Nanotechnology was the AI of the 2000s. What can we learn?
How Nanotechnology Survived The Failure of Its Metaphor
1. Sweet Dreams Were Made of This
In the 2000s, nanotechnology was spoken about the way AI is spoken about today: as an inevitability, a destiny, a force so transformative that even the metaphors were preloaded with awe.
When President Bill Clinton announced the National Nanotechnology Initiative (NNI) in January 2000, he promised a near-future of devices that would “translate foreign languages as fast as you can talk,” and—most memorably—“molecular computers the size of a teardrop with the power of today’s fastest supercomputers.” The rhetoric was backed by money and bolstered by promises of unparalleled riches. In 2002, NIST projected that the nanotech market would be “exceeding $1 trillion by 2015”. Sounds familiar, doesn’t it?
The promise of nanoscale went beyond the natural impulse to miniaturize technology. Nanoscale is small, incredibly so — a nanometer is a billionth of a meter, the length your hair grows by the time you comb it once. But at the nanoscale, the smallness has a quality.
Shrink a particle, and its surface area grows relative to its volume. A block of wood has six faces, but crush it into fine powder, and you’ve exposed thousands of times more surface to react with the fire. At the nanoscale, a material is almost all surface. This changes the rules. Gold, inert and noble in bulk, becomes a ferocious catalyst as a nanoparticle. This is what made the trillion-dollar predictions feel plausible: not just miniaturization, but a genuine change in what materials can do, an unlocking of capabilities and possibilities.
Yet a quarter-century after nanotechnology’s canonization as a national project, the word itself seems to have slipped out of ordinary conversation. However, this essay is not a claim that nanotechnology “failed,” nor is it a prediction that AI will. Rather, it is an attempt to name a particular kind of failure, a failure of metaphor, using nanotechnology as the prototype.
How a technology is spoken about is its metaphor. The metaphor represents the idealized outcome of a technology: “putting a man on the moon,” or “energy too cheap to meter,” or “curing cancer.” It is the storefront. The hype. Behind the metaphor is the real enterprise doing the real work. Can the enterprise survive if the storefront collapses?1
Nanotechnology is the prototype here because it can look, from a distance, like a punchline (“remember when nanotech was the next big thing?”) and still be objectively everywhere in the substrate of modern life. The disappearance of nanotechnology is a linguistic fact and not a technical one.
While the Google Trends searches for “nanotechnology” steadily trended downward in the last two decades, the technologies that nanotech enabled, like OLEDs (on most screens now), Lipid Nanoparticles (used in mRNA vaccines), and CRISPR (programmable molecular machinery for gene therapy), are all trending upward. Perhaps the most pervasive success of nanotech today is silicon chips, the ever more powerful CPUs and GPUs: the same technology that is the material substrate of AI. Not bad for a field that even Elon Musk declared is “100% synonymous with bs.”
In the case of nanotech, the science did not die. However, the public-facing picture of what the science would be turned out to be wrong, and the consequences of that wrongness were mostly absorbed rather than catastrophic. What exactly did the soothsayers get wrong? And why didn’t it matter?
2. What’s in a Metaphor?
It is a truth universally acknowledged that anyone writing about Nanotechnology must acknowledge Feynman’s 1959 talk There’s Plenty of Room at the Bottom. Feynman’s impromptu (and characteristically brilliant) post-dinner talk at the annual American Physical Society meeting at Caltech has been retroactively retconned as the birth of the field of Nanotechnology (the talk itself never uses the word, which had not yet been coined).
Feynman begins with a thought experiment about engraving the entire Encyclopedia Britannica on the head of a pin. And by the end of his lecture, we have been introduced to the concepts of manipulating individual atoms, building minuscule machines, and even a miniature surgeon that you can ingest to cure your ailments. Here was a metaphor everyone could grasp: a scaled-down version of our own world, if we could only figure out the engineering challenges.
Feynman’s talk is an intellectual romp. But its influence was almost entirely retrospective. There was very little discussion of this talk until the 1980s, when Feynman’s vision was resurrected and massively expanded by K. Eric Drexler in his 1986 book Engines of Creation: The Coming Era of Nanotechnology.
Drexler popularized the term “Nanotechnology” (which had been coined by now2). In his book, Drexler watered the long-dormant seed that Feynman planted. He took Feynman’s playful thought experiments and codified them into a full-blown mechanical paradigm. More importantly, Drexler ignited the public and scientific attention by adding the core hype-cycle concept of self-replicating assemblers to Feynman’s vision.
But the decisive act of this vision’s canonization was political, not scientific. President Clinton’s January 2000 speech announcing the National Nanotechnology Initiative (NNI) at Caltech directly quoted Feynman’s 1959 question—“What would happen if we could arrange the atoms one by one the way we want them?”—as the justification for the $500 million initial investment in the program.
