The AI “bubble” narrative persists because people are diagnosing the wrong thing. This isn’t a bubble. It’s the largest energy-driven infrastructure expansion in modern history.
getty
Why The AI Bubble Isn’t What It Seems
Michael Burry is shorting the AI market. The same investor who anticipated the subprime crisis now warns that we are living through an AI bubble defined by runaway valuations, aggressive speculation, cheap capital and faith in endless growth. The comparison to 1999 and 2008 is tempting. But this time he may be aiming at the wrong target.
The rise of AI is not being built on bad mortgages or inflated advertising metrics. It is being built on something far more fundamental: energy. Vaclav Smil, a leading scholar of energy systems and a thinker Bill Gates often cites as foundational, has written that energy is the universal currency of civilization. Every model, every inference, every output we describe as intelligent is ultimately a transformation of electricity into structured probability. When Burry shorted housing, he was betting against human leverage. When he shorts AI, he is betting against thermodynamics, and thermodynamics tends to win.
The problem is not that Burry is uninformed. The problem is that the mental models used to evaluate AI are outdated. This is not a bubble in any conventional sense. It only resembles one if we force AI into the logic of traditional software.
But AI does not behave like software. Its economics resemble the economics of infrastructure. Valuations may appear disconnected from productivity. Capital may look like it is circulating in a self-reinforcing pattern. Spending may appear excessive. Yet these dynamics appear irrational only through the lens of consumer technology.
Once energy becomes the primary input to intelligence, the logic changes. The true cost of intelligence is measured in watts and in the physical capacity to direct energy toward computation.
The Misread AI Bubble Narrative
The irony is that the “AI bubble” narrative is itself a bubble, inflated by people applying outdated analogies to a phenomenon that does not fit them. Critics point to OpenAI’s operating losses, its heavy compute requirements and the fact that its expenses dwarf its revenues.
Under classical software economics, these would indeed be warning signs. But AI is not following the cost structures of apps or social platforms. It is following the cost structures of infrastructure.
The early electrical grid looked irrational. The first telephone networks looked irrational. Railroads looked irrational. In every major infrastructural transition, society endured long periods of heavy spending, imbalance and apparent excess. These were not signs of bubbles. They were signs that the substrate of daily life was being rebuilt.
OpenAI’s spending is no more indicative of a bubble than Edison’s power stations or Bell’s early switchboards. The economics only appear flawed if one assumes the system they are building already exists.
What we are witnessing is not a speculative mania but a structural transformation driven by thermodynamics, power density and a global shift toward energy-based intelligence.
The bubble narrative persists because many observers are diagnosing this moment with the wrong conceptual tools. They are treating an energy-driven transformation as if it were a software upgrade.
The history of technological revolutions is full of high-profile misreads. In 1998, Paul Krugman declared that the growth of the internet would slow drastically and that by 2005 it would prove no more economically significant than the fax machine. His prediction failed because the framework on which it was based failed. The internet did not fall short of expectations, the expectations were built on the wrong mental model.
Much of today’s bubble discourse about AI suffers from the same problem.
That misunderstanding becomes even clearer when you look at the numbers under the surface. The world is not guessing about AI demand; it is signing for it. Enterprises and governments are locking in multi-year AI contracts, committing billions to the infrastructure that will power their operations over the next decade.
Speaking about the MAG6 – Microsoft, Apple, Google, Amazon, Meta and Nvidia, Michael Pecoraro, Head of Investment Risk at Voya Investment Management, told me, “The MAG6 are generating real revenue growth backed by multi-year contracts. That’s not speculative demand — that’s actual demand.”
His point is simple: In the late 1990s, investors were projecting demand. In the AI boom, demand is contractually guaranteed. That difference alone makes the historical analogy more complicated than most commentary suggests.
The Dot-Com Parallel That Clarifies The AI Bubble Debate
The closest historical parallel is not railroads or telephones. It is the dot-com boom.
That period is remembered as a bubble because thousands of companies failed, but that interpretation misses the larger truth. The infrastructure built during that frenzy created the modern internet. Much of today’s economic output is driven by a small group of winners, companies like Amazon, Google,and Meta, that emerged from the wreckage and now define the S&P 500.
The fact that most firms went to zero did not mean the internet was a bubble. It meant we could not yet see who the winners would be. The same principle applies to AI. Not every company needs to survive for the underlying transformation to be real.
