What If the AI Bubble Is Not a Bubble? Disentangling Speculation from Substance in the AI Revolution

Introduction: Beyond the Bubble Binary – Disentangling Speculation from Substance

The discourse surrounding Artificial Intelligence (AI) is sharply polarized. On one hand, it is hailed as a technological revolution on par with electricity or the internet, a foundational shift poised to reshape every facet of the global economy and human society. On the other, a growing chorus of financial heavyweights warns of a speculative bubble, characterized by irrational exuberance and unsustainable valuations reminiscent of past market manias. This report moves beyond this binary to ask a more nuanced question: Are we witnessing a speculative mania, a foundational paradigm shift, or, most likely, a complex interplay of both?

The analysis will demonstrate that the current AI landscape exhibits clear and undeniable characteristics of a financial bubble, including stretched valuations, extreme market concentration, and narrative-driven investment. However, unlike purely speculative manias of the past, this financial froth is layered upon a technological bedrock of unprecedented substance, general-purpose applicability, and long-term structural value. Therefore, a potential market correction would likely represent a painful but necessary recalibration of expectations, not an invalidation of the underlying technological shift.

Anatomy of the AI Bubble Narrative

The arguments supporting the AI bubble thesis are substantial, drawing on coordinated warnings from global financial institutions, historical parallels from past manias, and candid admissions from industry insiders. Together, they paint a comprehensive picture of the perceived risks and structural fragilities within the current AI investment boom.

The Chorus of Caution: Warnings from Global Financial Authorities

In recent months, the world’s most influential financial and regulatory bodies have begun to sound the alarm with increasing urgency. The Bank of England’s (BoE) Financial Policy Committee has issued what has been described as its “clearest warning yet,” stating that “The risk of a sharp market correction has increased”. The committee’s concern is rooted in data showing that “equity market valuations appear stretched, particularly for technology companies focused on artificial intelligence”. This is not an abstract concern; it is substantiated by the soaring valuations of private AI firms. For instance, OpenAI’s valuation skyrocketed to $500 billion from $157 billion in just one year, while Anthropic’s valuation nearly trebled to $170 billion in a matter of months. The BoE warns that this leaves equity markets “particularly exposed should expectations around the impact of AI become less optimistic”.

The International Monetary Fund (IMF) echoes these concerns in its Global Financial Stability Report, warning that AI-driven US stock markets are at risk of a “sudden, sharp correction”. While noting that overall market valuations have not yet reached the extremes of the dot-com bubble, the IMF highlights a “worryingly concentrated” market structure. A small cohort of mega-cap tech firms, dubbed the “Magnificent Seven,” now accounts for a historic high of 33% of the S&P 500 index’s total market capitalization. This concentration, according to the IMF, makes the broader market vulnerable to a downturn if these few firms fail to meet lofty expectations. IMF Managing Director Kristalina Georgieva has offered blunt advice to investors: “Buckle up,” stating that “uncertainty is the new normal”.

Leaders from major financial institutions have joined this chorus. JPMorgan CEO Jamie Dimon has warned of high asset prices and heightened uncertainty, while Goldman Sachs CEO David Solomon has pointed to historical cycles where markets inevitably “run ahead of the potential” of new technologies.

A warning from a single analyst can be dismissed as opinion. However, when the central banks of major economies, global financial stability monitors, and the CEOs of the world’s largest banks all issue coordinated warnings, the concern is elevated from market commentary to a formal assessment of systemic risk. This convergence indicates that the risk is not isolated to a few tech stocks but has the potential for “global spillovers” that could impact the broader economy. A sharp correction in AI-related equities could lead to a tightening of financial conditions, causing finance to dry up for households and businesses and potentially triggering a wider economic downturn. Furthermore, these high-profile warnings can themselves become a market-moving force, tempering investor enthusiasm and potentially precipitating the very correction they predict.

Echoes of the Past: The Dot-Com and Tulip Mania Analogies

To contextualize the current frenzy, experts and institutions are increasingly drawing parallels to historical speculative bubbles. The World Economic Forum and numerous analysts point to the dot-com bubble of the late 1990s as a powerful analogue. The similarities are striking: a compelling narrative of a world-changing technology, massive capital investment driven by a fear of missing out (FOMO), and company valuations that have become disconnected from current profitability or, in some cases, any revenue at all. The IMF explicitly states that the current investment boom “echoes the dot-com boom of the late 1990s”.

