We can't even agree on what number this is.

Klaus Schwab named the current era the Fourth Industrial Revolution in 2016. The European Commission countered with Industry 5.0 in 2021. NVIDIA CEO Jensen Huang calls AI factories "the engine of the next industrial revolution" without committing to a number. A growing group of historians and economists argues the whole "industrial" framing is wrong and that we're witnessing something categorically different: the first cognitive revolution in human history. Meanwhile, some of Carlota Perez's followers say we haven't even finished the third.

The number matters less than understanding why people are fighting over it. That debate is where the interesting thinking lives.

What the Previous Revolutions Actually Had in Common

Before assigning a number, it helps to understand what earns the "industrial revolution" label in the first place. Historians generally apply it when three things converge: a general-purpose technology that applies across many industries, a new energy source that powers it at scale, and a wave of infrastructure build-out that embeds the technology into everyday economic life. The technology alone is not enough. The energy alone is not enough. The combination, at scale, restructuring how the economy works at a fundamental level, is what qualifies.

The first revolution (roughly 1760 to 1840, centered in Britain) ran on coal and steam. The general-purpose technology was the steam engine, which mechanized textile production and then everything else. The infrastructure wave was canals, then railroads. Water power moved goods. Coal moved everything else. The second revolution (1870 to 1914) ran on electricity and petroleum. The general-purpose technology was electrical power and the assembly line. The infrastructure wave was the electrical grid, oil pipelines, and the modern factory. Henry Ford's moving assembly line in 1913 is usually cited as its symbolic peak.

The third revolution (roughly 1960s onward, still contested) was the digital revolution. The general-purpose technology was the semiconductor and later the microprocessor. The infrastructure wave was fiber optic networks, the internet, and enterprise computing. The energy source was cheap electricity feeding server rooms. Some historians fold this into the second revolution; others call it a distinct break. The argument about where to draw lines reflects a genuine ambiguity in how technological transitions compound on one another rather than neatly replacing each other.

The fourth revolution, as Schwab framed it in 2016, is the current fusion of physical, digital, and biological systems: robotics, AI, genomics, autonomous vehicles, 3D printing, all converging simultaneously. Industry 4.0 in manufacturing refers to this specifically: networked machines, digital twins, sensor-driven production, smart factories. It extends the third revolution rather than breaking from it cleanly.

The European Commission's Industry 5.0, published in January 2021, takes a different approach. Rather than marking a technological break, it defines a values-based evolution: manufacturing that puts human workers back at the center, emphasizes sustainability, and pursues resilience over pure efficiency. Under this framing, Industry 5.0 isn't more automated than 4.0, it's more consciously human.

Neither of those definitions is what most people mean when they describe AI as a new industrial revolution. They mean something bigger.

The Best Case for "Fifth": Jensen Huang's Factory Framing

Jensen Huang has been making the same argument for several years, and it has become the clearest industrial-revolution framing in circulation. The core claim: data centers are not storage facilities for the digital era. They are factories for the AI era. Raw materials go in (data and electricity). Finished goods come out (tokens, inferences, decisions, predictions). Every company, regardless of industry, is now in the business of manufacturing intelligence.

"The last industrial revolution was the manufacturing of software," Huang said at Dell Technologies World in 2024. "Previously, it was manufacturing electricity. Now we are manufacturing intelligence."

Every prior revolution had an equivalent: the cotton mill, the steel foundry, the assembly plant, the server farm. AI factories are the current version.

This framing holds up structurally. NVIDIA's Blackwell clusters, Microsoft's hyperscale campuses, the hundred-gigawatt data center buildout underway across the American South and West, these are the mills. They manufacture something that didn't exist as a commodity before: machine cognition at scale.

Data Center Frontier reported in 2025 that U.S. data center power demand could reach 106 GW by 2035, up from roughly 17 GW in 2022. Global data center power consumption, currently around 2 percent of total electricity use, is projected to exceed 6 percent by 2028. These are industrial-scale infrastructure numbers. They have infrastructure-era precedent. The question is which precedent applies.

