Based on a new report from S&P Global, a leading provider of financial market intelligence, the rapid expansion of AI infrastructure is beginning to resemble a classic late-cycle investment surge. The report’s findings echo what many AI sector experts are seeing: investment is pouring in at unprecedented scale without a clear path to profit, boosting the risk that today’s buildout may far overshoot economic reality.
It’s now a pressing question: is the pace of the current infrastructure buildout setting the stage for a major correction? Is this all one huge bubble?
“While aggressive spending without clear signs of reasonable returns could potentially lead to a bubble, we do not observe this phenomenon in the current environment,” said David Tsui, managing director, S&P Global Corporate Ratings.
The truth about AI infrastructure growth, it seems, is a bit more complicated than boom or bust.
Partners Become Rivals
The AI boom has entered a complicated phase. What began as an urgent scramble for computing power is turning into a cash-intensive competition that is straining balance sheets, not to mention local power grids across the country. According to S&P Global’s analysis, the industry is shifting from an early, cooperative expansion toward an adversarial stage, where execution and timing matter as much as tech leadership.
Over the past two years, jaw-dropping spending by hyperscalers and frontier AI labs such as OpenAI and Anthropic has created a self-reinforcing cycle. AI developers needed vast amounts of compute to train frontier models. The hyperscalers had the data centers and capital to support that demand (and even drive it), often through questionable spending within their own ecosystems—the infamous “circular spending” we see. Chipmakers and infrastructure vendors have enjoyed the sweet spot, benefiting from supply shortages that kept their stock prices galloping upward.
That structure is now changing. Competition is intensifying across every layer of the AI stack, from chips and data centers to models and applications. AI labs are seeking greater independence by exploring their own infrastructure and silicon strategies. Hyperscalers, in turn, are working to develop in-house chips and secure long-term power and data center capacity to avoid over reliance on external suppliers. A big example of this is Google’s in-house TPU chip, which gained major credence toward the end of 2025.
These shifts may strengthen individual players over time, but they also turn former partners into rivals as each dips into the others’ sector, compressing margins and raising the stakes.
A Troubling Gap
S&P Global’s report spotlights a central pain point: the now worrisome gap between investment and return. The AI ecosystem increasingly relies on heavy debt to support infrastructure spending that is arriving well before broad-based adoption, even as the enterprise market is aggressively experimenting with AI.
And while enterprise AI spend is rising, it still represents a small fraction of total enterprise software budgets. Productivity gains remain concentrated among early adopters, and the most lucrative use cases haven’t yet scaled across the wider consumer economy.
This mismatch creates obvious bubble dynamics. Investment is committed today on the belief that future demand will show up at sufficient scale and speed. If that demand arrives late or unevenly, which seems likely, assets designed for intensive AI training could become a large hole in corporate balance sheets. Data centers built for specialized workloads, such as the neo-clouds like CoreWeave, are not always easily repurposed.
Which brings us to the reality of physical limits. Power availability and grid interconnections are emerging as bottlenecks. Sometimes companies secure chips they can’t immediately deploy, which effectively strands capital.
Plus, and this is a big one, the lesson of DeepSeek in 2025 was that efficiency improvements in models or hardware could reduce compute requirements, undercutting the assumptions behind today’s largest infrastructure projects.
Possible Winners
Not all the players in this unfolding drama face the same level of risks. S&P Global’s credit analysis suggests hyperscalers remain best positioned due to their gargantuan cash flows, even as aggressive spending erodes free cash flow in the near term.
Frontier AI labs appear far more speculative, given their voracious funding needs and uncertain path to real profitability. Infrastructure providers that rely on a narrow customer base or heavy leverage, the CoreWeaves of the world, are highly vulnerable if spending slows.
To be sure, AI itself will continue to devour everything in its sight. AI is set to support a generational change in competitiveness for companies that can integrate it into core operations. But the S&P research reveals a narrowing margin for error. Both investors and operators are facing a hard truth: when hopes and dreams outpace actual dollars flowing in the door, painful corrections are a real possibility.
