The AI Bubble Debate: The Arguments on Both Sides

The AI bubble is one of the most-discussed economic-technological debates of recent years: the question of whether the investment, market valuations, and social expectations flowing into artificial intelligence exceed the technology's real productivity contribution today. On one side are those who say "this is an AI bubble and it will burst sooner or later," and on the other those who say "this time it is a genuine transformation, not a bubble." In this guide we examine both sides' arguments, their supporting grounds, and their counterarguments with a measured, balanced framework.

Our aim is not to take a side but to build a way of thinking that lets you reach your own decision. Because a one-word answer to "bubble or not" is misleading; the right question is more nuanced: in which layer is there a genuine overvaluation, and in which layer is lasting value accumulating? In this piece we examine, with a consultant's rigor, the source of the debate, the arguments of those who say bubble, the counterarguments, the expectations-curve perspective, the lessons from past technology bubbles, the mechanics of the investment cycle, the practical takeaway for organizations, and finally a balanced evaluation framework you can use to read your own context.

Definition

The AI Bubble (Debate)

An economic-technological debate describing the question of whether the investment, market valuations, and social expectations flowing into artificial intelligence exceed the technology's real and measurable productivity contribution today. Those who say bubble point to overvaluation, high cost, unproven return, and narrative inflation; the other side points to a genuine leap in capability, fast adoption, and lasting infrastructure value. The right reading is usually in between: inflated expectation and genuine transformation can coexist.

Also known as: AI bubble, AI overvaluation debate, AI hype, AI investment cycle

A "bubble," in the language of finance, is when an asset's price departs markedly from its fundamental (real) value and that departure rests largely on expectation, enthusiasm, and fear of missing out. The AI bubble debate asks exactly this: is the value ascribed to artificial intelligence supported by the concrete benefit it produces today, or by enthusiastic expectation of what it will produce in the future? The larger the gap between the two, the more solid the ground the bubble debate stands on.

There is a critical distinction here: a technology being real and transformative and a bubble forming around it do not exclude each other. The internet was real; there was still a dot-com bubble. Electricity was real; most early electricity companies still went bankrupt. So the question "is AI a genuine transformation" and the question "is there an AI bubble" are separate questions, and answering "yes" to both is not inconsistent. Every debate that misses this distinction gets stuck in a sterile "real/fake" dilemma.

To frame the AI bubble debate correctly, then, three separate layers must be distinguished. First, the technology layer: has the capability of the models really increased? Second, the adoption layer: is this capability really being used by real users and organizations? Third, the finance layer: are market valuations and the investment cycle supported by the real revenue this capability and adoption produce? Most of the bubble debate is actually about this third layer; the two sides usually disagree less than it appears about the first two.

This frame is the backbone of the rest of the piece. While listening to the arguments of those who say bubble and while evaluating the counterarguments, the question we keep asking will be: which layer is this argument about? Because an argument can be strong on the finance layer while weak on the technology layer, or vice versa. A realistic view requires reading these layers without mixing them.

One point must be clarified from the start: this piece is not investment advice and contains no buy-or-sell guidance about any asset, share, or venture. The aim is to clarify the AI bubble debate conceptually and to help enterprise decision-makers in particular build a sound thinking framework in their own context. Making precise predictions about market movements is also not this piece's job; because by the nature of the bubble debate, no one can know the exact timing of the peak or the trough. Our focus is not timing prediction but a decision discipline that needs no prediction.

There are several concrete sources for why the AI bubble narrative is spoken so loudly, and understanding them explains why the debate is peaking now. The first source is the extraordinary burst of attention in a short time: generative AI tools reached a broad user base very quickly, and almost every sector fell into a rush of "we too must do AI." This fast and widespread enthusiasm is, historically, the classic precursor of bubble debates.

The second source is the scale of investment. The size of the capital flowing into compute, data centers, chip production, and model development raised the question, for many observers, "will this money really come back?" In this early phase of the investment cycle, money can flow indiscriminately even into unproven business models; this feeds the overvaluation debate. The bulk and rapid flow of capital creates an uncertainty in which winners and losers have not yet separated.

The third source is narrative inflation. Sweeping, exaggerated statements like "AI will change everything," "every job will disappear," "general AI is a matter of time" widen the gap between real capabilities and expectations. When expectation runs far ahead of reality, an inevitable disappointment follows, and with it the "so it was a bubble" reaction. This narrative inflation is a phenomenon produced by both defenders and critics of the technology; both sides tend to exaggerate.

The fourth source is the gulf in enterprise experience. Many organizations started AI pilots; some produced concrete value, but a significant portion got stuck at the "impressive demo, uncertain business result" stage. Pilots failing to reach production, adoption being slower than expected, and returns not being clearly measured created a healthy skepticism among people in the field. We cover in detail why user adoption is difficult in the factors that determine user adoption; this experience is the most concrete field-level feeder of the bubble debate.

Taking seriously the arguments of those who argue there is an AI bubble is the first condition of a balanced evaluation. The strongest of these arguments are structural, not emotional; let us take each in its strongest possible form.

The first argument is overvaluation. Those who say bubble argue that most market valuations are supported not by today's revenue but by expectation of enormous future revenue. If a company's value is priced on assumed but not-yet-produced profit, the correction is sharp when that expectation is not met. The strength of this argument is that it fits a historical pattern: in every major technology wave, valuations first ran ahead of reality, then aligned.

The second argument is cost-revenue imbalance. AI infrastructure — compute, energy, data centers, expert people — is very expensive. Those who say bubble argue that these costs far exceed the revenue produced and that this imbalance is unsustainable. Especially the scale of energy and hardware cost sharpens the question "with what revenue model will this investment come back?" We cover where cost arises and how it is managed in LLM cost optimization; this cost reality is the most concrete leg of the bubble argument.

