Will AI Doom Fears End the AI Boom? BCA’s $10 Trillion Warning
The debate over artificial intelligence has entered a new phase. After years of enthusiasm around rapidly improving models, massive data-center construction and soaring AI investment, concerns about the technology's safety are increasingly being discussed alongside a more traditional financial question: Can the enormous investment in AI infrastructure eventually generate enough revenue and profits to justify the spending?
BCA Research Chief Economist Peter Berezin argues that fears about AI's potential dangers are only part of the reason some major AI companies are becoming more cautious. In a report published on September 15, Berezin said concerns that the industry is locked in a financially unsustainable race for AI dominance may be playing as much of a role in the emerging "go slow" approach as worries about AI's long-term risks.
That distinction matters for investors. A slowdown driven by safety concerns would look different from one caused by weak economics. And according to BCA's analysis, the financial challenge may ultimately be just as important as the "AI doom" debate.
BCA's Central Warning: AI May Need $10 Trillion in Annual Revenue
BCA's analysis focuses on the extraordinary amount of capital being committed to AI data centers and computing infrastructure.
The research estimates that major hyperscalers' annual depreciation charges could rise from about $255 billion in 2026 to $581 billion by 2029, as capital expenditure reaches approximately $1.16 trillion. Including spending outside their balance sheets, BCA estimates annual hyperscaler capex could approach $1.4 trillion toward the end of the decade.
Depreciation is an accounting expense that spreads the cost of equipment such as servers and data-center infrastructure over its useful life. Because companies do not record the entire cost of a capital asset as an expense immediately, today's heavy investment can have a larger effect on reported expenses in future years as depreciation increases.
That creates an important test for the AI boom: the infrastructure being built today eventually needs to generate sufficient economic returns.
BCA estimates that hyperscalers alone could need roughly $7.4 trillion in annual revenue to profitably support steady-state capital spending under assumptions of a 15% pre-tax return on invested capital and a 30% EBITDA margin.
Once AI spending by non-hyperscalers, Chinese companies, neocloud providers and private companies is included, BCA puts the potential global revenue requirement at around $10 trillion annually.
That is an enormous hurdle. BCA notes that $10 trillion is roughly comparable to annual global spending on healthcare or food.
The Problem Is Not Simply AI Safety
The phrase "AI doom" generally refers to concerns that increasingly capable AI systems could create severe social, economic or even existential risks.
Those concerns have become more prominent in September as some AI researchers and industry leaders have called for a slower and more controlled approach to frontier AI development. Reuters has reported on growing debate around slowing AI development, while Anthropic CEO Dario Amodei has publicly argued for greater caution around rapidly advancing AI capabilities.
But BCA's Peter Berezin highlights another explanation.
AI companies may have economic reasons to become more cautious because the infrastructure race is becoming extremely expensive.
In other words, a company can believe that AI is strategically important while simultaneously questioning whether every additional dollar spent on computing capacity will generate an adequate financial return.
That is a very different issue from whether AI technology itself works.
Why Current Profit Margins May Not Tell the Whole Story
BCA also raises concerns about how the AI investment cycle is affecting reported corporate profitability.
The S&P 500 was trading at roughly 19 times forward earnings in the analysis, broadly around its 10-year historical average. However, BCA noted that this valuation relies on forward profit margins of about 16.7%, which are historically high.
If margins returned to 2019 levels, BCA calculated that the implied forward price-to-earnings ratio would rise to about 26.7 times.
One factor behind this issue is accounting treatment.
When Microsoft, Amazon, Alphabet, Meta and Oracle purchase large amounts of AI hardware, the spending is generally recorded as capital expenditure rather than being immediately deducted as an operating expense. The cost is then recognized over time through depreciation.
That can make current earnings look stronger than they might if all infrastructure spending were treated as an immediate expense.
This does not mean the accounting is improper. Capital assets are normally depreciated over their useful lives. The investment question is different: will the assets produce enough future cash flow to justify their cost?
The Hyperscalers Are Building at an Unprecedented Scale
The financial challenge becomes clearer when the size of the AI infrastructure cycle is considered.
Microsoft, Amazon, Alphabet, Meta and Oracle are investing heavily in data centers, GPUs, networking infrastructure and energy capacity.
