenging tasks and those ones that can be done more cheaply,” Craig says.

“You can see this with the OpenRouter data, which shows you the token usage for Chinese models has been going up significantly as people are now making that distinction.”

But he argues the competition is not simply about price, with businesses needing to look beyond finding a model that's smart, complex, and fast enough, but is also secure enough.

“That's the big issue at the moment.”

JP Morgan Asset Management Portfolio Manager, EMAP Equities, Sonal Tanna added that Asian companies are currently capturing approximately 20% of global AI capex spend, with Korea and Taiwan dominant through memory chips and semiconductor production. Crucially, these companies are largely unaffected by which AI model wins.

"Whether it's Anthropic that wins, or OpenAI that wins, or even Meta, these companies are agnostic to who the winner is. As long as AI is going to continue to be a theme in the markets, the Asian supply chain will continue to benefit,” Tanna says.

China's chip localisation

A major point in favour of the US models is that they are operating on the top end chips, which are unavailable in China thanks to sanctions on their sale. However, Tanna says a campaign of chip localisation could change things.

JP Morgan now estimates that within a decade, about 90% of chips used in China will be produced domestically, up from just over 30% today.

"That's a pretty incredible acceleration over the next decade or so," Tanna says.

“Clearly there are restrictions in terms of China getting access to the same technology, so it's been fairly broad-based, particularly in areas like lithography where companies like ASML are not allowed to sell some of their most advanced machinery to China.

“The starting point is obviously low, but … there's a lot of effort on the ground in terms of China trying to catch up or be more creative with the availability of the chips that they have to catch up in some of these key areas to help with semiconductor production.”

The most visible recent example is CXMT, the Chinese memory chipmaker that completed its IPO earlier this year. At launch, CXMT's entire production was booked out by domestic Chinese orders alone, which is a sign of both the scale of internal demand and the speed at which Chinese technology companies are pivoting toward local supply chains.

The diversification opportunity

The central message from both Craig and Tanna was not that investors should reduce AI exposure, but that concentration is now high enough that meaningful diversification within and around the theme is warranted.

"Concentration has been the right call across this. Diversification is the next thing we really think we should look at within portfolios to make sure they're resilient to some of these shocks that could be out there,” Craig says.

“More broadly, there are beneficiaries for that AI coming through in other sectors as well, so there's a concentration risk here. It just means being diversified across asset classes, across geographies, and really trying to think about mitigating some of that risk. I think now you do actually have the opportunity, especially when you look at fixed income, to try and balance things out that don't really have AI exposure to them.”