The third argument is unproven large-scale return. Those who say bubble argue that, despite impressive demos, a measurable and repeatable return on investment at the enterprise level has not yet been broadly proven. Pilots failing to reach production, productivity gains not being clearly measured, and the "everyone is doing it but no one can show net profit" situation feed this argument. Indeed, a technology working and it producing net financial value for an organization are separate things.

The fourth argument is narrative inflation and expectation swelling. Those who say bubble argue that the "AI" label has turned into a marketing magic word, that even products with no real innovation are priced with this label, and that this narrative inflation calls for an inevitable correction. The more expectation swells, the sharper the disappointment.

The four main arguments of those who say bubble and which layer they belong to

Argument	Essence	Relevant layer
Overvaluation	Price rests on future expectation more than today's revenue	Finance
Cost-revenue imbalance	Infrastructure and energy cost far above revenue	Finance / operations
Unproven return	Net enterprise-scale ROI not yet common	Adoption
Narrative inflation	Expectation ran ahead of reality, correction inevitable	Perception / narrative

The common strength of these arguments is that none of them says "the technology does not work." The soundest bubble arguments target not the technology but the price, the timing, and the expectation. So taking them seriously does not mean belittling AI; on the contrary, it is a must for a realistic view.

Now let us turn to the other side of the coin. Those who oppose the AI bubble narrative also have strong arguments, and these too must be taken in their strongest possible form. "This time is different" is one of the most dangerous sentences in the history of finance; but in some cases it really is different, and ignoring the difference is also a mistake.

The first counterargument is a real and measurable leap in capability. What distinguishes this technology from earlier ones is that it produces not a lab promise but a concrete capability that millions of people already use daily. In tasks like text generation, summarization, writing code, translation, and data extraction, there is a measurable productivity contribution. If a technology is already used broadly and voluntarily, calling it "just narrative" becomes hard. This is a fundamental difference from many earlier bubble candidates.

The second counterargument is the speed of adoption. Many transformative technologies in history took long years to reach mass adoption; generative AI tools reached a very broad user base in a very short time. This speed is a sign of real demand. For Türkiye in particular this adoption is especially high, and this is concrete evidence that there is real usage in the field.

The third counterargument is the durability of infrastructure. Even if the bubble bursts, what remains behind? The other side argues that the investment in infrastructure — compute capacity, data centers, energy infrastructure, accumulated expertise — leaves lasting value however the bubble debate resolves. This fits the historical pattern: the railway bubble burst but the tracks remained and carried the economy for decades. Infrastructure is the core that survives after the froth of the bubble subsides.

The fourth counterargument is the technology's general-purpose nature. AI is a horizontal technology touching almost every knowledge task rather than a single sector, just like electricity or the internet. The value of general-purpose technologies emerges not in a single application but in the sum of countless applications, and the maturing of this value takes time. This argument answers the "we do not see net ROI today" criticism with "because the return of general-purpose technologies comes with a lag."

The clearest way to see the debate is to place each argument alongside its grounds and its counterargument. The table below summarizes the main axes of the AI bubble debate in this three-part structure; it is built not to pick a side but to let you weigh each claim together with its opposite.

The AI bubble debate: argument × grounds × counterargument

Bubble argument	Grounds	Counterargument
Valuations are excessive	Price rests on future expectation more than today's revenue	In general-purpose tech return comes with a lag; early valuation prices the future
Cost is unsustainable	Infrastructure and energy cost far above revenue	Unit costs fall fast; infrastructure is lasting value even if the bubble bursts
Return unproven	Net enterprise-scale ROI not common, pilots stall	Adoption is very fast and real; when value is measured there are apps with net ROI
Just narrative	'AI' label became a marketing word, expectation swelled	Millions use it daily and voluntarily; usage is real independent of narrative
Correction is near	In every tech wave expectation swells then falls	Correction hits price, not the technology; a lasting productivity plateau follows the trough

The most important message of this table is that in each row neither side is entirely right or entirely wrong. Valuations can really be excessive and at the same time the technology can really be valuable. This seeming contradiction is, in fact, the nature of the bubble debate: price and value are not the same thing. An asset's price can be excessive; that does not show it is worthless, only that today's price over-discounts its future value.

The practical lesson for organizations from this table is this: you do not have to know which side "wins." It is enough to see the tension in each row and make your own decision aware of that tension. In the coming sections we will make this decision concrete; but first let us look at a powerful perspective that explains the debate on the time axis.

The concept that frames the AI bubble debate most illuminatingly is the expectations curve (hype cycle). The expectations curve is a pattern describing how society's expectation of a new technology changes over time, and it lifts the bubble debate out of the "bubble or not" dilemma into a much richer time perspective.

The curve works roughly like this. First a trigger comes: a new capability becomes visible and draws attention. Then expectation rises rapidly and reaches a peak of enthusiasm; at this point the narrative is far ahead of reality, presented as the solution to every problem. Then the inevitable disappointment comes: when the technology cannot immediately do everything hyped, expectation collapses and descends to a trough of disappointment; it is in this phase that the "so it was a bubble" reaction rises. But the story does not end here: after the trough the technology quietly matures and reaches a real, measured plateau of productivity. On this plateau the narrative has faded but real productivity is at its highest.

The most valuable insight this perspective adds to the bubble debate is this: excessive expectation and genuine transformation do not exclude each other; they are different points on the same curve. The narrative at the peak of enthusiasm may be unrealistic — this is true, those who say bubble are right. But the same technology can, years later, pass the trough of disappointment and reach the plateau of productivity — this is also true, the other side is right. The two sides speak from different points in time. Those who say bubble look at the peak, the other side at the plateau.