A separate September analysis from UBP estimated that the five major hyperscalers could spend around $820 billion in 2026, with projected spending potentially reaching $1 trillion to $1.3 trillion in 2027. It also estimated cumulative investment could reach $5.6 trillion by 2030.
Meanwhile, the Bank for International Settlements has warned that the AI investment boom is becoming large enough to create financial-stability risks. The BIS said the world's five largest technology companies were expected to invest more than $1 trillion in AI across 2025 and 2026.
The financing side is therefore becoming an important part of the story.
If companies increasingly rely on debt or external financing to fund AI infrastructure, the cost of capital and investor appetite can become just as important as demand for GPUs.
Could an AI Investment Slowdown End the AI Boom?
Not necessarily.
A reduction in capital spending would not automatically mean AI adoption has failed.
In fact, technological investment cycles often contain periods of overbuilding. Companies may build too much capacity initially, leading to lower prices and cheaper access to infrastructure. That can eventually make the technology more accessible to a broader range of businesses.
The more immediate risk for investors is therefore not necessarily "AI disappears."
It is that AI infrastructure spending grows faster than monetization.
If revenue from AI products, cloud services, enterprise software and other applications fails to keep pace with infrastructure investment, returns on capital could deteriorate.
That could put pressure on valuations even if AI usage continues to grow.
Nvidia and AI Infrastructure Stocks Face a Different Test
For semiconductor and infrastructure companies, the key issue is whether AI demand remains strong enough to support the enormous spending plans of hyperscalers.
Companies supplying GPUs, networking equipment, memory, power systems and data-center infrastructure can benefit from continued investment even before end users generate massive AI profits.
But that creates a second-order risk.
If hyperscalers eventually decide that additional computing capacity is producing inadequate returns, they could slow new orders. Such a change would affect the broader AI supply chain.
That is why investors should distinguish between AI adoption and AI infrastructure spending. The first could continue rising even if the second eventually grows more slowly.
What About the "AI Doom" Argument?
BCA's research does not establish that AI will or will not create catastrophic risks. Those are questions involving technology, governance and security rather than conventional market forecasting.
What the research does highlight is that the industry faces two separate forms of uncertainty.
The first is technological and societal: how quickly AI capabilities advance and whether safety measures can keep pace.
The second is financial: whether the enormous capital being invested today can earn adequate returns over time.
The two issues could also interact.
More cautious AI development could reduce the pace of infrastructure deployment. But greater spending on safety, monitoring and evaluation could also create additional demand for computing resources. The economic effect therefore depends on how AI companies choose to manage the trade-off between capability development, safety and capital intensity.
What Investors Should Watch Next
For investors tracking the AI boom, several indicators may be more useful than headlines about whether AI is in a "boom" or "bubble."
Watch:
Hyperscaler capital-expenditure guidance
AI-related revenue growth
Cloud utilization and pricing
Data-center occupancy and power availability
GPU and networking demand
Depreciation expense growth
Free cash flow after capital expenditure
Debt issuance by major technology companies
Returns on invested capital
Enterprise adoption and willingness to pay for AI services
The most important metric over time may be the relationship between incremental AI investment and incremental revenue.
If every additional dollar of infrastructure produces substantial new cash flow, the investment cycle can remain economically productive. If returns decline sharply as spending scales, the market could become more sensitive to the size and pace of capital expenditure.
The Bottom Line
BCA Research's latest analysis adds a financial dimension to the growing debate over AI risks. Peter Berezin argues that the industry's emerging caution may reflect not only concerns about AI's potential dangers but also worries that the current infrastructure race could become financially unsustainable.
BCA estimates that the broader AI industry could ultimately need around $10 trillion in annual revenue to justify the scale of infrastructure investment under its assumptions. That is not a forecast that AI will generate $10 trillion, but an illustration of the enormous economic hurdle created by today's spending plans.
For investors, the central question is becoming more specific: not simply whether AI will transform the economy, but whether the profits generated by that transformation will be large enough to justify the capital being committed today.
That question will become increasingly important as depreciation rises, financing needs grow and technology companies move from building AI capacity toward proving its long-term commercial returns.
Follow the blog for more updates on AI, technology companies, global markets and the business trends shaping the next investment cycle.
This article is for informational and educational purposes only and should not be considered investment advice.