<figure class="infographic"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1000 360" role="img" aria-label="US data center power demand growth 2022 to 2035"> <defs> <linearGradient id="bargrad" x1="0" y1="0" x2="0" y2="1"> <stop offset="0%" stop-color="#D84824"/> <stop offset="100%" stop-color="#3C90CC"/> </linearGradient> </defs> <text x="20" y="40" font-family="ui-monospace, SFMono-Regular, Menlo, Consolas, monospace" font-size="11" fill="#9aa0b0" letter-spacing="3">U.S. DATA CENTER POWER DEMAND</text> <text x="20" y="68" font-family="Inter, sans-serif" font-size="22" fill="#FFFFFF" font-weight="700">17 GW to 106 GW in 13 years</text> <g transform="translate(60,140)"> <rect x="40" y="100" width="120" height="60" fill="#3C90CC" opacity="0.6"/> <text x="100" y="92" text-anchor="middle" font-family="Inter, sans-serif" font-size="14" fill="#FFFFFF" font-weight="600">17 GW</text> <text x="100" y="180" text-anchor="middle" font-family="ui-monospace, SFMono-Regular, Menlo, Consolas, monospace" font-size="11" fill="#9aa0b0" letter-spacing="2">2022</text> </g> <g transform="translate(60,140)" stroke="#D84824" stroke-width="1.5" fill="none" opacity="0.5" stroke-dasharray="3 4"> <path d="M180 130 Q 380 60 580 30"/> </g> <g transform="translate(60,140)"> <rect x="600" y="-20" width="120" height="180" fill="url(#bargrad)" opacity="0.85"/> <text x="660" y="-32" text-anchor="middle" font-family="Inter, sans-serif" font-size="14" fill="#FFFFFF" font-weight="600">106 GW</text> <text x="660" y="180" text-anchor="middle" font-family="ui-monospace, SFMono-Regular, Menlo, Consolas, monospace" font-size="11" fill="#D84824" letter-spacing="2">2035</text> </g> <text x="500" y="340" text-anchor="middle" font-family="Inter, sans-serif" font-size="12" fill="#9aa0b0">6.2x growth in projected demand. Source: Data Center Frontier, 2025.</text> </svg> </figure>

The Stronger Argument: This Isn't Industrial

Here is where the "fifth industrial revolution" framing starts to show its limits.

Every previous revolution, including the ones called "digital," automated physical or procedural work. Steam engines replaced animal and human muscle in moving materials. Electrical assembly lines replaced hand assembly. Computers replaced manual calculation and record-keeping. Even software automation replaced the procedural steps humans used to execute by hand. In every case, the things being automated were tasks the body or lower-level cognition performed. The human mind remained the essential, unreplaceable input for judgment, strategy, creativity, and decision-making.

AI is different in kind, not just degree.

The cognitive revolution framing, developed by researchers across economics and computer science, rests on a single observation: for the first time in economic history, the thing being automated is thinking itself. Not the procedural thinking of calculation or rule-following, but analysis, synthesis, language, inference, and increasingly, reasoning. Humans built machines to replace muscles. Humans built computers to replace calculation. We have now built systems that replace portions of the cognitive work that defined the human economic advantage.

That is a categorically different transformation. It's the second major structural shift in how humans generate economic value, following the shift from agriculture to industry. The first shift moved value from nature to muscle. This shift moves value from muscle (already automated) to mind, and now it is beginning to automate that too.

Previous industrial revolutions changed what humans did. This one changes what humans are for.

Carlota Perez and the Question of Where We Actually Are

Carlota Perez, the economic historian whose work on technological revolutions has aged better than most, offers a framework that clarifies the timing question.