The expectations curve also contains a warning: the trough of disappointment is real, and in this phase many ventures, many investments, and many projects will be eliminated. So "the technology is lasting" and "all these investments will return" are not the same thing. The curve says at once that the technology will survive but that many players will fall along the way. A realistic view accepts that both the plateau and the trough are real.

Past technology bubbles provide a strong reference when evaluating the AI bubble debate, because patterns repeat. Two historical examples are especially instructive: the dot-com bubble and, much older than it, the railway bubble. Both show that a bubble hits not the technology but timing and price expectations.

The dot-com bubble taught this: the internet was a real and transformative technology; yet many companies reached astronomical valuations without revenue, on the expectation given merely by being an "internet company." When the correction came most of these companies were wiped out — but the internet itself did not vanish; it developed and became the backbone of today's economy. The lesson is clear: a bubble bursting does not show the underlying technology is worthless; it only shows that players with no revenue model, relying on narrative alone, will be eliminated. After the bubble burst, the winners were those producing real value with a measurable business model.

The railway bubble is older still and even more instructive. During the railway mania speculative capital flowed in bulk into laying tracks; many investors lost their money because prices had run ahead of real return. But the tracks left behind — physical infrastructure — carried the economy for decades and became the backbone of industrialization. So the bubble drew capital into infrastructure; when the froth subsided the infrastructure remained. This is the historical grounds for the "infrastructure leaves lasting value" counterargument in the AI bubble debate.

The common lesson from these two examples is threefold. First, a bubble and a technology are separate things; the bubble bursts, the real technology remains. Second, infrastructure investment often leaves lasting value behind; so investment in infrastructure is more durable than investment in speculative business models. Third, the winners are the players relying not on enthusiasm but on real revenue and measurable value. These three lessons guide organizations directly: invest by value not by narrative, prioritize lasting layers (infrastructure, capability) where possible, and measure return.

But historical analogies have a limit, and this must be said honestly: unlike many ventures of the dot-com era, AI today already produces measurable productivity with a broad and real user base. So while the similarities are instructive, the real usage on the adoption layer separates this wave from pure speculation. Past bubbles give us the pattern; but each wave also has its own specific conditions. A realistic view uses the analogy but does not worship it.

The most common conceptual confusion that muddies the debate is mistaking "bubble" and "overinvestment" for the same thing. Yet these are different phenomena, and seeing the distinction lets you understand the overvaluation debate much more clearly. Overinvestment in an area does not necessarily mean there is a bubble; and a bubble does not always mean the total investment is waste.

Overinvestment is more capital flowing into an area than it can absorb in the short term. This is an almost natural feature of new technology waves: when the future is uncertain, capital, unable to know the winner in advance, invests in many players at once. Part of this "excess" investment is wasted (the losing players), but part produces lasting value (the winners and the infrastructure). So overinvestment produces both waste and progress; the two are parts of the same process. In economists' language this is partly "creative destruction."

A bubble is a narrower phenomenon: the price of an asset departing from its real value on expectation. The distinguishing feature of a bubble is that the price rests largely on the expectation that "others will buy at a higher price" — not on fundamental value. In an area there can be both overinvestment and a bubble; but it is also possible to have overinvestment without a bubble (if the return is lagged but real), or for the investment to be reasonable while a narrative bubble still forms.

Why does this distinction matter? Because from the standpoint of enterprise decision-making the response to the two phenomena differs. If there is overinvestment, the right strategy is to stay calm and wait for the winners to separate, because infrastructure and real value are lasting. If there is a pure price bubble, the right strategy is to stay away from speculative pricing but keep using the technology. An organization that mixes the two either panics and misses the real opportunity or gets swept into enthusiasm and takes on speculative risk.

Bubble, overinvestment, and genuine transformation: distinguishing three phenomena

Phenomenon	What it means	Right enterprise response
Price bubble	Price departed from fundamental value on expectation	Stay away from speculative pricing, keep using the technology
Overinvestment	More capital flowed in than the area can absorb short-term	Stay calm, wait for winners to separate, trust infrastructure
Genuine transformation	The technology produces measurable and lasting value	Measure in your own context, scale what works

In real life these three phenomena usually appear intertwined: a price bubble, overinvestment, and a genuine transformation can exist together at the same time. The difficulty and richness of the AI bubble debate is exactly this. An organization that can distinguish the three escapes the trap of both panic and enthusiasm and makes its decision on realistic ground.

The financial heart of the AI bubble debate lies in understanding how the investment cycle works. In new technology waves, capital flows not randomly but in a recognizable pattern; seeing this pattern gives a realistic frame for "where does the money flow and when does it return."

The typical investment cycle is layered. At the bottom is the infrastructure layer: compute, data centers, chips, energy. This layer is the most capital-intensive and its return is the longest-term; but it is also the one that leaves the most lasting value even if the bubble bursts. Above it is the model and platform layer: foundation models, development tools, infrastructure services. At the top is the application layer: products solving specific business problems. Capital usually flows in bulk into the infrastructure and model layers first, then winners separate in the application layer.

The most misleading side of this cycle is timing. The return of a general-purpose technology comes with a lag: investment is made first (cost up front), and value emerges only when the technology matures and applications spread (return at the back). This lag produces, in the middle of the cycle, the reaction "look, there is still no return, so it is a bubble." Yet the lag can be a sign not of a bubble but of the nature of general-purpose technologies. When electricity spread, the productivity gain too appeared years later.

