The recent selloff in AI stocks has brought a growing investor concern into sharper focus. Companies are pouring eye-watering sums into chips, data centers, and the infrastructure needed to power AI, but there’s a larger question hanging over all that spending: whether it\u2019ll actually pay off.<\/p>\n\n\n\n
AI keeps getting smarter and more capable, but the financial rewards are proving a lot harder to pin down.<\/p>\n\n\n\n
That’s one of the big takeaways from a new white paper from L&G economist Matthew Rodgers. In the research, he digs into the sheer scale of AI investment and the uncertainty surrounding its payoff. And he makes an important point: a technology can transform the economy without every company behind it making a fortune.<\/p>\n\n\n\n
For investors, that means it\u2019s essential to look beyond the hype to see where the cash is going, who’s making money, and whether the profits<\/a> can last.<\/p>\n\n\n\n The amount of money flowing into AI is staggering. L&G estimates that AI infrastructure spending will be equal to about 1.3% of the total US economic output this year. That\u2019s hundreds of billions of dollars going toward data centers, chips, and the electricity to run them.<\/p>\n\n\n\n For perspective, that’s already slightly above the 1.2% reached during the telecom infrastructure boom at the turn of the millennium. It’s still well below the railroad boom of the 19th century, when investment peaked at nearly 5%. But make no mistake: this is one seriously hefty infrastructure spending spree.<\/p>\n\n\n\n Past performance is not indicative of future results.<\/p>\n\n\n\n AI investment recently surpassed peak early-2000s telecom spending but remains well below historical railroad investment. Data as of July 2026. Sources: NBER, FCC\/US Telecom, company filings, L&G.<\/p>\n\n\n\n History comes with a warning, though.<\/p>\n\n\n\n The railway boom transformed transportation, and the telecom boom helped lay the foundations for the modern internet. But neither guaranteed handsome returns for the investors financing them. In some places, companies built far more infrastructure than they needed. Competition drove down prices, and plenty of companies struggled to earn enough to cover what they’d spent.<\/p>\n\n\n\n That doesn’t mean AI is destined for the same fate. In fact, L&G is siding with the optimists, at least for now. AI developers are still paying top dollar for every new chip, and each new generation of AI models seems to bring another leap in what the technology can do. So far, there’s little evidence that the industry’s appetite for computing power is fading.<\/p>\n\n\n\n Still, that doesn\u2019t necessarily mean that strong demand will lead to strong investment returns. Companies could be right about AI’s enormous potential and still end up spending more than they can profitably recover. After all, the ultimate test isn’t how much infrastructure gets built or even how useful it becomes \u2013 it’s whether the companies financing it can earn enough to justify the cost.<\/p>\n\n\n\n And that’s where the story gets particularly interesting, because the profits<\/a> from the AI boom aren’t being shared equally.<\/p>\n\n\n\n There\u2019s an awful lot of money changing hands in AI, but the companies involved aren’t benefiting equally.<\/p>\n\n\n\n L&G’s research highlights an interesting divide between three parts of the industry \u2013 infrastructure suppliers like Nvidia<\/strong> and TSMC<\/strong>, data center and cloud computing providers like Microsoft<\/strong> and Amazon<\/strong>, and AI developers like OpenAI<\/strong> and Anthropic<\/strong>.<\/p>\n\n\n\n And the differences in profitability are striking. Infrastructure suppliers in L&G’s analysis have operating profit margins of around 51%, while the group containing major cloud computing providers has margins of roughly 30%. AI developers, meanwhile, are losing around 45 cents from operations for every dollar of sales<\/a> \u2013 despite their rapidly growing revenue<\/a>. In other words, the companies building the brains behind AI aren’t necessarily the ones making the most money from it.<\/p>\n\n\n\n Past performance is not indicative of future results. <\/p>\n\n\n\n Profitability across the AI supply chain. Infrastructure stocks are Nvidia, TSMC, and ASML. Cloud and data centers are Microsoft, Google, Amazon, and CoreWeave. AI developers are OpenAI, and Anthropic. Sources: Bloomberg, The Information, FT, L&G.<\/p>\n\n\n\n That makes sense for now.<\/p>\n\n\n\n AI developers are spending heavily on research, computing power, and talent as they race to build better products and attract customers. According to L&G, their computing costs alone eat up about three-quarters of their revenue<\/a>.<\/p>\n\n\n\n Of course, today’s winners won’t necessarily be tomorrow’s. If infrastructure spending slows, chipmakers could face weaker demand and greater competition. AI developers have their own challenges, particularly as increasingly capable, freely available models make it harder to charge premium prices. And data center operators could eventually find themselves fighting harder for customers if the industry builds more computing capacity than it needs.