4. Technological and efficiency tailwinds
Finally, the bulls argue that concerns over power shortages and hardware obsolescence will be mitigated by efficiency gains. NVIDIA reports a 40,000% improvement in their chips’ energy efficiency over time, suggesting that compute density and performance per watt will continue to outpace demand growth. Meanwhile, rapid improvements in model performance may create deflationary effects, allowing AI to boost productivity across the global economy faster than it consumes capital or energy. In this view, AI spending becomes self-reinforcing: productivity gains fund further investment, drawing more capital into the ecosystem—a classic virtuous cycle.
The Bear Case: Unsustainable Leverage, Thin Revenues and Physical Limits
The bear case challenges the notion that AI’s financing machine can run indefinitely without seeing sufficient revenues to justify the investment. This view suggests that beneath the headline numbers, the business models supporting the sector remain fragile, while physical and regulatory bottlenecks loom large.
1. Fragile revenue foundations
OpenAI’s monetization dilemma is a central concern. Despite its vast user base, only a small fraction of users pay directly, leaving revenue concentrated in subscriptions and application programming interface (API) fees. While some may assume adoption by businesses will close this gap, the path to sustained free cash flow is quite uncertain. With projections suggesting OpenAI may seek $1.6 trillion in new capital, skeptics question whether such ambitions can coexist with a largely free consumer model resulting in revenues of just $13 billion per year. The risk is that a prolonged mismatch between spending and revenue could erode balance sheets, triggering refinancing stress and equity underperformance.
2. Leverage masks fragility
Heavy reliance on debt markets—especially structured finance—could obscure true risk concentration. The Meta financing deal, cited by bulls as a model of efficiency, could come to resemble the over-leveraged telecom infrastructure of the early 2000s. If rates rise or utilization lags, the yield advantage could quickly reverse, pressuring balance sheets across the ecosystem.
3. Power as the hard constraint
Perhaps the most tangible risk lies in power infrastructure. Between 2028 and 2035, data centers could add a projected 15% to 20% strain on global grids. Even with efficiency gains, power transmission, transformer production and permitting timelines create bottlenecks that financial engineering cannot solve. The result could be stranded capital: data centers built faster than utilities can deliver power.
4. Governance and concentration risks
Finally, investors need to consider governance. The market’s overreliance on visionary figures—OpenAI CEO Sam Altman foremost among them—creates fragility in leadership and strategy. Meanwhile, the distinction between “defensive” spending to protect existing platforms—such as Google adding Gemini to protect its search business—and “offensive” expansion, like Microsoft introducing and charging more for Co-Pilot, remains blurred. This raises concerns that much of today’s capital expenditures serve short-term competitive positioning rather than long-term profitability.
Implications for Investors
Much of the U.S. equity market’s gains since ChatGPT’s launch in late 2022 have been driven by companies at the center of the AI ecosystem. As investors look to the future, a key question is whether this powerful trend can continue or is vulnerable to a sharp pullback. One of the key challenges is in identifying which companies can convert AI infrastructure into recurring, high-margin revenue streams, and which are merely relying on increasingly risky financing.
AI may indeed catalyze the next productivity supercycle. But as this debate made clear, the path from vision to value will depend less on the amount of capital that is raised—and more on how productively that capital is deployed.