So when does the money return? The honest answer: it varies by layer and player, and some of it never returns. Most of the investment in infrastructure leaves lasting value however the bubble debate resolves. The winning application-layer players produce strong returns. But a significant part of the investment in players with no revenue model, relying on narrative alone, does not return — and this is not a failure but the normal working of the cycle. The investment cycle always produces both winners and losers; the existence of losers is not proof of a bubble but a sign of separation.

The practical lesson for organizations from this cycle is clear. First, investment in lasting layers (infrastructure, capability, data) is more durable, because its value remains however the cycle resolves. Second, narrow projects producing measurable value on the application layer are safer than broad, uncertain commitments. Third, the return coming with a lag is normal; so instead of expecting return in the very short term, one must set up the right measurement and be patient. We cover the general framework for how to assess return in the AI risk assessment document and the maturity-based approach in digital maturity.

Now let us come to the most practical question of the debate: as an organization's manager, how should the AI bubble debate change your decision? The short and perhaps surprising answer is this: in a well-designed strategy, the outcome of the debate should not greatly change your decision. Because a sound approach is designed to survive whether the bubble inflates or bursts.

The key to this is deciding by measurable value in your own context, not by narrative. The side of the bubble debate is a distant and general question; whereas your question is concrete and local: "does this application produce measurable value in this process of mine?" If you can answer yes, a bubble existing in the world does not change the value you get from that application. If you answer no, the world's enthusiasm does not make that application valuable for you. The decision must rest on internal measurement, not external narrative.

Several concrete rules of conduct follow from this principle. First, proceed with narrow-scope, return-measured pilots; instead of "let us transform everything," prove measurable value in a single process. Second, tie every project to a business outcome; not technology for technology's sake but an application that improves a defined metric. Third, avoid large, irreversible commitments; proceed with incremental, reversible steps. Fourth, avoid vendor lock-in; keep components replaceable so you stay flexible however the cycle resolves.

Fifth and most important: invest in capability. Technology, model, and vendor can change; but people and process capability is lasting and produces value independent of the bubble debate. Your team's competence in reading AI correctly, verifying it, and tying it to a business problem works whichever model wins. So even in the middle of the bubble debate, the safest investment is the one in people and learning. We cover capability transformation in career and skill transformation in the AI age and enterprise learning structure in the enterprise AI academy.

So far we have covered the two sides of the debate, the expectations curve, and the practical takeaway. This final framework turns all of this into a thinking tool you can apply to your own organization. The aim is not to give a general answer to "is there an AI bubble"; it is to let you read which decisions are right in your specific context.

The framework is built around four questions. The first is the layer question: which layer is the thing you are talking about on? An investment decision can be risky on the finance layer (valuation, speculation) while using the technology in your own process is safe on the adoption layer. Every decision made without clarifying the layer is blurry. The second is the value question: does this application measurably improve a defined metric of mine, or does it only look impressive? Distinguishing between an impressive demo and measurable value is the key to escaping the bubble trap.

The third is the time question: where on the expectations curve is this application for me? Is it still at the peak of enthusiasm (everyone talking, no one showing a clear result), or on the plateau of productivity (repeatable, measured value)? Be cautious with applications at the peak, invest in those on the plateau. The fourth is the durability question: what happens to this decision if the bubble bursts? If your decision is valuable only because the narrative is true, it is fragile. If it is valuable because it solves a real problem, it is durable. Durable decisions survive independent of the outcome of the debate.

Asking these four questions in every investment decision takes you out of both panic and enthusiasm and gives you a realistic view. A realistic view neither belittles nor deifies the technology; it sees it as a tool and evaluates every tool by the single question it asks: "does it measurably solve my defined problem?" This question holds whether or not there is a bubble in the world, and only your context can give the right answer.

The balanced evaluation framework: four questions and what they look for

Question	What it asks	Sign of a sound answer
Layer	Is the thing I am talking about finance, technology, or adoption?	Decision made by the risk of the right layer
Value	Does it measurably improve a defined metric?	A measured result, not an impressive demo
Time	Where am I on the expectations curve?	Invest on the plateau, caution at the peak
Durability	Does this decision survive if the bubble bursts?	Value rests on a real problem, not narrative

The beauty of this framework is that it is applicable to any new technology wave. Even if the AI bubble debate gives way to another debate tomorrow, the "layer, value, time, durability" questions retain their validity. Because these are questions about decision discipline, not about the technology. A realistic organization keeps this discipline even as the narrative changes.

Up to here we have built the framework; now let us turn it into a concrete sequence of steps. The roadmap below summarizes the practical steps an organization can follow to build an evidence-based and durable approach, unaffected by the noise of the AI bubble debate.

How to

A durable roadmap in the face of the bubble debate

Practical steps for an organization to build an evidence-based and durable AI approach independent of the AI bubble debate.

1. 1

   Separate narrative from context

   List not the general bubble debate but the concrete problems in your own processes; tie the decision to internal need, not external enthusiasm.

2. 2

   Choose a narrow, measurable pilot

   Start with a narrow-scope pilot that produces measurable value with a defined metric in a single process; avoid the goal of 'transforming everything'.

3. 3

   Define the return up front

   Before starting the pilot, decide which number will show success and measure the current baseline; unmeasured value cannot be argued.

4. 4

   Prioritize lasting layers

   Instead of speculative business models, invest in layers that leave value even if the bubble bursts, like infrastructure, data, and capability.

5. 5

   Avoid vendor lock-in

   Keep components replaceable; avoid binding irreversibly to a single model or vendor.