Her argument, developed across decades of research, is that major technological revolutions follow a two-phase pattern. First comes the installation period: new technology bursts onto the scene, speculative capital floods in, infrastructure gets built well ahead of actual productive use, and a financial frenzy inflates asset prices before a crash. Then comes the deployment period: the technology diffuses broadly, institutions adapt to it, and the productivity gains that the installation period promised finally materialize across the whole economy.

By her framework, AI looks like a textbook installation period. The frenzy markers are all present: hundreds of billions in capital expenditures, valuations decoupled from near-term revenue, infrastructure buildout racing ahead of demonstrated ROI, and widespread speculation about which companies will capture the value. She has argued, in a 2024 piece for Project Syndicate, that AI is best understood as a new development within the still-evolving ICT revolution rather than a clean break into a sixth wave. The microprocessor started a technological revolution in the 1970s that has not yet fully deployed.

What that means practically: the transformation most people are anticipating from AI, the broad productivity gains, the economic restructuring, the labor market shifts, has not actually happened yet. The IMF published a piece in December 2025 noting that the economic historian Nicholas Crafts found steam power's productivity impact in the 19th century was slower and smaller than previously believed, because steam-powered sectors initially made up only a small fraction of the economy. Reaping the full gains from general-purpose technology requires reorganization of production, and that takes decades.

We are, by most credible estimates, somewhere between the peak of the installation frenzy and the beginning of the deployment phase. The factories are being built. The intelligence is not yet broadly deployed in ways that show up in GDP and wage statistics.

The Infrastructure Pattern

Every industrial revolution has been gated by its infrastructure buildout. This is the most consistent pattern across all five (or four, or six, depending on your numbering).

Steam needed coal mines and railroads. You could not run a steam-powered factory without coal delivery and goods transport. The railroad-building frenzy of the 1840s in Britain was infrastructure speculation that preceded the actual productivity gains of the second industrial period. Mass production needed the electrical grid. Electrification of American factories took from the 1880s to the 1920s, and it was only after factories were redesigned around electric motors (not just replacing steam with electric, but redesigning the whole production layout) that the productivity gains showed up. The digital revolution needed fiber and reliable baseload power. The internet as consumer infrastructure did not exist at scale until the fiber overbuild of the late 1990s.

The current build is the grid and the data center. Both are undersized for what is being planned.

Data Center Frontier documented the grid bottleneck in 2025: the breakthrough technology is landing on a grid that was never designed for this kind of load, at this kind of speed. New transmission projects face multi-year interconnection queues. Large-scale data center projects in high-demand markets are stalling not because of capital shortages or technology gaps, but because the megawatts aren't available fast enough.

The technology exists. The capital is willing. The physical infrastructure is the constraint.

This is the exact pattern of previous infrastructure-constrained revolutions.

In Arizona, this plays out directly: the state's emergence as a top-three data center market is driven as much by available land and growing power infrastructure as by any technology factor. The factories of this revolution are being sited by the same variables that sited the factories of previous ones: energy access, land cost, regulatory environment, and labor availability.

The Speed Problem

The most genuinely unprecedented feature of this revolution is not the technology. It is the pace of diffusion.

ChatGPT reached 100 million users in two months after its November 2022 launch. No previous technology in history came close to that adoption curve. The internet took seven years to reach the same mark. The telephone took 75 years. The television took 13. Previous industrial revolutions unfolded over decades, giving societies time for generational adaptation.