Dot-com created dark fiber, which we didn’t yet know how to use. AI is the opposite: The infrastructure is already saturated before it is even built. And just as the S&P today is dominated by a small group of winners few could have predicted in 1999, the same will likely be true for AI.
Apple’s dominance rests on an ecosystem made possible by internet-era infrastructure, something virtually no one foresaw during the dot-com frenzy. In that era, many believed the safest bet was Cisco because it supplied the plumbing for the entire networked world. In hindsight, that conviction proved misplaced.
Nvidia occupies a similar position today: the seemingly obvious “picks and shovels” choice. Maybe that intuition is correct. Perhaps it isn’t. The point is that the ultimate winners of a technological revolution are almost always visible only in retrospect. AI will produce its own giants, companies so deeply embedded that they reshape the S&P, but we won’t recognize them until the dust settles.
The difference is that the internet’s bottleneck was demand. AI’s bottleneck is energy. Which means this transition will be larger, faster and more constrained by physics than the last.
Why AI Isn’t Software? Here’s Why That Matters For The AI Bubble Conversation
The most persistent misconception about AI is the belief that it operates like the software of the past half-century. Traditional software was a static artifact. Engineers wrote it, compiled it and executed it endlessly without requiring new thought from the machine. All intelligence was determined in advance. The computer ran instructions.
AI reverses this logic entirely. A model must interpret every request. It must weigh context, generate meaning, and construct an answer that did not exist a moment earlier. Intelligence is not stored; it is produced. There is no prewritten script. There is only potential, activated through computation.
Each response has a physical cost. Every analysis, prediction, and sentence requires GPUs operating continuously, not intermittently. AI behaves less like software and more like a living cognitive industry, a factory that manufactures intelligence in real time using electricity as its raw material.
This is why the infrastructure buildout is so vast. It is not vanity or excess. It is the scaffolding for a global system of real-time cognition.
The demand is not hypothetical. On October 29, 2025, Microsoft reported that Azure’s AI services had once again exceeded supply. The company also disclosed nearly $400 billion in contracted future revenue, with an average commitment of two years. These are not pilots or experiments. They are binding agreements from enterprises that now consider AI compute essential to daily operations. If this were a bubble, we would see unused hardware piled in warehouses. Instead, we face persistent scarcity.
Predictions that AI will soon commoditize overlook the reality of the industrial stack. Nations are spending billions to replicate the capabilities of Nvidia and ASML.
Yet the technological gap is not closing; it is widening. These are not commodity components. They are among the most advanced and intricate systems humanity has ever engineered.
And we are early. We have not seen AI deployed at the scale of a physical workforce. We have not yet entered the era of hybrid quantum–classical discovery. We have not experienced what happens when real-time intelligence becomes a baseline expectation across enterprises.
If AI is a long game, we are still in the early innings.
AI should not be understood as software. It is a new industrial sector built on continuous, energy-intensive cognition. Each computation strengthens the system. Each deployment expands capability. The intelligence curve rises as the cost curve bends downward.
The Trajectory Of An AI Bubble
This is not the trajectory of a bubble. It is the early architecture of a new industrial age.
This is why so many traditional bubble comparisons fail. Previous infrastructure booms were notorious for building far more capacity than anyone could use. Railroads were laid before freight existed. Housing developments sat empty. Telecom companies ran miles of dark fiber with nothing flowing through it. AI is the opposite. The world cannot get enough compute or power.
As Nisa Amoils, founder and managing partner of the venture firm A100X, told me, “Demand for data centers, power, and tokens is insatiable. Unlike the frameworks of previous infrastructure booms — whether railroads, housing, or fiber — that were initially defined by empty train cars, vacant houses, and miles of dark fiber with no traffic, the capacity and frameworks being built for AI are being used. It is prudent to strategically invest in AI-adjacent startups that are safe and usable by enterprises and consumers.”
Her point underscores the reality that this is not speculative capacity waiting for demand. The demand has arrived, and the infrastructure is struggling to keep pace.
The Psychology Behind Every ‘AI Bubble’ Claim
Every apparent bubble, whether political, financial or technological, follows a similar pattern. Systems do not drift into extremes. They react into them. When structures weaken or destabilize, they provoke exaggerated responses as people and institutions search for stability.
Political theorists have long observed that extreme movements rarely appear in isolation. They emerge in response to prior disorder. When people feel unanchored, they seek certainty. When systems fail, they gravitate toward whatever form of stability appears strongest.