The more distant, yet potent, comparison to the 17th-century Dutch tulip mania has also resurfaced. While AI is not a flower bulb, the analogy serves as a historical touchstone for the psychological dynamics of a speculative bubble, where an asset’s price escalates far beyond its intrinsic value based on the belief that “buyers at high prices always seemed to be able to find someone willing to pay even more, until they suddenly couldn’t”.

However, a crucial lesson from these historical comparisons, particularly the dot-com era, is that a bubble bursting does not necessarily destroy the underlying technology. The dot-com crash wiped out countless companies like Pets.com, but the fiber-optic networks and internet protocols that were overbuilt during the boom laid the essential groundwork for the web we use today. This suggests that a financial bust in the AI sector may serve to rationalize the market without invalidating the technology’s long-term importance.

The Skeptics’ Dossier: Hype Cycles, Circular Financing, and the Profitability Gap

Beyond broad market warnings, specific critiques target the internal mechanics and financial health of the AI industry. Harvard-affiliated authors David Gray Widder and Mar Hicks argue that the generative AI boom reflects a classic tech hype cycle. In this pattern, a technology is aggressively marketed as “inevitable and necessary” to spur massive, premature infrastructure investment and create “path dependence,” effectively entrenching the technology before its significant flaws, such as hallucinations, biases, and high operational costs, become widely apparent and cause the hype bubble to deflate.

Yale School of Management scholars Jeffrey Sonnenfeld and Stephen Henriques highlight the danger of “tangled corporate deals” and systemic overinvestment. A primary concern is the emergence of what critics call “circular financing” or “vendor financing”. In this arrangement, a company invests in or lends to a customer, who then uses those funds to purchase the investor’s products, creating an illusion of organic demand and revenue growth. The $100 billion deal where Nvidia invests in OpenAI, which will then use the capital to purchase Nvidia’s chips, has been cited as a potential example of this dynamic.

This practice creates a fragile, self-reinforcing loop that inflates the perceived value of the entire ecosystem. A hardware giant like Nvidia invests billions in a foundation model company like OpenAI. OpenAI then places massive orders for Nvidia’s GPUs, which are booked as revenue, driving Nvidia’s stock price up and validating its investment strategy. This, in turn, helps justify OpenAI’s high valuation, which is predicated on its access to massive compute infrastructure. The entire system becomes a house of cards. If OpenAI’s business model fails to materialize and its demand for GPUs evaporates, Nvidia’s revenue projections would crater. This would cause its stock to collapse, wiping out the value of its investments in other AI startups and triggering a rapid, cascading failure across the ecosystem, a structural vulnerability that could lead to widespread contagion.

This structural risk is compounded by a stark profitability gap. A startling study from MIT found that 95% of organizations investing in generative AI are currently seeing zero return on their investment. Leading companies like OpenAI, despite a staggering $500 billion valuation, have never turned a profit and are projected to incur tens of billions in cumulative losses. This profound disconnect between narrative-driven valuation and financial fundamentals is a classic warning sign of a bubble.

Confessions from the Epicenter: Insider Admissions

Perhaps the most compelling evidence for the bubble narrative comes from the leaders at the very heart of the AI revolution. In a striking admission, OpenAI CEO Sam Altman has conceded that the sector may be in a bubble. “When bubbles happen, smart people get overexcited about a kernel of truth,” he stated, before answering with a direct “My opinion is yes” when asked if investors were overexcited about AI. He explicitly drew the parallel to the dot-com bubble, where the underlying technology, the internet, was real and important, but investors simply “got overexcited”.

Former Intel CEO Pat Gelsinger has been even more blunt. When asked in an interview if the industry was in an AI bubble, he responded, “Of course. Of course we are”. He described the market as “hyped” and noted that “enormous leverage” is being put into the system. However, Gelsinger qualifies this by predicting the bubble won’t burst for “several years,” arguing that businesses have yet to begin materially benefiting from their massive AI investments.

