- AI is advancing so fast that billions invested in its infrastructure could be overtaken before they pay off.
- Amazon has already shortened the estimated lifespan of some equipment, citing rapid technological advances.
- Experts warn that AI's biggest financial winners may not be the companies bankrolling its buildout.
Investors are pouring billions into AI infrastructure, but the technology is evolving so quickly that today's investments could be overtaken before they pay off.
"I think without question, the technology will be wildly successful. The technology will be wildly adopted, but it doesn't mean the investment returns will be what people expect," Brent Tomlinson, president of Risk Advisory at global financial advisory firm Kroll, told Business Insider.
A key challenge is timing. Data centers can take years to build, while advances in computing technology could erode the commercial value of the hardware inside them.
"Investors are financing assets with very different economic lives," Volker Pfirsching, a partner at management consultancy Arthur D. Little, told Business Insider.
"Buildings and power connections can support several generations of technology, while computing hardware can lose commercial value much sooner," he said.
For investors, the question is how often that hardware will need replacing and how much money it can earn before then.
The financial implications are already showing up in companies' accounting decisions.
Amazon shortened the estimated useful lives of a subset of its servers and networking equipment from six years to five years in 2025, citing faster technological development, particularly in AI and machine learning.
The accounting change increased depreciation and amortization expenses by $1.4 billion and reduced net income by $1 billion that year, primarily affecting AWS, according to the company's annual report.
The cost of betting on AI
The sums at stake are staggering.
Amazon, Microsoft, Alphabet, and Meta were planning to spend up to $725 billion in combined capital expenditure in 2026.
Financing that expansion is already making investors nervous. Shares of AI infrastructure providers CoreWeave and Nebius tumbled in October as expectations of higher interest rates weighed on the sector.
Beyond financing costs, there's also uncertainty over how much infrastructure the industry will ultimately need.
Goldman Sachs CEO David Solomon has questioned the eventual scale of the buildout.
"Whether we're going to spend $5 trillion on the buildout or $10 trillion on the buildout or more, I don't think anybody really knows," Solomon said in the bank's "Exchanges" podcast published on October 27.
Solomon remains optimistic about AI's long-term potential and said demand for computing power currently exceeds supply — but there are still questions about how much data center capacity the industry will need.
A race against time
The difficulty is that infrastructure decisions made years ago are still playing out as AI technology races ahead.
Tomlinson said data center projects planned two years ago are still being built, even as AI technology continues to evolve.
"The pace and timing of technology change versus the pace and timing of CapEx investment and infrastructure investment is so different that it's hard to fully understand how those emerge at any given point in time," he said.
"I think overbuilding can happen even while the technology keeps improving," he added.
However, more efficient AI could fuel demand rather than undermine it.
Pfirsching said cheaper, more efficient AI could encourage wider adoption, potentially increasing overall computing demand even as models become less resource-intensive.
"For infrastructure investors, returns depend on whether growth in paid usage compensates for falling prices and covers electricity, financing and equipment replacement," he said.
Who really wins?
That raises the stakes for investors trying to pick the right parts of the AI boom.
Pfirsching sees opportunities for infrastructure providers with secured power supplies, creditworthy customers, and contracts that support their financing commitments.
But some of AI's biggest financial winners may not be infrastructure companies at all.
In an October report, State Street Investment Management drew parallels with the fiber-optic expansion of the late 1990s.
Those networks became essential to the internet economy, but some companies that financed their construction struggled financially.
AI infrastructure could follow a similar pattern, with businesses using increasingly affordable computing power capturing more of the economic value, the State Street strategists wrote.
Tomlinson said investors are becoming more discerning, assessing different parts of the AI ecosystem separately rather than treating the industry as a single investment opportunity.
That's particularly important because many AI-related businesses need to keep spending on infrastructure and research and development, he said, even as investors become more selective and higher interest rates push up their return expectations.
Pfirsching said investors should also consider the possibility of slower customer adoption, falling rental prices, and earlier hardware replacement.
"The greatest vulnerability is borrowing heavily to buy hardware, relying on a handful of customers and assuming today's rental prices will last," he said.
The post The race to build AI's future could leave investors holding yesterday's bets appeared first on Business Insider
