Brad DeLong cuts through the fever dream of artificial intelligence to ask a question few are willing to pose: who actually pays for the massive data centers being built today, and will the productivity gains justify the cost? While the market chases the promise of a "digital god," DeLong argues we are witnessing a "Rashomon" moment where the reality of narrow, incremental tools clashes with the hype of a general-purpose revolution. This piece is essential because it refuses to choose between utopian optimism and doomsday pessimism, instead offering a grounded economic analysis of why the current boom might be driven more by platform monopolies fearing obsolescence than by genuine consumer demand.
The Gap Between Hype and Utility
DeLong begins by dismantling the idea that modern advanced machine-learning models are about to upend daily life. He leans on the experience of physicist Chad Orzel to illustrate that for most professionals, the technology remains a "squishy lukewarm" addition rather than a transformative force. Orzel's experiments with summarizing data and extracting numbers from spreadsheets resulted in "a complete hallucination" or required so much verification that the time saved was negligible. DeLong uses this to highlight a critical friction point: when tasks are high-stakes and infrequent, the cost of checking an AI's work often exceeds the cost of doing it yourself.
The author suggests that the current utility is "narrow and incremental," serving mostly as a natural-language interface for structured data or a syntax checker for programmers. This is a sobering correction to the narrative that we are on the brink of a singularity. As DeLong puts it, "MAMLMs may be of great use, but they will not upend my workflow and daily experience, let alone that of people who are not part of the tech-clerisy." This framing is effective because it grounds the discussion in the mundane reality of administrative work, where the promise of automation often crashes against the need for accuracy.
Critics might argue that early adoption curves for technologies like the internet or electricity were similarly unimpressive before the "killer apps" emerged, suggesting DeLong is underestimating the speed of adaptation. However, the distinction here is the sheer scale of capital currently being deployed before the value proposition is proven.
"The theme of the 2025-26 academic year is clearly 'Fretting About the Bag of Words'... A complete hallucination. Wrong names, wrong number of columns, made-up comments."
The Economics of Fear and Platform Power
The commentary shifts from the limitations of the technology to the motivations of the builders. DeLong identifies a powerful dynamic: the "cloud oligopolists" are not building these data centers because they are certain of the returns, but because they are terrified of being left out. He notes that "everyone with a platform monopoly (except Apple) is working diligently and spending whatever is needed to eliminate OpenAI's ability to exist anywhere near its consumer space." This is not a market driven by organic demand, but by a defensive war to capture the "application-layer rents."
The author draws a sharp parallel to the historical "Netscape-meets-Microsoft" era, but with "unbelievable scale datacenter investments added on." The result is a scenario where Microsoft, Google, Amazon, and Meta are pouring hundreds of billions into compute capacity, effectively turning regions like Northern Virginia into "GPU-powered company towns." DeLong argues that this investment wave is "bootstrapped into existence by a concentrated investment wave driven by fear of being the one big player left without a chair when the music stops."
This analysis is crucial for understanding the current economic distortion. The "goldfield is fenced," meaning the upside is real, but the "digital shovels and picks" are controlled by the same handful of firms funding the boom. As DeLong writes, "It is not rational for all to flood into AI startups... given that no more than a small fraction of them will ever earn back the opportunity costs of their time and capital once the dust settles."
"The history of Netscape-meets-Microsoft, rhyming, but this time with unbelievable scale datacenter investments added on."
The Productivity Paradox and the "Digital God"
The final section of the argument tackles the macroeconomic expectations. DeLong contrasts the optimistic "consultant deck" narrative—which promises trillions in GDP growth—with the cautious reality that we are currently in a "something will turn up" mode. He references the concept of a General-Purpose Technology (GPT), noting that while AI has the potential to be one, the current build-out is fueled by "narrative-driven exuberance."
The author warns that we risk repeating the "curse of dimensionality" where the complexity of the model outstrips the clarity of its application. He points out that the "weaknesses of LLMs are being addressed by exponential increases in compute," but questions whether this solves the fundamental issue of value creation. The risk is that the industry is "leveraged into an investment wave far ahead of demonstrated cash flows," creating a scenario where the sunk costs of data centers may never be recovered if the productivity gains remain "highly uncertain."
DeLong's personal reflection on his past underestimation of Uber serves as a humble reminder that "technological change plus very patient capital can sometimes hold together arrangements that look... unsustainable." Yet, he cautions against universalizing this lesson, noting that "anyone offering you guarantees is selling snake oil." The core tension remains: is this a genuine productivity revolution or a massive bubble propped up by the fear of missing out?
"Hence right now we are still in 'something will turn up' mode; hence right now 'WE ARE BUILDING DIGITAL GOD!!!!' is still playing an enormous role here as an energizer, for hard numbers do no."
Bottom Line
DeLong's strongest contribution is his refusal to let the "digital god" narrative obscure the economic reality that platform monopolies are driving a defensive, capital-intensive arms race rather than a consumer-led revolution. The argument's greatest vulnerability lies in its reliance on current utility metrics, which may fail to capture the non-linear breakthroughs that often define true General-Purpose Technologies. Readers should watch for whether the massive datacenter build-out can actually generate the productivity surplus required to pay for itself, or if we are simply witnessing a costly correction in the making.