These insider admissions are not simple confessions; they are sophisticated strategic communications. First, they temper public and investor expectations, providing air cover if short-term results disappoint. Second, they function as a warning to weaker competitors and undisciplined investors, with Altman noting that “someone is going to lose a phenomenal amount of money”. Most importantly, these statements are declarations of confidence. By acknowledging the bubble, leaders like Altman are implicitly positioning their own companies as the future Amazons or Googles that will survive the inevitable crash and consolidate the market. Gelsinger’s prediction of a multi-year timeline serves to keep capital flowing in the short term while still acknowledging the long-term risk. This reveals the internal game theory of the AI race: its leaders are not just building technology; they are actively shaping the investment narrative to their advantage.

The Fad Fallacy: A Historical Perspective on Technological Skepticism

Before concluding that AI is merely a bubble destined to burst, it is critical to place current skepticism in historical context. The annals of technological innovation are replete with examples of transformative technologies that were initially dismissed as fads, niche toys, or impractical novelties by contemporary experts. This pattern, which can be termed the “fad fallacy,” serves as a powerful cautionary tale against underestimating the potential of exponential change.

“The Internet? Bah!”: The Ultimate Cautionary Tale

Perhaps the most famous and instructive example is the early skepticism surrounding the internet. In a now-infamous 1995 Newsweek article titled “The Internet? Bah!”, physicist and author Clifford Stoll declared the burgeoning network a passing fad. He confidently dismissed predictions of online shopping (“Uh, sure”) and e-commerce, famously asking, “how come my local mall does more business in an afternoon than the entire Internet handles in a month?”. Stoll characterized the web as a “wasteland of unfiltered data” and concluded that “no online database will replace your daily newspaper”. His critique, while not entirely without merit regarding the loss of human contact, stands as the quintessential example of an expert extrapolating from a technology’s nascent, clunky state and failing to envision its exponential evolution in usability, speed, and economic significance.

The Personal Computer: From Hobbyist Toy to Ubiquitous Tool

A similar pattern of skepticism surrounded the personal computer in the 1970s and 1980s. Early microcomputers were widely viewed as niche products for technicians, hobbyists, and “amateur electricians who enjoyed building their own computers with soldering irons and spare parts”. The primary barriers to mass adoption were high cost, operational complexity, and a perceived lack of clear utility for the average household. This sentiment was famously captured by Ken Olsen, the founder of the successful minicomputer company Digital Equipment Corporation (DEC), who stated in 1977, “There is no reason anyone would want a computer in their home”. This view failed to anticipate the powerful confluence of exponentially falling costs (Moore’s Law), dramatically improved user-friendliness (the graphical user interface), and the emergence of “killer applications” like word processing and spreadsheets that would ultimately make the PC an indispensable tool.

Niche Experiments to Mainstream Realities: EVs and Mobile Phones

The histories of electric vehicles and mobile phones offer further examples. Early EVs in the 20th century were dismissed as impractical due to their limited range, sluggish performance, and the lack of charging infrastructure, especially when compared to the convenience and affordability of gasoline-powered cars. They faded into obscurity for decades, only re-emerging as a central technology when enabling factors, battery performance, cost reductions, and climate imperatives, finally aligned.

Similarly, the first mobile phones were far from mass-market devices. The Motorola DynaTAC, launched in 1984, was a bulky device weighing over a kilogram, costing the equivalent of nearly $10,000 today, and was seen as a status symbol for “wealthy financiers and entrepreneurs”. Early network providers projected a tiny market limited to business elites, completely underestimating the explosive mass-market appeal that would be unlocked by subsequent decades of plummeting costs, miniaturization, and network expansion.

The core error made by early skeptics of these technologies was the linear extrapolation of an exponential phenomenon. They analyzed a technology in its primitive, expensive, and difficult-to-use form and, projecting that state forward in a straight line, concluded it would never be cheap, easy, or useful enough for the mass market. They failed to account for the exponential nature of progress in enabling areas: processing power, network bandwidth, battery density, and data storage. They also underestimated the power of ecosystem development, where new infrastructure and unforeseen “killer applications” emerge to create value that was previously unimaginable. The lesson for the current AI debate is profound: critics who focus on the present-day limitations of large language models, such as their high energy costs or propensity for hallucination, may be repeating this same historical error of judgment, underestimating the potential for future architectural breakthroughs and efficiency gains to solve these problems far faster than anticipated.

The Case for Durability: Why AI is a Foundational Shift

Despite the clear signs of financial speculation, a compelling case can be made that the underlying AI technology is not a transient fad but a durable, structural transformation. This argument rests on AI’s nature as a general-purpose technology, the powerful economic and infrastructural feedback loops it creates, and the fundamental change it represents in the paradigm of computation itself.