6. 6

   Measure, learn, scale

   Measure the pilot's return; if it does not work, stop; if it does, scale incrementally. Grow by evidence, not enthusiasm.

The most important feature of this roadmap is that it does not depend on the outcome of the bubble debate. If the bubble bursts, you are left with an application that solves a real problem and whose value is measured. If the bubble keeps inflating, you are protected because you did not get swept into enthusiasm and enter irreversible commitments. In both scenarios the winner is the organization that measures value in its own context, not the one that follows the narrative.

One point must be underlined: this roadmap does not mean "be slow." On the contrary, narrow, measured pilots usually produce value faster than broad, uncertain programs, because the learning loop is short. A realistic view does not see caution and speed as opposites; it advocates evidence-based fast movement. To design your enterprise roadmap according to your context, you can start with AI consulting, and review corporate training options for your teams' competency.

To fully understand the AI bubble debate we must look not only at economics but at human psychology, because bubbles are fundamentally collective psychological phenomena. Both the inflation and the deflation of a bubble are determined by people's shared emotions and reflexes as much as, and often more than, technical facts. Recognizing this psychology enables both reading the narrative more clearly and cleansing your own decisions of these reflexes.

The first psychological force is FOMO (fear of missing out). In a new technology wave, managers' biggest fear is falling behind while competitors catch an advantage. This fear triggers investments made without measurement or justification: "everyone is doing it, so we must too." FOMO is the strongest psychological fuel of the overvaluation debate, because it feeds price not from fundamental value but from "fear of missing out." Decisions made with FOMO share a common feature: their justification is external movement, not internal need.

The second force is herd behavior. Under uncertainty, people use others' behavior as a source of information: "if this many smart investors are investing, they must know something." This creates a self-reinforcing loop — everyone joins the flow assuming others are right, and no one questions fundamental value. Herd behavior amplifies both the swelling in the enthusiasm phase and the sudden collapse in the panic phase. The sharp descent of bubbles often stems not from real bad news but from the herd changing direction.

The third, often overlooked force is the counter-reflex. As a reaction to enthusiasm, an excessive and generalizing pessimism also forms: "it is all a bubble, none of it is real." This counter-reflex is as misleading as the enthusiasm, because it too erases nuance and ignores the technology's real value. Healthy skepticism and wholesale denial are different things. A realistic view neither joins the herd nor opposes the herd blindly out of spite; it recognizes both reflexes and sets them aside, and looks at the evidence.

The lesson of this psychology for enterprise decisions is this: when making your own decision, ask "is this decision driven by FOMO, by the herd, or by a measured need?" If your justification is "everyone is doing it" or "let us not fall behind," you are most likely deciding by narrative. If your justification is "it improves this metric in this process by this much," you are deciding by evidence. This simple internal questioning is the most practical way to escape the traps of bubble psychology.

A significant part of the confusion in the AI bubble debate actually stems from two entirely different questions being asked with the same words. "Is AI a bubble?" means completely different things depending on who is asking; and every debate held without separating these two questions misunderstands itself.

The first question is the investor question: "is it wise to invest in AI companies at today's prices?" This is an asset-pricing question; it is about valuation, timing, risk, and return. The answer to this question is largely on the finance layer, and saying "yes, some prices are excessive" is consistent. For an investor a bubble is a real risk, because a price correction directly hits their capital. The cautious answer to the investor question is a requirement of a realistic view.

The second question is the user question: "as an organization, does using AI in my processes add value to me?" This is a tool-utility question; it is about productivity, quality, cost, and business outcome. The answer to this question is largely on the technology and adoption layers, and saying "yes, used correctly it adds value" is consistent. For a user a bubble is not a direct threat, because if the tool they use solves a real problem, that value remains in place even if market valuations fall.

Here is the critical point: a person can say "yes, there is a bubble" to the investor question and at the same time "no, the technology adds real value" to the user question — and this is not at all contradictory. Because the two questions are on different layers. The investor asks about price, the user about utility. Most of the bubble debate consists of people who confuse these two questions misunderstanding each other: one speaks of price, the other of utility, and they appear to be arguing about the same thing.

For most organizations, the directly relevant one is the user question. An industrial company, a hospital, or a municipality is not investing in the share price of AI companies; their question is "how do I turn this technology into value in my own work?" So for these organizations the right reading of the bubble debate is to set aside investor panic and enthusiasm and focus on the user question. Whatever the price bubble in the market, the measured value in your process is real and should be the foundation of your decision.

The most concrete and least-discussed dimension of the cost argument of those who say bubble is the energy and physical resource side. AI infrastructure is expensive not only financially but physically: compute consumes electricity, data centers require cooling and space, hardware production demands resources. Understanding this dimension deepens both the cost argument and the counterargument to it.

The energy dimension of the cost argument is this: if there is significant energy consumption underneath every AI query and the value produced by that consumption is uncertain, this imbalance is unsustainable in the long run. This argument is strong because it rests on physical reality; it is measured not by narrative but by resources. And this dimension is not only a financial matter but also a sustainability matter: resource consumption carries an environmental and social cost too.

The counterargument to this is two-layered. The first is the efficiency trend: the compute cost and energy consumption per unit historically tend to fall rapidly as the technology matures. Doing the same task with ever fewer resources becomes possible; hardware, model, and software efficiency improve together. So today's high unit cost may be a sign not of a lasting feature of the technology but of its early phase. The second is the net-effect argument: if an AI application saves more energy or resources than it consumes (for example by optimizing a process), the net effect can be positive.

