There’s a curious paradox at the center of technology’s biggest story right now. The GPUs (graphics processing units) and other essential hardware that tech giants are spending so lavishly on to equip their data centers are quickly becoming obsolete. That’s the view detailed in an excellent new report from Research Affiliates, a firm that oversees about $200 billion in investment strategies for its RAFI index funds and ETFs.
Author Chris Brightman — RA partner and senior consultant — argues that the race for AI has, in effect, created a new industrial era. In this transformed ecosystem, companies are not “investing” in the traditional sense. Instead, they are turning equipment at such an incredibly fast pace to generate revenue that it is changing the very definition of capital expenditure.
Also read:
Continues after advertising
“They look more like supermarkets than traditional technology or industrial companies, but their pivot isn’t in things like groceries. It’s what drives their large language models, vector search and other products,” Brightman said in a phone interview.
“They are in a race where they need to replace their hardware very quickly, i.e. restock their shelves in a hurry.” The problem, says Brightman, is that the giants are making losses on the language models, vector databases and other products they sell to businesses and consumers; So, the more hardware they buy, the more money they lose.
“Right now, everyone uses AI to maintain crucial dominance in their field, and that makes sense,” notes Brightman. But, he adds, the immense expense required to maintain these competitive advantages and fend off rivals can generate negligible returns in the future and harm your overall profitability.
In the article, Brightman highlights the historic rise in capital expenditure (capex) on AI, which jumped from US$250 billion in 2024 to US$650 billion this year, according to a Bloomberg estimate, equivalent to 2% of GDP.
This historic appetite for capital has generated the view that AI is becoming the new steel or the new railroads. But as Brightman points out, the equipment and infrastructure that underpinned these businesses are very different from those that power AI.
“Steel plants and railroad tracks depreciated over 40 to 45 years,” he writes. He then contrasts these decades-long lifespans with the AI scenario. Companies like Microsoft, Amazon, Alphabet and Meta are depreciating their GPUs and other hardware over the course of about five or six years in their income statements. Although these deadlines already seem very short, he claims that their real “lives” are still much shorter.
Continues after advertising
In economic terms, assets become fully depreciated, or obsolete, when the revenue they generate no longer covers their acquisition cost (reflected in annual depreciation), operating expenses and the cost of capital. According to Brightman, industry figures show that AI hardware loses value in about three years.
As proof, he cites data on the profitability of Nvidia’s industry-standard H100 GPUs. In the second year, an H100 generated US$36,000 in annual profit, with a return on investment of 137%.
But, in the fourth year, the product recorded a loss of more than US$4,400, with a negative ROI of 34%, and results deteriorated quickly from there. Writes Brightman: “The economic life of AI hardware is much shorter than its accounting life.”
Continues after advertising
It’s not that the equipment wears out. Physically, it can function much longer. The reason AI hardware is losing power so quickly is that Nvidia, AMD, and other manufacturers are releasing new offerings that each year deliver huge increases in computing power per watt used.
As companies face severe energy constraints, they are constantly seeking large volumes of new processing power with additional doses of electricity.
Normally, if traditional manufacturers were adding capital at the rate that the giants are pouring into AI, they would have already built a gigantic base of equipment and infrastructure that they could use for years, without needing to keep buying more.
Continues after advertising
That is not the case in this new business model. AI equipment evolves so quickly that every year, companies need to replace a huge portion of their capital base just to maintain the same ability to drive AI innovation.
“Most of the spending is not growth capex, it is ‘maintenance’ capex,” says Brightman. Still, the total numbers are so large that even though only about a third goes to expansion, this is enough to massively increase the volume of products and services they can offer each year.
Companies are accumulating large losses
In our phone conversations, Brightman explained the AI giants’ dilemma. “As they expand their processing power, they lose more and more money,” he says. “But for now, they have enough reasons to do so.”
Continues after advertising
All of the Big Four seek to offer the best AI capabilities to enhance their core products and recognize that they will lose leadership in these segments if the AI component is not cutting-edge.
Amazon makes most of its money by providing cloud computing and storage. According to Brightman, it cannot recover nearly the cost of AI additions from customers. “But it makes sense, because if Amazon doesn’t stay in the race, it will lose the cloud business. It needs AI services as part of that component.”
As for Microsoft, its main product is office software that generates subscription revenue, especially on its 360 platform. This franchise now faces stiff competition from Google’s Docs and Sheets products.
“To protect its existing business and retain its customers, Microsoft needs to offer AI model services, even if it is losing money on its AI capex,” says Brightman.
Alphabet is dominant in search and leads as the world’s largest online ad seller. Microsoft threw down a challenge by creating its own search engine. “To continue its profitable line of business and maintain its edge, Alphabet needs the AI component, and that requires large investments in data centers,” says Brightman.
Meta has to worry about the other three encroaching on its highly profitable social media advertising business.
“People go to the platform to look at photos and videos, and it costs Meta a lot of money to produce this content that supports the ads,” notes Brightman. Meta uses AI to personalize feeds, classify content on Instagram and Facebook, and check the security of posts, and it needs these uses to maintain its lead.
Yet once again, Brightman says, it still can’t charge enough for its ads to pay for the huge new spending needed to offer these features.
Brightman concludes that the flood of investments in AI does not mean that this revolutionary advancement will be highly profitable for Big Techs. It is more of a weapon for each giant to defend their domain.
“When capital turns over quickly and competition forces continuous reinvestment, extraordinary spending can sustain competitive position without creating value for shareholders,” he states in the article.
Again, the lifespan of what’s filling our data centers is so short that buying GPUs, for example, feels more like restocking grocery stores than building factories that last decades.
AWS CEO Andy Jassy has a very different view. In its most recent annual letter to shareholders, it stated that AWS chips, servers and networking equipment have a useful life of 5 to 6 years.
On the other hand, Brightman told me that the resources that are costing these champions a lot helped him prepare his analysis. “A year ago, this project would have taken nine months to research and model. But I used the best of Claude, ChatGPT and Gemini, synthesized their feedback and did everything from start to finish in three weeks”, he reports.
Brightman’s little story illustrates the point: This new industrial era could be far more beneficial to the people and companies that use AI-enhanced products than to those that provide them.
2026 Fortune Media IP Limited