The Generality Engine: Deep Integration Across All Economic Sectors

Unlike many niche technologies that have fueled past bubbles, AI is a “general-purpose technology” with the potential for deep and transformative integration across virtually every industry. Its applicability is not confined to a single market or use case.

• In healthcare, AI is already revolutionizing diagnostics by analyzing medical images with superhuman accuracy, streamlining complex administrative tasks, and dramatically accelerating drug discovery. The prime example of this is DeepMind’s AlphaFold, which solved the 50-year-old grand challenge of protein folding. This breakthrough, which allows scientists to predict the 3D structure of a protein from its amino acid sequence, has monumental implications for understanding diseases and designing novel therapeutics.
• In logistics and manufacturing, global giants like Amazon, Walmart, and FedEx are deploying AI for sophisticated demand forecasting, real-time route optimization, and the automation of warehouse management. These applications have led to documented, large-scale efficiency gains, cost reductions, and decreased carbon emissions.
• In science and engineering, AI is being used for the generative design of new materials with desired properties, to improve the accuracy of complex climate models, and to accelerate research by parsing vast scientific datasets that would be impossible for humans to analyze. This demonstrates a fundamental value that extends far beyond purely commercial applications.

The Flywheel Effect: Data, Scale, and Infrastructure Entrenchment

The economics and infrastructure of modern AI create powerful self-reinforcing cycles that entrench the technology and its leading platforms.

• Data Feedback Loops: AI systems, particularly large language models (LLMs), operate on a principle of continuous improvement through use. Every user interaction provides a stream of data, both explicit feedback (like ratings) and implicit feedback (like user follow-up questions), that is used to refine and retrain the models. This creates a powerful feedback loop: the more a model is used, the better it gets, which in turn attracts more users and generates more data, spinning the flywheel faster.
• Network Effects and “Stickiness”: As leading AI models become de facto platforms, an entire ecosystem of downstream applications is built upon their APIs. This creates strong network effects and high switching costs for businesses that have integrated these platforms into their core workflows. This “stickiness” makes the leading platforms difficult to displace, creating deep economic moats. A fad is typically a single product with a finite lifecycle, like the Pet Rock. Even many companies in the dot-com bubble were essentially single “products,” like Pets.com. Foundational AI models, however, are not products; they are platforms, akin to an operating system or a utility grid, upon which an entire ecosystem of other products and services is built. This structural difference suggests a much higher degree of durability and resilience compared to past bubbles built on less foundational technologies.
• Infrastructure Investment: The massive, multi-trillion-dollar global investment in the physical infrastructure of AI, data centers, specialized GPUs, and power generation facilities, creates a tangible, hard-to-reverse momentum. This physical buildout, representing a concrete commitment to a new technological paradigm, is a key differentiator from more ephemeral, software-only bubbles.

A New Computational Paradigm

Crucially, AI is not merely an incremental improvement on existing software. It represents a fundamental shift in the nature of computation itself. For decades, software has been built on a paradigm of deterministic, rule-based logic, where a human programmer must explicitly code every step and contingency. AI, and specifically machine learning, introduces an era of probabilistic, learning-based inference. In this new paradigm, a system learns patterns, relationships, and capabilities directly from data, without being explicitly programmed for each task. This allows us to solve problems, such as image recognition, natural language understanding, and complex pattern detection, that were previously intractable under the old, deterministic paradigm.

This Time is Different? Quantitative and Qualitative Divergences

Several pieces of evidence suggest that the current AI boom has structural differences from past tech bubbles. A 2025 preprint on arXiv that examined scientific citation networks found that the patterns of scientific publishing in the AI era do not cleanly mirror those that preceded the dot-com bubble. This divergence led the authors to suggest that “this is not the same kind of bubble” and that the underlying scientific foundations may be more robust.

Furthermore, the financial underpinnings are different. Unlike the many dot-com startups that were burning through venture capital with no revenue, the current AI boom is being heavily financed by some of the most profitable and cash-rich companies in history (the “Magnificent Seven”). These giants are funding a significant portion of the AI buildout from their own massive internal cash flows, providing a more stable financial foundation than the purely speculative external capital that fueled the dot-com era.

