The Reality Check: BIS Warns of Unsustainable AI Costs
The Bank for International Settlements has issued a stark warning regarding the massive spending on artificial intelligence infrastructure. Economists at the organization state that annual revenue of approximately $3.7 trillion will be needed by 2032 to prevent these investments from ending in losses.
This required revenue amount is equivalent to about 85% of Japan’s annual GDP, a figure that highlights the sheer scale of the financial gap facing the industry.
The Revenue Gap and Off-Balance-Sheet Burdens
Despite the ambitious projections, current market reality suggests a significant disconnect between spending and income. Combined revenue for OpenAI and Anthropic in 2026 is estimated at only around $0.1 trillion.
Hyperscalers are currently paying cloud usage amounts that far exceed their current revenue, with funding provided by the investors themselves. Money is simply circulating within the system, and there is still little income coming in from the outside world.
The financial pressure is compounded by massive off-balance-sheet obligations. The future leases and purchase commitments for semiconductors and other items for four major companies already total $2.4 trillion. If AI usage revenue does not quickly approach the necessary levels, this burden will become unsustainable for the industry players.
Market Corrections and Historical Parallels
The financial markets are reacting to these concerns with tangible losses. The Philadelphia semiconductor index posted its eighth straight loss, marking its longest run since 2008.
NVIDIA stock slid 1.43% and Microsoft fell 1.93% following Bank of America’s warning note on October 16, 2026. Alphabet fell 2% and Meta dropped nearly 4% on Raedler’s warning as well.
These declines mirror monetization failures in past booms like railroads and fiber networks. Token prices fell roughly 50% since summer, reflecting the same pattern where underlying technology succeeded but investor returns suffered.
Bank of America investment strategist Sebastian Raedler argued that AI is tracking the same path as previous capital cycles. He noted that most major capital cycles of the past have ended in disappointment, even though the underlying technology often lived up to expectations.
Investing for the End Users
Columbia University estimates that cumulative AI infrastructure investment from 2025 to 2032 is around $9 trillion. This spending must generate more than $3.5 trillion in annual revenue by 2032 just to break even.
Historical examples show telecom companies built too much fiber optic infrastructure while the businesses using the finished product became the ultimate winners. The quality of the equipment and infrastructure improved dramatically, helping internet-based businesses become financially feasible much more quickly.
The 19th-century railroad expansion resulted in over-construction that hurt ROI for builders yet the broader economy benefited enormously from cheaper shipping rates. Again, it was the businesses that ended up heavily using the new infrastructure that benefited the most, not the ones that were deeply involved and invested in its development.
BIS economists argue that as AI automation progresses, investment in machines will swell while the proportion of the economy going toward consumption decreases. Since profits from new AI are generated from consumer-facing businesses, the source of funding for these massive infrastructure projects will eventually dwindle.
The warning has shifted from a historical comparison to the real-world problem of whether the investment can actually be recovered. As is usually the case in such booms, the winners may not be the ones spending all the money but rather those who implement the technology into their businesses.
Sources
- The Warning on AI Investment Has Shifted from 'Historical Parallels' to the 'Reality of Return on Investment'｜かんたん
- BofA Warns AI Investors Could Lose Even If the Technology Works
- Wall Street's AI Boom Just Reached a Dangerous Milestone. History Says the Winners Could Change. | The Motley Fool
Photo: Winston Chen / Unsplash (Unsplash License)
