Field Note
This Time It's Different. Until It's Not.
AI may change the cost structure of cognition. That does not mean it has escaped the older financial pattern of overinvestment, leverage, and institutional hardening.
AI optimists are right about something important: this time really is different. The reason is deeper than the current run of productivity demos, enterprise pilots, agent frameworks, and increasingly elaborate charts about token costs. For the first time, cognition itself has become cheap enough to distribute at scale. A person who could not write production software can now build useful software. A founder can use a model to get working traction on Riemannian geometry, not as decorative intellectual furniture, but as scaffolding for a real project. A lawyer can reason through technical architecture. A scientist can draft code. A coder can interrogate finance. A student can sit beside a machine that explains, translates, simulates, summarizes, and generates on demand.
That does not mean expertise has vanished. It has not. A model can explain a field without living inside its standards, and it can produce plausible work without knowing which parts are brittle. Anyone who has used these systems for real work learns this quickly, usually by stepping on something sharp. But the threshold has moved. Deep access that once required a specialist, a department, or years of apprenticeship can now be reached in seconds. First contact with difficult knowledge is cheaper. Iteration is cheaper. Translation across domains is cheaper. That is not another SaaS feature. That is a change in the cost structure of thought.
So when people say AI will change the world, the correct response is not a smug lecture about tulips. AI may change the world. It may matter more than the internet because the internet changed access to information, while AI changes access to cognition. That is a larger claim, and it deserves to be treated seriously.
But real transformation has never protected a market from overinvestment. If anything, real transformation is what makes overinvestment attractive.
Minsky's Warning
This is where Hyman Minsky becomes useful. Minsky was an American economist whose central insight was not that people are fools, or that markets are always irrational, or that capitalism is one large machine for manufacturing regret, although one can understand the temptation. His deeper claim was that stability itself changes behavior. When a financing pattern works for long enough, participants gradually take on more risk because the world has been rewarding them for believing the structure is safe. The system becomes fragile not only through failure, but through success.
Minsky's Financial Instability Hypothesis is worth understanding because it gives us a better language for technological booms than the usual bull-versus-bear shouting match. In his framework, financial systems tend to migrate through three broad forms of financing. In hedge finance, borrowers can cover both principal and interest from expected cash flows. This is the boring world, which is often where civilization hides its best work. In speculative finance, borrowers can cover interest but need to refinance principal. This can still function if markets remain open and confidence holds. In Ponzi finance, borrowers cannot cover principal or interest from cash flows and must rely on rising asset values or continued refinancing. At that point, the future has to arrive on schedule.
That last sentence is the hinge. Every real technological boom creates a plausible future. Railroads made a new transport economy imaginable. Electricity made a new industrial economy imaginable. Telecom made a new communications economy imaginable. The internet made a new distribution economy imaginable. AI is making a new cognitive economy imaginable. The danger does not come from imagining the future. The danger comes when financing begins to treat the future as if its timing, margins, demand, infrastructure, and adoption curve have already been settled.
The Financing Form
This is why "this time is different" is usually half right. The technology can be genuinely different. The mistake is assuming that technological difference cancels financial pattern. Real innovation expands the believable future, and finance then converts that belief into exposure. It does so through whatever machinery the period makes available: shares, scrip, bonds, trusts, vendor financing, structured products, venture rounds, private credit, power contracts, data-center leases, or some new wrapper with better typography and worse covenants.
The historical record is not subtle. The British Railway Mania was not a mass hallucination about imaginary trains. Railways were real. They changed Britain's economic geography. Gareth Campbell's work is useful here because it keeps us away from the lazy morality play. Investors were not simply buying "railway optimism." They were buying particular forms of exposure: scrip, partly paid shares, future capital calls, and claims on profits that required the buildout to unfold in a particular way. The technology was real, but the financing structure still mattered.
The same pattern appears in different clothes across later cycles. The telecom boom financed a real communications revolution, but vendor financing and debt-funded network buildouts assumed a demand curve that did not arrive quickly enough. The internet boom funded companies built around a real transformation in distribution, but public equity markets turned many plausible futures into tradable claims before the underlying businesses were ready. The 2008 crisis involved housing, which was real enough, but mortgage-backed securities, CDOs, CDS, repo financing, ratings, guarantees, and off-balance-sheet vehicles transformed local mortgage exposure into a globally distributed structure that looked safer as it moved.
Different tools. Similar function. Financial instruments broaden participation, move risk, make the story easier to own, and delay the moment when the system has to decide whether the story has outrun the structure beneath it. This is not always bad. Financial innovation can fund infrastructure, distribute risk, and bring capital to useful work. A society that cannot finance new futures is not prudent. It is stuck. But financial innovation can also act as a pressure valve, allowing a system to keep expanding when direct adjustment would be painful.
The AI Cycle
That is the useful way to look at the current AI cycle. The serious question is not whether AI is real. It is real. The serious question is what kind of financial structure is forming around the belief that AI is real. Who is financing the buildout? Who is absorbing duration risk? Who is underwriting future utilization? Who is assuming demand arrives on time? Who owns the downside if model architectures shift, inference economics change, or customers adopt more slowly than projected? Who needs the story to remain intact for one more funding cycle?
Minsky helps because he makes this question precise without requiring cynicism. A system can move toward fragility through decisions that look reasonable one at a time. If utilization is rising, capacity expansion looks prudent. If capital is available, refinancing risk looks manageable. If valuations keep rising, collateral looks stronger. If every serious competitor is spending, restraint starts to look like strategic negligence. Stability teaches the system the wrong lesson: that the structure is safer than it is.
At some point, the belief stops being merely a belief. It becomes structure. Contracts. Debt. Data centers. Power agreements. Semiconductor commitments. Hiring plans. Investor letters. Board decks. Valuation models. Public narratives. Once enough structure accumulates around a belief, changing the belief is no longer a private act of cognition. It becomes a public act of reversal, and institutions are bad at those. They prefer revisions that can be described as continuities. They prefer pivots that still sound like the original strategy if read quickly by someone tired.
This is also why financial markets cannot be understood as one giant mind processing information. Markets are made of different professional populations with different salience filters. Equity investors see growth. Credit investors see repayment. Venture investors see optionality. Infrastructure investors see duration. Executives see strategic necessity. Regulators see systemic boundaries. Journalists see the story that will travel. The same information passes through different identities and produces different conclusions. That is not an error in the machine. That is the machine.
When a new technology opens a plausible future, these groups do not move together. Some move first because they can. Others wait because they must. Some need proof. Some need permission. Some cannot participate until an instrument translates the exposure into something they are allowed to hold. Then, very often, the system creates that instrument. The next pool of belief becomes reachable. The story travels farther. The structure hardens.
This is how real transformations can end in financial breaks without making the transformation false. The technology opens the future. Capital races to own it. Instruments broaden the race. Institutions harden around the story. Then the future arrives unevenly, late, or in a form different from what the financing structure required. The crash, if it comes, does not prove the technology was fake. It proves that real transformations are among the easiest things in the world to overfinance.
Different at Which Layer?
So yes, this time is different. AI changes the cost structure of cognition, and that may be one of the most important economic facts of this century. Minsky would not require us to deny that. He would ask a more disciplined question: what forms of financing are being built around that difference, and how much of the system now depends on the future arriving exactly on time?
That is where "this time is different" becomes useful again. Not as a slogan. As a warning to specify the layer.
Technology may be different. The financing pattern may not be.