<figure class="infographic"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1000 380" role="img" aria-label="Time to 100 million users by technology"> <text x="20" y="40" font-family="ui-monospace, SFMono-Regular, Menlo, Consolas, monospace" font-size="11" fill="#9aa0b0" letter-spacing="3">TIME TO 100 MILLION USERS</text> <text x="20" y="68" font-family="Inter, sans-serif" font-size="22" fill="#FFFFFF" font-weight="700">A logarithmic compression of adoption</text> <g transform="translate(220,120)" font-family="Inter, sans-serif"> <g> <text x="-180" y="22" font-size="14" fill="#FFFFFF" font-weight="600">ChatGPT</text> <rect x="0" y="6" width="14" height="22" fill="#D84824"/> <text x="22" y="22" font-size="13" fill="#D84824" font-weight="700">2 months</text> </g> <g transform="translate(0,46)"> <text x="-180" y="22" font-size="14" fill="#FFFFFF" font-weight="600">Internet</text> <rect x="0" y="6" width="84" height="22" fill="#3C90CC" opacity="0.85"/> <text x="92" y="22" font-size="13" fill="#FFFFFF" font-weight="500">7 years</text> </g> <g transform="translate(0,92)"> <text x="-180" y="22" font-size="14" fill="#FFFFFF" font-weight="600">Television</text> <rect x="0" y="6" width="156" height="22" fill="#3C90CC" opacity="0.65"/> <text x="164" y="22" font-size="13" fill="#FFFFFF" font-weight="500">13 years</text> </g> <g transform="translate(0,138)"> <text x="-180" y="22" font-size="14" fill="#FFFFFF" font-weight="600">Telephone</text> <rect x="0" y="6" width="660" height="22" fill="#3C90CC" opacity="0.4"/> <text x="668" y="22" font-size="13" fill="#FFFFFF" font-weight="500">75 years</text> </g> </g> <text x="500" y="350" text-anchor="middle" font-family="Inter, sans-serif" font-size="12" fill="#9aa0b0">Bar length scaled to months. ChatGPT compressed adoption from decades to weeks.</text> </svg> </figure>

Children who grew up after the steam era had different skills than their parents. Children who grew up with electricity took it as given. The labor market adapted through generational replacement as much as through retraining.

AI is not giving society that generational buffer. The people whose cognitive tasks are being automated are the same people alive right now. The handloom weavers of the first industrial revolution suffered badly in the transition, with wages in some Lancashire towns falling by more than half over five years in the early 1800s. But those weavers were one occupation among many, and the broader economy was still mostly agricultural. Today's cognitive workers represent a majority of employment in high-income economies.

The reinstatement effect, the historical tendency for new technologies to eventually create more jobs than they displace, has held across every previous revolution. Roughly 60 percent of workers in 2018 held occupations that did not exist in 1940. The net count has always been positive. But reinstatement is a long-run phenomenon. The question this revolution poses is whether the pace of displacement and the pace of reinstatement will be close enough together to avoid sustained disruption at scale.

That is genuinely unknown. Anyone claiming certainty in either direction is wrong.

What to Watch For

If Carlota Perez's framework holds, the installation phase ends with a financial correction before the deployment phase begins. The deployment phase is when the technology's economic benefits actually materialize broadly. That is when productivity statistics move, wages rise in technology-adjacent occupations, and new industries emerge that weren't imaginable during the installation period.

The signals that the deployment phase is beginning: AI stops being a capital expenditure conversation and becomes an operating model conversation. Companies restructure operations around AI capabilities rather than buying AI as an add-on. Labor markets start showing wage premiums for roles that integrate AI judgment with human expertise. Productivity statistics begin moving faster than the historical baseline.

None of those signals are dominant yet. The infrastructure is being built. The productivity is being promised. The deployment is ahead of us.

Whether this deserves to be called the fifth industrial revolution, the first technological revolution, or the first cognitive revolution in human history, depends on which framing you find most useful. The "industrial" label grounds it in historical precedent and makes the infrastructure parallels legible. The "cognitive" label emphasizes what is genuinely new about automating mind rather than muscle. The "technological revolution" framing, Perez's framing, places it correctly in time as still-in-progress rather than already-accomplished.

All three framings are partially right. What none of them should do is obscure the core fact: the thing being built right now, the data centers, the power infrastructure, the AI factories, is the physical substrate of whatever this turns out to be. Previous revolutions were eventually judged by whether their infrastructure managed to keep pace with their transformative potential.

This one will be too.

Sources: AI Business / NVIDIA / Data Center Frontier / IMF Finance and Development / Carlota Perez, Project Syndicate / Columbia Business School / JHU Hub