But we must be honest: these counterarguments rest on a trend and a potential, not a guarantee. Efficiency can indeed increase rapidly, but this is not automatic; the net effect can indeed be positive, but this must be measured. So the energy and sustainability dimension is one of the areas of the bubble debate that most needs balanced handling: neither the "the energy problem will end everything" pessimism nor the "efficiency will solve it anyway" optimism is right on its own.

The practical takeaway of this dimension for organizations is this: when assessing the cost of an AI application, consider not only the license or usage fee but also the real resource cost behind it and its trend over time. Applications that produce measurable value, whose resource efficiency increases, and whose net effect is positive, stand on solid ground however the bubble debate resolves. Sustainability is now an inseparable part of technology decisions; and a realistic view measures cost not only by the bill but also by resources.

Although the AI bubble debate is spoken of as a general narrative, in reality it carries different meanings for each sector. The answer to "is it a bubble" is one thing for a software company, another for a hospital, and yet another for an industrial enterprise. Seeing this differentiation is another way of reducing the debate to your own context.

In high-maturity, data-intensive sectors (for example finance, telecom, e-commerce) AI has long entered operations and produces measurable value; here the bubble debate is more about the finance layer, that is, provider valuations, not the technology itself. For an organization in these sectors the right question is not "is it a bubble" but "which application should I scale." For them the debate is largely behind; the technology is already part of daily work.

In medium-maturity, regulated sectors (for example healthcare, public sector, insurance) the situation is more nuanced. Here the technology's potential is high, but adoption is slower because of trust, compliance, and verifiability requirements. For an organization in these sectors the bubble debate feeds the dilemma of "should I hurry or should I wait." The right answer usually varies with the expectations curve: as regulation and trust mature, value materializes, so being cautious but prepared — learning with pilots and standing ready to scale — is the soundest stance.

In traditional, physical-process-heavy sectors (for example manufacturing, construction, logistics) AI often comes as an auxiliary layer: visual inspection, forecasting, optimization. Here the "bubble" narrative usually finds less echo, because value is easy to measure when it can be tied to a concrete operational improvement. For an organization in these sectors the debate is almost entirely a user question: "does this application provide a measurable improvement on my production line, in my field, in my warehouse?"

The common lesson of this sectoral differentiation is this: you must read the general bubble narrative multiplied by your own sector's maturity level and value equation. The same headline can mean "do not fall behind" for you or "do not hurry"; what determines which is your sector's concrete relationship with the technology. For a sector-specific reading we cover the maturity-based approach in digital maturity and organizational readiness in the enterprise AI academy.

The practical value of the bubble debate lies in being prepared for the question "what if it bursts?" But a "burst" is not a single thing; it can happen in different forms and intensities. Thinking through three possible scenarios lets you see where your organization stands in each and is the most concrete test of a realistic view.

The first scenario is soft alignment: expectation aligns with reality gradually rather than in a sharp collapse. The narrative fades, speculative prices correct, but the technology quietly continues to mature and produce value. This is the classic course of the expectations curve — a soft descent from peak to plateau. In this scenario, the organization that invested in measurable value continues uninterrupted; only those who invested in narrative alone experience disappointment. In soft alignment the loser is the one swept into enthusiasm; the winner is the one who measures.

The second scenario is a sectoral correction: a sharp valuation correction occurs on the finance layer, many speculative players are eliminated, and investment slows for a while. But just like after dot-com, the technology and infrastructure remain; indeed, after the correction the winners producing real value become more clearly visible. In this scenario your organization's position depends on how much you invested in lasting layers (infrastructure, capability, data). An organization not locked to a vendor and having built its capability internally weathers the correction relatively soundly, because its value is not tied to a single player.

The third scenario is a long winter: expectation breaks very sharply, investment dries up for a long time, and progress slows. This scenario is less likely because a real and broad usage base already exists; but it is not wholly impossible and one must be prepared. Even in a long winter, applications that solve a real problem and whose cost is sustainable survive, because their value is tied not to narrative but to a working business outcome. The victims of a long winter are those with no revenue, dependent solely on future investment.

The common lesson of these three scenarios is striking: in all three, the same strategy protects the organization. Investing in measurable value, prioritizing lasting layers, avoiding vendor lock-in, and building capability internally — these four principles survive whatever the scenario. This is exactly why the best answer to "what do I do if the bubble bursts" is not to try to predict the burst but to build a strategy that needs no prediction. A realistic view focuses not on knowing the future but on being durable in every future.

There are a few thinking mistakes that people following the debate and making enterprise decisions often fall into; seeing them enables both reading the debate more clearly and making sounder decisions.

• Binary thinking: Asking "bubble or not" and seeking a one-word answer. The reality is layered: an overvaluation on the finance layer can be real while a genuine leap on the technology layer can also be real. Failing to accept both answers at once is the most common mistake.
• Confusing price with value: An asset's price being excessive does not show it is worthless. A bubble is a problem of price, not of value. Every judgment made without separating these two is misleading.
• Deciding by narrative: Making an investment decision based on external enthusiasm or pessimism. The right decision rests on measurable value in your own context, not on external narrative.
• Mistaking return lag for a bubble: In general-purpose technologies return comes with a lag; saying "there is still no net ROI" does not always mean "so it is a bubble." The lag can be a natural feature of the cycle.
• Worshipping the historical analogy: The dot-com or railway analogy is instructive but not identical. Using the analogy for the pattern is right; copying the outcome mechanically is wrong.
• Swinging to one of the two extremes: Both indifferent denial and unbounded enthusiasm are mistaken. A realistic view is in the evidence-based, incremental middle.

The AI bubble debate directly affects not only investors and organizations but also ordinary employees and individuals planning their careers; because the question "is it a bubble" intertwines with "is this skill worth learning" and "is my job at risk." For individuals too the right frame is the same as for organizations: look not at the narrative but at the real value in your own context.