Technology evolves the same way. The dot-com boom was a reaction to a new digital frontier. Cryptocurrency surged after the 2008 financial crisis. Today’s enthusiasm around AI is a reaction to something broader: a recognition that our informational and economic systems are struggling to keep pace with reality.
AI is rising in response to disorder. The world feels chaotic, and computational certainty becomes seductive. Markets are not merely pricing future earnings. They are pricing the hope that scaled intelligence will restore balance.
This moment feels inevitable not because it is pure speculation but because it channels a deeper psychological and structural tension. The desire for stability is expressing itself through capital, computation and energy.
The AI Bubble Speculation Loop
Capital moves through the AI ecosystem in a loop that reinforces itself. OpenAI purchases chips from Nvidia to train models. These purchases increase Nvidia’s revenue and valuation. Nvidia invests in startups that rely on its chips. Microsoft funds OpenAI and hosts its workloads. Money circulates, reinforcing the expectation of exponential growth.
In the dot-com era, companies traded ad inventory. Today, companies trade compute. But compute is not symbolic. It is energy consumed at global scale. Training requires megawatts. Every model draws from the same grid that sustains society.
The output is not a traditional product. It is a probability distribution shaped by energy. We are converting electricity into intelligence.
Reflexivity, Energy And The AI Bubble Illusion
In traditional capitalism, capital becomes labor, labor becomes product, product becomes value, and value becomes reinvestment. AI inverts this cycle. Capital becomes narrative, narrative becomes valuation, and valuation attracts more capital.
Markets are not pricing what intelligence does. They are pricing what intelligence might become.
George Soros called this reflexivity. AI turns reflexivity into physics. The speculative unit is no longer attention or real estate. It is compute. More precisely, it is the electrons used to drive it.
Sam Altman has said plainly that this entire transformation is about electrons. Models are structured electricity. Intelligence is becoming a thermodynamic process.
Energy Is the Missing Lens In The AI Bubble Debate
For decades, conversations about technology centered on user interfaces — the points where humans interact with machines. Visionaries such as Peter Diamandis emphasized that progress accelerates when these interfaces become more intuitive. Touchscreens, voice assistants, spatial interfaces and neural links were framed as the gateways to technological adoption.
Yet the deeper truth is that compute itself is the user interface of energy.
The great technological breakthroughs of history were breakthroughs in how humans interface with energy. Fire was the first interface, a controlled release of combustion. Mechanical tools shaped force. Electricity extended human perception and communication across distance. The microchip compressed energy into logic. The internet distributed that logic globally.
Diamandis described the visible layer. The deeper interface is the substrate, the way energy arranges itself into intelligence.
Each shift in how humans harness energy opens a new layer of capability. Electrification enabled radio, which expanded human awareness. Semiconductors enabled computation, which reorganized society. Now large-scale compute capacity is enabling predictive intelligence.
These transformations were never about the surface interface. They were about the underlying substrate of energy. That is why the “AI bubble” framing collapses. AI is the next interface in a lineage that began with fire. When intelligence becomes an expression of energy, its economics resemble a long thermodynamic ascent rather than a short speculative spike.
We are not witnessing the rise of artificial intelligence. We are witnessing the rise of energy-formed intelligence. Nothing about that resembles a bubble. It resembles a shift in the fabric of civilization.
This Isn’t An AI Bubble It’s An Energy Revolution
Energy and intelligence have always been intertwined. Fire was our first algorithm, a controlled release of stored energy that reshaped what humans could do. Electricity extended human capability across continents. Civilization itself is the story of arranging energy into meaning.
This connection is personal. Years ago, I invented the Luci solar lantern to bring light to communities without reliable electricity. It was an effort to democratize photons, to capture sunlight and convert it into opportunity. In Africa, I watched children hold a Luci lantern with awe. Light became cognition. Energy became hope.
Today, AI does not illuminate potential. It simulates cognition at industrial scale. We have built machines that think about thinking, and then charge for the effort. The result is not only financial inflation; it is thermodynamic inflation.
When energy stops producing proportionate value, bubbles form. When those bubbles burst, they strain the systems that supply power, the grid, supply chains, the environment, and public trust.
The Geopolitics Of The AI Bubble Myth
Every discussion about AI eventually becomes a discussion about energy. Every discussion about energy becomes geopolitical. The new arms race is electrical.
China is securing rare earth minerals for GPUs and batteries. The United States is attempting to expand chip manufacturing and strengthen the grid. Europe struggles with energy constraints. The Middle East is becoming a global data-center hub.