The Human Imperative: Why We Need Artificial Intelligence

Beyond the commercial opportunities and financial speculation, a powerful argument for the durability of AI is that it is becoming a necessary tool for addressing humanity’s most pressing and complex challenges. AI is not just a luxury; it is an imperative for progress and, in some cases, survival.

A Cognitive Prosthesis for Grand Challenges

Humanity faces a class of existential-scale problems characterized by immense complexity, non-linear dynamics, and vast datasets that overwhelm human cognitive capacity. AI offers a form of “cognitive prosthesis” that is indispensable for tackling these challenges.

• Climate Change: AI is essential for improving the accuracy of climate models, which are critical for policy-making. It is also being deployed to optimize the performance of renewable energy grids, manage energy storage systems, and monitor deforestation and polar ice melt with high precision using satellite imagery analysis.
• Epidemiology and Public Health: In a world facing the threat of future pandemics, AI can help forecast disease spread, accelerate the discovery of new drugs and vaccines, and enable the development of personalized medicine tailored to an individual’s genetic makeup.
• Resource Optimization: AI is critical for managing fragile global supply chains, optimizing agricultural yields to feed a growing population, and ensuring the efficient use of vital resources like fresh water and energy.

In these domains, human-only methods are insufficient to process the complexity and scale of the data. AI provides an indispensable toolkit for modeling, prediction, and decision support.

The Co-Pilot for Creativity and Discovery

Contrary to the dystopian narrative of machines replacing human creativity, AI is emerging as a powerful collaborative partner that augments and expands human ingenuity. AI tools can enhance the creative process by expanding the “search space” of possibilities, proposing novel patterns, combinations, and designs in fields like art, music, engineering, and scientific research. By automating tedious and repetitive tasks, AI frees up human creators to focus on higher-level conceptual work and strategic thinking. This is most powerfully demonstrated in science, where tools like AlphaFold provide new lenses for inquiry, helping scientists achieve breakthroughs that were previously out of reach.

Democratizing Expertise and Opportunity

One of AI’s most profound potential impacts is its ability to democratize access to expert-level capabilities, making specialized knowledge and services more affordable and accessible to a broader population, thereby helping to reduce inequality.

• Personalized Education: AI-powered tutors can provide individualized instruction tailored to each student’s unique pace and learning style. This can help bridge educational gaps caused by geography, socioeconomic status, or limited resources, breaking down traditional barriers to high-quality learning.
• Accessible Healthcare: AI diagnostic tools can serve as a “co-pilot” for clinicians in underserved or remote regions, enabling earlier and more accurate detection of diseases like diabetic retinopathy, tuberculosis, or certain cancers. This democratizes access to specialized medical knowledge that is currently concentrated in urban centers.

This potential for democratization offers a solution to a persistent macroeconomic challenge. The economist William Baumol identified a “cost disease” where costs in labor-intensive sectors with low productivity growth, such as education and healthcare, perpetually rise relative to goods-producing sectors. This makes high-quality education and healthcare increasingly unaffordable and inequitable. AI introduces a non-human factor that can dramatically scale the delivery of expertise. An AI tutor or diagnostic tool can serve millions of users simultaneously with minimal marginal cost. Therefore, AI is not just a technological enhancement; it represents a potential macroeconomic solution to the unsustainable cost growth of essential human services, framing the need for its development as a fundamental economic imperative.

A Clear-Eyed View: Navigating the Risks and Realities

While the case for AI’s long-term value is strong, a balanced perspective requires acknowledging the significant risks, limitations, and costs associated with the current AI revolution. Complacent optimism is unwarranted; the path forward is fraught with technical, financial, and societal challenges.

The Perils of Hype and Misallocated Capital

Even if the underlying technology is transformative, the risk of a severe financial correction remains high. The extreme overvaluation in parts of the market means that, as Sam Altman noted, “someone is going to lose a phenomenal amount of money”. A major danger of any bubble is the misallocation of capital, where vast sums flow into weak ideas, fraudulent startups engaged in “AI-washing,” and projects with no viable business model, at the expense of genuinely promising but less-hyped research and development.

The Technical Frontier: Current Limitations

Today’s AI models, particularly LLMs, have significant and well-documented technical limitations that undermine their reliability and trustworthiness in high-stakes applications.