For an employee the bubble narrative produces two opposing panics. On one hand the "every job will disappear" narrative creates excessive anxiety; on the other the "it is all a bubble, never mind" narrative creates the risk of missing a genuine skill transformation. Both are extremes. A realistic view sees that the technology really does change some tasks, but that this is a gradual shift at the task level rather than a sudden, wholesale disappearance. Both panic and indifference misguide the individual.

The practical takeaway for the individual, just as for the organization, is to prioritize durable investments. Whichever model wins, whether the bubble inflates or bursts, there are competencies that retain their value: defining a problem clearly, critically verifying the technology's output, tying different tools to a business outcome, and a habit of continuous learning. Because these competencies rest not on a particular product but on a way of thinking, they work wherever you are on the technology wave. We cover them in detail in career and skill transformation in the AI age.

The individual's biggest mistake in the face of the bubble debate is deciding by external narrative: "everyone is learning AI, I must hurry too" (FOMO) or "it is all a passing fad, I need not learn it" (counter-reflex). The right question is none of these; the right question is "in my job, in my role, which tasks does this technology change, and with which competency do I increase my value?" This question leads to a personal and measurable answer independent of external noise.

In short, the bubble debate is for individuals too not a paralysis but an opportunity to find direction. Reading the real change in your own role without being swept into the narrative, and investing in durable competencies, makes you strong however the debate resolves. With a real and realistic view, even if the bubble bursts these competencies stand as lasting value in your career.

The AI bubble debate runs largely through media and social networks; and in these channels headlines of both "the bubble is bursting" and "everything is changing" can stand side by side on the same day. For a manager the real skill is to read this flood of narrative with a critical eye and separate the signal from the noise. Here are a few practical questions to ask when evaluating this narrative.

The first question is the source's position: who is producing this content, and with what interest? An investor, a provider, a critic, or a journalist — each brings a viewpoint from a different layer and a different incentive. A provider's "everything is changing" narrative and a short-selling investor's "the bubble is bursting" narrative can both be colored by interest to the same degree. When reading the narrative, seeing the incentive behind the claim is the first step to weighing it correctly.

The second question is the type of evidence: what does this claim rest on — measured data, an anecdote, or a prediction? "A company saved this much" is concrete evidence; "AI will soon take every job" is a prediction; "a friend's trial did not work" is an anecdote. All three enter the debate but their weights differ greatly. A sound reading distinguishes the type of evidence and does not confuse a prediction with measured data.

The third question is the degree of generalization: how broad an area does the claim sweep? A sweeping generalization like "AI is a bubble" is almost always an oversimplification, because it puts the technology's different layers and application areas into a single basket. The narrower and more specific a claim (in a particular application, a particular sector, a particular metric), the more testable and the more valuable it is. A natural skepticism toward broad generalizations is the protective reflex of a realistic view.

When you apply these three questions — source, evidence, generalization — to every headline, you will find that most of the flood of narrative filters itself out. What remains is content that is narrow, evidenced, and transparent about its incentive; and that is exactly what is valuable for a decision. Filtering the narrative this way protects you from both the wave of enthusiasm and the wave of pessimism and directs your attention to what really matters — measurable value in your own context.

In the bubble debate two opposing narratives constantly clash, and both contain a thinking trap. The first narrative is "this time is different": the advocates of every new technology say the lessons of past bubbles do not apply this time, because this technology is truly unique. The second narrative is "same thing": critics say this is an exact repeat of past bubbles and will end the same way. Both are partly true and partly a trap.

The "this time is different" trap is known as the most expensive sentence in the history of finance, because it is usually used at the peak of the bubble to justify no longer questioning fundamental value. The advocate of every bubble has said "the rules do not apply this time," and most have been wrong. So when you hear this sentence, a natural alarm bell should ring: maybe it really is different, but the claim requires evidence, not enthusiasm.

But the "same thing" narrative is also a trap: a mechanical historical analogy forces every new wave into the mold of the past and ignores real differences. Each technology wave has its own specific conditions; adoption speed, real usage base, revenue model, and maturity level can differ. Saying "it is all the same" is as lazy a thought as saying "it is all different"; both resort to a template instead of looking.

The way beyond these two traps is neither to worship the analogy nor to reject it entirely; it is to use the analogy as a hypothesis and test it with evidence. "Past bubbles showed this pattern; where on that pattern is this wave, at which point does it resemble it, at which point does it diverge?" This question both uses the historical lesson and stays open to today's unique conditions. A realistic view sees both narratives as tools and surrenders to neither.

The practical takeaway of this debate for organizations is this: stop seeking a general answer to "is this time different"; instead ask "which one holds in my context." In some application areas there is genuinely lasting and different value; in others the familiar bubble pattern of the past repeats. Distinguishing the two with your own measurement is always sounder than binding blindly to one of the two narratives.

The most concrete test of all this debate happens at the budget table. A manager must, under the noise of the AI bubble debate, decide how to allocate a limited budget. It is at this decision moment that the framework we have built so far turns into a practical tool, and the narrative comes down to a concrete allocation decision.

The first principle is to distribute the budget not into a single big bet but into measured small bets. Under the uncertainty of the bubble debate, putting the entire budget into a single broad transformation program is fragile both if the bubble bursts and if the technology matures more slowly than expected. Instead, a budget split across several narrow, return-measured pilots both distributes the risk and speeds up learning. Winning pilots are scaled, losing ones are stopped early; the budget is reallocated by evidence.

