Wall Street is missing an AI cost that could hit Google, Meta, Amazon

Wall Street is fixated on how much artificial intelligence is boosting Big Tech’s revenue, but it is still underestimating a quieter threat sitting on the balance sheet. The next hit to earnings may not come from slowing demand for chatbots, but from how quickly the physical infrastructure behind those models wears out and has to be written down.

As Google, Meta and Amazon race to dominate generative AI, they are pouring unprecedented sums into data centers, chips and power, while stretching accounting assumptions to make the profits look smoother than the underlying economics. I see a growing risk that these choices will collide with reality, forcing a painful reset in valuations when the true cost of keeping the AI engine running finally shows up in depreciation and cash flow.

The depreciation time bomb behind AI data centers

The core problem is simple: the useful life of AI hardware is shrinking just as companies are extending it on paper. On Nov 26, 2025, one analysis of Michael Burry’s warnings described how “When useful lives are extended while technological obsolescence accelerates, every extra year delays roughly $30 billi,” a blunt way of saying that stretching depreciation schedules flatters current earnings at the expense of future ones, and that the adjustment “will snap back hard tomorrow” when the assets have to be written down, a point that directly affects how investors should view the AI buildout at Google, Meta and Amazon Nov 26, 2025. If the servers and accelerators that power large language models become obsolete in three or four years, but the companies depreciate them over five or more, the gap between economic reality and reported profit widens with every quarter.

That is exactly the blind spot Bank of America highlighted on Nov 12, 2025, when it warned that Google, Meta and Amazon’s massive AI spending could trigger unexpected depreciation costs that Wall Street has not fully calculated, a concern that goes beyond headline capex and straight into the earnings models that underpin trillion dollar market caps Bank of America. If those models assume a gentle, linear expense profile, they are likely missing the cliff that arrives when a generation of AI chips must be replaced all at once, compressing margins just as growth expectations are highest.

A $600 Billion and 250 Billion infrastructure bet

Behind the accounting choices sits a staggering physical buildout that is still accelerating. In 2025, one detailed breakdown of the AI race described how “In 2025, te” largest technology platforms embarked on what it called “The Hidden Cost of AI: How Big Tech’s $600 Billion Gamble Is Reshaping America,” arguing that this $600 Billion wave of spending is transforming everything from industrial parks to local labor markets, and that “The Numbers Don” actually “Add Up” once you factor in the full lifecycle cost of the hardware and energy involved $600 Billion. That same analysis, published on Oct 20, 2025, underlined how “Oct,” “The Hidden Cost of AI,” “How Big Tech,” and the “Billion Gamble Is Reshaping America” are not abstractions but concrete factories, substations and server halls that will need to be refreshed long before their concrete foundations wear out.

Another assessment on Nov 25, 2025, framed the issue as “Understanding the 250 Billion Dollar Question Behind Big Tech Artificial Intelligence Infrastructure Spending,” arguing that the real strategic challenge is not whether companies can raise the money, but whether the returns will justify a 250 Billion outlay on assets that may have to be replaced faster than any previous generation of computing gear Understanding the. When I look at Google, Meta and Amazon, I see that same 250 Billion scale of commitment embedded in their data center roadmaps, and I see how little room there is for error if AI revenue growth falls even slightly short of the heroic assumptions now baked into their share prices.

Creative financing and the AI bubble risk

As the sums involved balloon, the financing structures are getting more complex, which makes it even easier for investors to miss the underlying cost. On Nov 19, 2025, one column described the AI buildout as “Fast and Furious” and pointed to “Exhibit A” in Meta Platforms Inc’s massive data center project in rural Louisiana, which is being financed off balance sheet through bonds that shift some risk away from the company’s headline debt metrics but do nothing to change the economic reality that the facility must eventually earn its keep Fast and Furious. I see similar patterns emerging across the sector, with joint ventures, leasebacks and structured deals that spread the capital burden but also obscure the true depreciation profile of the assets involved.

Those structures sit on top of what one Nov 22, 2025, report called “Enormous” spending that hinges on returns that could be a fantasy, warning that Silicon Valley is taking on new debt with the assumption that AI data centers will stay full and profitable for years to come, even though the revenue models are still unproven Enormous. In the same Nov 22, 2025, coverage, the design of one marquee deal was described as Blue Owl taking out a loan for $27 billion for the data center, with repayment tied to the future value of the data center itself, a structure that amplifies the risk if AI demand or pricing disappoints Blue Owl. When I connect those dots, I see not just a financing story, but a setup where any downward revision in asset values could cascade through both earnings and credit markets.

The power and operating costs Wall Street is glossing over

Even if the hardware lasted forever, the operating costs of AI data centers are rising in ways that are hard to square with the smooth margin trajectories in many analyst models. On Aug 13, 2025, a detailed look at the grid impact of AI reported that “New” data centers require utilities to spend billions of dollars on power lines and plants, and that these facilities are already consuming a significant share of all new electricity generated since 2022, a trend that will eventually flow back into higher power prices for both operators and local residents New. When I look at hyperscale campuses in places like central Ohio or northern Virginia, I see not just server racks but long term commitments to pay for upgraded substations, transmission lines and backup generation, all of which will show up as higher operating expenses over time.

Those costs are being layered on top of already aggressive capital plans. One Nov 25, 2025, earnings update described “Heavy” duty spending on AI data centers and projected that the global data center market will reach $939 billion by 2028, up from roughly half that level today, with much of the incremental growth driven by AI workloads rather than traditional cloud hosting $939 billion. If the cost of electricity and cooling rises faster than expected, the operating leverage that investors are counting on for Google, Meta and Amazon could instead turn into a drag, especially in regions where regulators or communities push back on subsidizing power for AI at the expense of other users.

Why Google, Meta and Amazon are especially exposed

Not every tech company is equally vulnerable to this hidden AI cost, but the trio of Google, Meta and Amazon sits at the center of almost every risk vector. A Nov 26, 2025, analysis of the sector’s capital intensity estimated that Meta, Amazon, Microsoft, Google and Tesla will by the end of this year have collectively spent $560 billion on AI related infrastructure, a figure that captures both the scale of the bet and the concentration of exposure in a handful of names $560. When I break that down, I see Google leaning heavily on its search and YouTube franchises to absorb AI costs, Meta banking on advertising and virtual reality to justify its data center push, and Amazon threading AI into everything from its cloud to its retail logistics, including the infrastructure behind its flagship marketplace at Amazon.

Each of those strategies assumes that AI will either unlock new revenue streams or at least deepen existing ones enough to cover the depreciation, financing and power bills that are now locked in. Yet the same Nov 22, 2025, reporting that flagged Silicon Valley’s “Enormous” spending also raised the possibility that AI revenues could fall short of the hype, leaving companies with underutilized facilities and a need to accelerate write downs just as growth slows Silicon Valley. When I put that alongside the 250 Billion and $600 Billion figures already committed, and the specific warning from Bank of America about uncounted depreciation at Google, Meta and Amazon, the conclusion is hard to avoid: the market is still pricing these companies as if AI infrastructure is a one time investment, when in reality it is a recurring, accelerating cost that could hit earnings much harder than most models currently allow.

More From Alpine Times