<\/p>\n\n\n\n There\u2019s an important caveat here, too. Companies like Microsoft and Amazon rake in profits<\/a> from businesses that have little to do with AI. So their overall profit margins don’t tell us exactly how profitable their AI investments are.<\/p>\n\n\n\n And while spreading your money across chipmakers, cloud providers, and<\/em> AI developers might seem like a sensible way to manage these risks, there\u2019s a potential snag there: those businesses are more closely connected than they might appear. A slowdown in AI spending could hurt several of them at once.<\/p>\n\n\n\n There’s another possibility worth considering.<\/p>\n\n\n\n As AI becomes cheaper and more widely available, some of the biggest winners could ultimately be the companies that use the technology, rather than those that develop it. After all, businesses that can boost productivity<\/a> without spending billions building AI models could be well-positioned to benefit.<\/p>\n\n\n\n And that’s an important insight \u2013 because AI could become an enormous economic success even if some of today’s biggest AI investments disappoint.<\/p>\n\n\n\n You don\u2019t have to be an AI skeptic to be cautious about AI stocks.<\/p>\n\n\n\n L&G remains broadly optimistic that the technology will boost productivity and transform industries over time. And I largely agree. The uncertainty lies in how much of that economic value today’s AI companies will actually capture.<\/p>\n\n\n\n Take the falling cost of AI tools.<\/p>\n\n\n\n That shift is already underway \u2013 L&G’s data shows that average prices for one group of AI models meeting a minimum performance standard fell by more than 80% between early 2025 and mid-2026. As those costs decline, more businesses will adopt AI, improving their productivity and potentially creating entirely new products and services.<\/p>\n\n\n\n Past performance is not indicative of future results. <\/p>\n\n\n\n Average price per million tokens for AI models scoring 20 or more on the Artificial Analysis Intelligence Index. Data as of June 2026. Sources: Artificial Analysis, L&G.<\/p>\n\n\n\n That’s good news for the economy \u2013 but lower prices could also squeeze the profits<\/a> of companies selling those tools. Consumers and businesses using AI might end up capturing much of the benefit instead.<\/p>\n\n\n\n Something similar could happen with data centers, too. If providers build capacity with the expectation that demand will keep soaring, they may eventually find themselves competing harder for customers, putting pressure on what they can charge. The infrastructure could remain enormously useful even if its owners struggle to earn attractive returns.<\/p>\n\n\n\n And that’s why share prices matter just as much as the technology itself. A company could grow rapidly, increase its profits, and still turn out to be a disappointing investment if investors paid a price that assumed even greater success. Conversely, businesses with lasting competitive advantages and more reasonable valuations could continue to reward shareholders as AI spreads.<\/p>\n\n\n\n That\u2019s why investing in AI isn\u2019t simply about identifying the most exciting technology or the fastest-growing company. It\u2019s about finding businesses that can turn the opportunity into lasting profits \u2013 and buying their shares at a reasonable price.<\/p>\n\n\n\n The AI boom still offers plenty of opportunities, but picking the winners will require looking beyond impressive technology and soaring sales<\/a>.<\/p>\n\n\n\n Here are three things worth keeping in mind.<\/p>\n\n\n\nWelcome to the biggest infrastructure boom in decades.<\/h2>\n\n\n\n
![\"AI](\"https:\/\/finimize.com\/_next\/image?url=https%3A%2F%2Ffinimize-img.imgix.net%2Fhttps%253A%252F%252Fchivas-assets.s3-eu-west-1.amazonaws.com%252Fstatic%252Fimages%252FInsight_031126_2_BB.original.png%3Fixlib%3Dpython-3.1.2%26s%3D88e874ec1d792fa7987451af06e6e2bc&w=3840&q=75\")
<\/figure>\n\n\n\nFollow the profits, not just the spending.<\/h2>\n\n\n\n
![\"Profitability](\"https:\/\/finimize.com\/_next\/image?url=https%3A%2F%2Ffinimize-img.imgix.net%2Fhttps%253A%252F%252Fchivas-assets.s3-eu-west-1.amazonaws.com%252Fstatic%252Fimages%252FInsight_031126_1_BB.original.png%3Fixlib%3Dpython-3.1.2%26s%3D9e50d8587ce38487349bc2ff0f3ba5ba&w=3840&q=75\")
<\/figure>\n\n\n\nA winning technology isn’t always a winning investment.<\/h2>\n\n\n\n
![\"Average](\"https:\/\/finimize.com\/_next\/image?url=https%3A%2F%2Ffinimize-img.imgix.net%2Fhttps%253A%252F%252Fchivas-assets.s3-eu-west-1.amazonaws.com%252Fstatic%252Fimages%252FInsight_031126_3_BB.original.png%3Fixlib%3Dpython-3.1.2%26s%3Ddbb621c48c6e6eb3c4164afc4a5529f6&w=3840&q=75\")
<\/figure>\n\n\n\nAnd finally, here\u2019s what this means for investors.<\/h2>\n\n\n\n