When intelligence depends on electricity, sovereignty depends on the grid. The next global contest will be energetic, not ideological. Control the electrons and you control the intelligence. Control the intelligence, and you influence the future.
Why Scaling Intelligence Doesn’t Look Like An AI Bubble
Many assume intelligence grows smoothly as models grow larger. But intelligence has a physical boundary. It is constrained by energy.
As models expand, their marginal returns diminish. Performance slows long before the power curve does. The illusion is that intelligence scales like software. In reality, it scales like biology. Human intelligence is metabolically expensive. Machine intelligence is electrically expensive. The cost curve has not disappeared. It has simply been hidden.
The Coming Correction That Isn’t An AI Bubble Burst
Every major technological shift begins with an irresistible truth. The internet was inevitable. Electrification was inevitable. Intelligence augmentation is inevitable. But inevitability does not guarantee rationality.
The AI bubble, such as it is, will cool. The essential question is what remains. Will we inherit abandoned data centers and stranded GPUs? Or will we inherit a planetary intelligence infrastructure capable of generating lasting value?
The Real Asset Behind The So-Called AI Bubble: Energy Stewardship
If energy is the foundation of intelligence, the most important capability of the next era will be stewardship. Scale for scale’s sake will not endure. The systems that survive will be built by people who understand the cost of their electrons.
The story of intelligence has always been the story of light, from fire to fiber optics to the glow of a Luci lantern. The next digital divide will not concern information. It will concern power. Energy inequality will become intelligence inequality.
So, If the AI Bubble Isn’t Real What Should Business Leaders Do Now?
If this moment is being misread, then so are the opportunities and risks. AI is not a feature to bolt onto existing systems. It is a new layer of infrastructure, and operating in that environment requires a shift in how organizations think and act.
Understand your energy footprint. Intelligence now has a measurable physical cost. Any company that cannot quantify the power demands of its models and data workflows does not yet understand its own intelligence layer.Elevate data quality above model size. The organizations that endure will not be those with the largest GPU clusters, but those whose data is structured, verified, and anchored in reality. Clean inputs will outperform brute-force scale.Build architectures that account for energetic cost. Computation is no longer effectively free. Systems that allocate compute thoughtfully and avoid unnecessary inference will prove more stable and resilient than those that assume infinite power.Invest in true internal literacy. Not prompting tricks or surface-level fluency, but an organizational understanding of how intelligence systems function, where they break, and how energy, data, and risk move through the enterprise.Adopt the timeline of infrastructure, not software. The companies that succeed in the next decade will be the ones built for endurance, alignment, and stewardship. Quarterly cycles cannot guide decisions about a foundational technology wave.
After The AI Bubble: What Jevons And Perez Tell Us
Carlota Perez, the economist whose work on technological revolutions and financial cycles is considered definitive in innovation studies, has shown that every major technological era begins with an investment bubble. The bubble finances the infrastructure, the crash resets the logic, and the infrastructure becomes the foundation of the next era.
William Stanley Jevons, the British economist who first recognized that improvements in energy efficiency can increase total resource consumption, demonstrated that more efficient steam engines intensified rather than reduced coal use. Efficiency expands demand.
The same dynamic applies to intelligence: As AI becomes more efficient at thinking, the world will not use less thinking. It will use more.
History reflects this pattern. The dot-com crash erased speculative capital, but the infrastructure it built became the modern internet. AI will follow the same arc. When speculation cools, the compute base, energy systems, and societal literacy will remain. Intelligence will not be a novelty. It will be an environmental layer.
Beyond The AI Bubble Narrative
If energy is the foundation of intelligence, the next era is not merely about accelerating computation. It is about understanding consequence. The intelligence that matters will be the intelligence that knows its own cost, its effects, and its role in the world.
The bubble is building the scaffolding. The correction will build the purpose. The breakthrough will not be artificial intelligence but responsible intelligence, grounded in physics, aligned with truth, and aware that it exists because human beings directed energy into form.
We may be the first civilization to understand, in real time, that thought requires energy, that energy creates consequence, and that consequence requires care. Progress has never been linear. It evolves through feedback, loops, and return. Energy behaves this way. So do markets. So do minds.
The real test of this era is not whether machines accelerate. It is whether we learn to guide the power behind them. Do that well, and the idea of an AI bubble will fade, replaced by a deeper understanding that intelligence itself is evolving. The future may not be artificial in the narrow sense at all. It may be something that feels remarkably close to life.
