The AI Boom Rewards Applications, Not Valuations
The AI investment cycle concentrates capital in infrastructure, not in the applications that generate lasting value. OpenAI carries a $500 billion valuation against projected 2025 revenues of around $13 billion, a multiple that exceeds 38 times forward revenue. The pattern is not new, and the dot-com era's outcome offers a precise map of what tends to follow.
The Circular Investment Structure of the AI Ecosystem
A distinctive feature of the current cycle is the degree to which major players invest in each other. Bloomberg has documented how Microsoft, OpenAI, and Nvidia are effectively paying each other in circular deals: Microsoft holds a roughly 49% stake in OpenAI, OpenAI has signed a $300 billion cloud infrastructure agreement with Oracle, Oracle buys chips from Nvidia, and Nvidia has committed to invest up to $100 billion back into OpenAI.
OpenAI's revenues remain small relative to the scale of its commitments, with $13 billion projected for 2025, yet it has already signed a $300 billion cloud infrastructure agreement with Oracle and a $90 billion deal with AMD. AMD, in turn, offered OpenAI purchase options on up to 160 million of its shares.
At the heart of this structure is Nvidia, which recently reached a $5 trillion market capitalization and near-monopolistic pricing power in AI chips. Each participant reinforces the other's position, but the circularity also means that a confidence shock anywhere in the chain propagates everywhere.
¿En Qué se Parece Esto al Ciclo Punto-com?
💡 The right comparison: The dot-com bubble's most instructive lesson is not that technology was wrong, but that infrastructure investment ran ahead of real demand. The survivors were the companies that solved a specific user problem better than anyone else.
By 2001, the dot-com collapse had erased over $5 trillion in market value and sent numerous companies into bankruptcy. Yet Amazon, Google, and a handful of others not only survived but went on to define the next decade. The difference was not scale of investment: it was whether the company had built something that solved a concrete problem for real users.
The dot-com bubble inverted the natural order of innovation: instead of demand pulling infrastructure forward, infrastructure pushed ahead of demand. The crash destroyed valuations but left behind the physical networks that made Web 2.0 possible. Overbuilt assets sold at cents on the dollar to future winners like Google, Amazon Web Services, and Equinix.
The current AI cycle follows the same sequence. Base models and GPU capacity are receiving the majority of capital. Specific, industry-facing applications that could generate measurable returns today receive comparatively little.
Infrastructure vs. Applications: Where the Overinvestment Actually Sits
Guinness Global Investors notes that the downside of overinvestment in AI is industry-wide overcapacity, depressed returns on invested capital, and potential write-downs, losses that large technology companies can absorb, but that smaller infrastructure players cannot.
The pattern that repeats across technology cycles is that infrastructure is overvalued early and applications are undervalued. The companies building vertical solutions for specific industries, automating concrete processes, and generating measurable ROI from AI deployment are doing what Amazon and Google did in 1999-2002: solving a real problem rather than betting on a valuation.
Three Scenarios from Here
The source data supports three plausible outcomes, which are not mutually exclusive.
Gradual adjustment. Compute costs fall as competition and efficiency improvements (including model compression and smaller architectures) drive down the cost of inference. Access broadens. Practical applications proliferate because the barrier to deploying them drops.
Sharp correction. A confidence shock deflates valuations rapidly. Speculative infrastructure players fail. The narrative briefly becomes "AI was overhyped." But companies with real products and paying customers consolidate their positions during the correction, exactly as Amazon did between 2000 and 2003.
New dominant infrastructure layer. The massive compute buildout creates a permanent infrastructure layer analogous to what Amazon Web Services became after the dot-com bust, but more concentrated and more subject to state intervention given the geopolitical stakes around AI capability. A combination of this scenario and the sharp-correction scenario is the most plausible near-term path.
What the Cycle Means for Companies Adopting AI Now
The macro question of whether valuations correct is largely separate from the operational question of whether a specific AI deployment generates value. Even in a bubble-correction scenario, tech giants with diversified revenue streams are likely to absorb the write-downs, and the underlying models and infrastructure will remain available.
The companies that will look back on this period as an advantage are those that treated AI adoption as a process question rather than a market question: which specific workflows can be automated or improved, what does a measurable outcome look like, and how do you build internal knowledge as you go? That approach produces compounding returns regardless of where OpenAI's valuation sits in 2027.
Waiting for valuations to stabilize before beginning is the same bet that most dot-com-era companies made on infrastructure: they waited for the "right" platform and missed the window to build the user relationships that Amazon and Google were quietly accumulating.