• Hallucination: LLMs are notoriously prone to generating “plausible yet nonfactual content”. This tendency to “hallucinate” is arguably the “biggest hindrance to safely deploying” these models in critical applications like medicine or finance. Some computer science research even posits that, given the probabilistic nature of current architectures, hallucination may be an “inevitable” limitation that cannot be fully eliminated.
• Bias: Trained on vast swathes of internet data, AI models inevitably learn and can amplify the societal biases related to gender, race, and culture embedded in that data. This can lead to unfair, discriminatory, or stereotypical outputs, posing a significant ethical risk.
• Lack of Robust Reasoning and Explainability: LLMs often struggle with logical consistency, causal reasoning, and common-sense understanding. Their “black box” nature makes it difficult, if not impossible, to fully understand why they produce a certain output. This lack of explainability is a critical failure for high-stakes domains where accountability and transparency are paramount.

The Societal Ledger: Ethical and Governance Risks

The deployment of AI at scale introduces profound ethical and societal risks that require careful governance.

• Job Displacement: While AI will undoubtedly create new jobs, it is also expected to automate a significant portion of existing tasks. A Goldman Sachs report estimates that up to 300 million full-time jobs could be impacted globally, threatening widespread economic dislocation and exacerbating inequality if not managed through proactive policy, reskilling efforts, and social safety nets.
• Disinformation and Manipulation: The ability of generative AI to create highly realistic text, images, and video at scale poses a severe threat to the global information ecosystem. This technology can be weaponized for sophisticated propaganda, personalized influence campaigns, and the mass production of “deepfakes,” potentially destabilizing democratic processes and eroding social trust.
• Concentration of Power: The immense cost of data and computation required to train frontier AI models risks concentrating unprecedented power in the hands of a few large corporations and governments. This power could be used for mass surveillance, censorship, and social control, posing a direct threat to open societies and individual liberties.

The Physical Footprint: Environmental Costs

The AI boom has a substantial and rapidly growing environmental footprint. Training and running large models consumes vast amounts of electricity and water, creating a central paradox for a technology often touted as a solution to climate change.

• Energy Consumption: The energy required for AI is staggering. The training of GPT-3 is estimated to have emitted 552 metric tons of CO2, equivalent to 123 gasoline-powered cars driven for a year. The computational power needed for top models has been doubling every few months, leading to an exponential growth in energy demand that has already caused some utility providers to delay the retirement of coal-fired power plants to meet the needs of new data centers.
• Water Usage: Cooling the massive data centers that power AI requires millions of liters of fresh water, placing a significant strain on local water resources, particularly in drought-prone regions.

This creates a fundamental tension: the very tool being developed to help solve our greatest environmental challenge is currently a significant and rapidly growing contributor to that same problem. This reframes the core challenge of AI development. The goal cannot simply be to create more powerful models; it must be to create more efficient models. The future viability and ethical justification of AI may depend as much on breakthroughs in algorithmic efficiency and energy-saving hardware as it does on raw capability.

Conclusion: A Bubble of Speculation on a Bedrock of Substance

The evidence presented in this report leads to a clear, albeit complex, conclusion: the AI market of 2025 displays all the classic hallmarks of a speculative financial bubble. Valuations are stretched to historic levels, market concentration is extreme, and a powerful, seductive narrative is driving investment far ahead of proven profitability. The coordinated warnings from the world’s most senior financial guardians should not be ignored.

However, to dismiss AI as just a bubble is to repeat the historical error of the fad fallacy. The speculation is occurring on top of a genuine and durable technological revolution. AI’s nature as a general-purpose, platform-level technology, its potential for deep integration across all economic and scientific sectors, and its demonstrated ability to deliver tangible breakthroughs like AlphaFold provide a bedrock of substance that many past bubbles lacked.

A market correction, perhaps a severe one, is not only possible but may be a healthy and necessary event. Such a reckoning would serve to wash out the speculative excess, discipline capital allocation, and separate the viable long-term players from the short-term opportunists. Such a correction would be a financial event, not a technological one. It would burst the bubble of inflated expectations and unrealistic timelines, but it would not erase the underlying technology. The core architecture of intelligent systems, models that learn, adapt, and augment human capability, is here to stay. It will remain integral to our economic and societal future long after the current froth has settled.

Next time someone dismisses AI as “just a bubble,” flip the question: What should we be doing to make sure it isn’t? Because the truth is, AI is already too convenient, and too embedded, to ignore.