The second principle is to strike a conscious balance between lasting layers and speculative layers. A portion of the budget should go to layers that leave value however the bubble debate resolves (capability building, data infrastructure, process improvement), because these investments remain with the organization whatever the technology wave does. A portion can go to application trials that carry higher uncertainty but high return potential — but this portion must be kept at a size whose loss is affordable.

The third principle is that every budget item has an exit criterion. Defining, from the start, when you will stop an investment is as important as starting it. The sentence "if this pilot does not improve this metric by this much in this time, we stop it" saves the budget from being at the mercy of the narrative. Investments with no exit criterion are the easiest prey of bubble psychology, because FOMO and herd behavior enlarge losses by saying "let us wait a little more."

These three principles — distributed small bets, the lasting-speculative balance, defined exit criteria — let a manager turn the bubble debate from a source of paralysis into a manageable decision problem. Ultimately the budget decision does not require knowing whether the world is a bubble; it only requires placing evidence-based, distributed, and exit-defined bets in your own context. With this discipline, however the debate resolves, your budget works in the organization's favor.

Is AI a bubble?

The short answer is not "yes" or "no" but "partly and depending on the layer." In the AI bubble debate the most honest answer is to accept that in some layers of the market (unproven business models, valuations given without real revenue, narrative-based pricing) an overvaluation is real; but that the technology itself has also undergone a real and lasting leap in capability. Inflated expectation and genuine transformation can coexist. The existence of a bubble does not mean the technology is worthless; it only means price and timing expectations have run ahead of reality.

Is AI investment excessive?

In some areas probably yes, in others no. That is the nature of the investment cycle: in a new technology wave capital first flows in bulk and indiscriminately, then winners and losers separate. Most of the investment in infrastructure (compute, data centers, energy) leaves lasting value however the bubble debate resolves. By contrast, some of the investment in ventures relying on narrative alone, without a real revenue model, will most likely not return. There is no single answer to "is it excessive"; it varies by layer, business model, and time horizon.

What should organizations do if the AI bubble bursts?

The right strategy is not to try to predict how the bubble debate resolves, but to make decisions that survive whatever the outcome. That means investing by measurable value in your own context rather than by narrative: starting with narrow, return-measured pilots; tying every project to a business outcome; avoiding vendor lock-in; and building capability independent of the technology. Even if the bubble bursts, an application that solves a real problem and whose value is measured retains its worth. The right answer is the same in both scenarios: follow an evidence-based, incremental, measured path.

Is the AI bubble the same as the dot-com bubble?

There are similarities but they are not identical. Similarity: in both, a real and transformative technology combined with an unrealistically inflated short-term market expectation. In the dot-com era the internet was real and lasting; but many companies' valuations were not backed by revenue, and when the correction came those companies were wiped out while the internet remained. Difference: unlike many dot-com-era ventures, AI today already produces measurable productivity contribution and a broad base of real users. The common lesson of past bubbles: a bubble hits not the technology but timing and price expectations.

How does the expectations curve explain the AI bubble debate?

The expectations curve (hype cycle) describes how society's expectation of a new technology changes over time: first a rapid peak of enthusiasm, then a trough of disappointment, and then a realistic plateau of productivity. This perspective lifts the bubble debate out of the "bubble or not" dilemma: the technology can be real even while the narrative is at its peak; value does not vanish even when disappointment arrives, expectation merely aligns with reality. So the expectations curve shows that excessive expectation and lasting transformation are different points on the same curve.

What should a realistic AI investment strategy look like?

A realistic view is neither being swept into bubble talk and staying indifferent, nor spending without measuring in unbounded enthusiasm. A sound strategy rests on these principles: tie every investment to a business outcome and a measurable metric; proceed with narrow, incremental pilots instead of large, irreversible commitments; prioritize layers that leave lasting value independent of the bubble debate, like infrastructure and capability; avoid vendor lock-in; and measure return continuously. This discipline protects the organization whether the bubble inflates or bursts.

In short: the AI bubble debate is not a question to be answered in one word with "bubble or not." The right frame is to separate the debate into layers: an overvaluation on the finance layer can be real while a genuine and measurable transformation on the technology and adoption layers can also be real at the same time. The arguments of those who say bubble — overvaluation, cost imbalance, unproven return, narrative inflation — are mostly about price and timing; the counterarguments — a real leap in capability, fast adoption, lasting infrastructure — are mostly about technology and usage. Both are partly right.

The expectations curve unites this picture: excessive expectation and lasting transformation are different points on the same curve, and past technology bubbles teach that a bubble hits not the technology but timing and price expectations. The practical takeaway for organizations is not to predict the outcome of the debate but to make decisions that survive whatever the outcome: investing by measurable value in your own context, not by narrative; proceeding with narrow, measured pilots; prioritizing lasting layers and capability. This is the essence of a realistic view — neither indifferent denial nor unbounded enthusiasm, an evidence-based and incremental discipline.

To underline it one last time: being right in the AI bubble debate comes not from picking the correct side in advance but from building a discipline that survives whichever side prevails. No one can predict the peak or the trough exactly; but an organization that measures value in its own context, invests in lasting layers, and defines its exit wins without needing any prediction. This is exactly what a realistic view is: not a claim to know the future but a discipline of being durable in every future. An organization that internalizes this discipline once will stand on the same solid ground in the bubble debate of the next technology wave too.

To deepen the core concepts you can see career and skill transformation in the AI age, digital maturity, and LLM cost optimization. To design an evidence-based roadmap tailored to your organization, independent of the bubble debate, you can start with AI consulting, review corporate training options for your teams' competency, and deepen all concepts in the learning center. To follow a balanced and current perspective, you can join the newsletter and reach evidence-based content instead of the noise of the debate.