This was Drexler’s “Dry” metaphor of miniature machines, nanobots, and assemblers. This metaphor is based on a simple, mechanically intuitive premise: “build machines on the molecular scale” atom-by-atom. The most radical implication of this top-down, atom-by-atom vision is that you could print an entire human being if you had a 3D printer with one-atom resolution.
In 1989, following this vision, Don Eigler and Erhard Schweizer at IBM’s Almaden Research Center used a scanning tunneling microscope (STM) to position 35 individual xenon atoms on a nickel surface, spelling out “IBM.” It became one of the most iconic images in the history of science — and arguably the most important piece of nanotech propaganda ever produced. Behind the scenes, this required 22 hours of work by human operators using the STM at 4 degrees Kelvin. Still, the promised implications had a real physical hook, and every technology boom needs an image that fits inside a human skull.
The assembler metaphor was powerful because it promised mastery: a machine that could build anything, precisely, cleanly, on command. This was the “dry” metaphor in the user’s sense: dry as in mechanical, top-down, visibly engineered, and dry also as in separated from the messy chemistry of liquids, ions, proteins, and membranes. And the IBM logo, made atom-by-xenon-atom, was the dry metaphor brought to life.
But nanotechnology’s eventual reality would prove different. The cleanest artifact of that mismatch is the Drexler–Smalley debates, which became one of the most visible boundary fights about what nanotechnology was and was not.
3. What Happens at the Nanoscale
Richard Smalley, a Nobel laureate in Chemistry, took the mechanical imagery seriously and then showed why it collapses at the nanoscale. Smalley’s critique attacks the “Dry” picture on its own terms. He explained why the laws of physics are non-cooperative at the nanoscale for such a vision. He drew our attention to the pesky aspects of reality, such as viscosity, Brownian motion, and Van der Waals forces, that render gears and levers useless at the nanoscale.
Smalley’s two arguments were later formalized as “fat fingers” and “sticky fingers.” The “fat fingers” argument begins with an almost childlike point that becomes fatal in miniature: manipulator “fingers” are made of atoms, so they have “a certain irreducible size,” and “there just isn’t enough room in the nanometer-size reaction region” to do the kind of universal, atom-by-atom mechanical control implied by the assembler metaphor. The “sticky fingers” problem follows: even if you could fit the manipulators, “the atoms of the manipulator hands will adhere to the atom that is being moved.” This is why writing “IBM” was such an arduous process. Imagine manipulating not 35 but billions of atoms this way.
Some of the most worthwhile parts of my otherwise misspent youth were spent coming to terms with Smalley’s critiques of Feynman’s and Drexler’s vision. In the early 2010s, Nanotechnology programs were cropping up in India as the hot new thing. Hot, new, and with questionable career prospects. But the lure of calling myself Feynman’s intellectual descendant proved irresistible, and I chose to major in Nanotechnology. I started on Drexler’s side, and slowly, begrudgingly, had to admit that Smalley had a point. And empirically, the parallel “Wet” vision of nanotech inspired by Smalley’s critiques seems to have won out.3
Drexler’s vision was built on a mechanical engineer’s intuition: that the rules governing a wristwatch also govern a molecule (more or less), just at a different scale. But the nanoscale is the threshold where classical behavior gives way to quantum effects. Electrons start behaving like probability clouds. And their behavior becomes sensitive to small changes in scale.
A nanoparticle of cadmium selenide (CdSe) doesn’t have a fixed color — its color depends on its size, because quantum confinement forces the electrons into discrete energy states that shift with the particle’s diameter. Shrink the particle, and it glows blue; grow it, and it glows red. No gear or lever does that. Top-down manipulation doesn’t work here. Bottom-up chemistry is how things get done in this world. That’s how you manipulate the size of these ‘quantum dots.’
Drexler-Smalley debates did not only concern Drexler and Smalley but also the plausibility of the entire dry metaphor of atom-by-atom mechanical manufacture. The argument forced a distinction: nanoscale engineering is real, but the mechanical picture of nanoscale engineering might be a misdescription. The dry metaphor failed because the nanoscale isn’t merely our world in miniature. It’s a different world, with different rules.
Quantum confinement also provides a way to see why “fat fingers” and “sticky fingers” were not mere rhetorical jabs. If nanoscale behavior depends on confinement and surface states, then nanoscale engineering does not involve placing atoms like Lego bricks, but controlling growth conditions, ligands, surfaces, and environments so that the system settles into a desired nanoscale structure. That is, the path to control runs through chemistry.
The nanotech reality today consists of bottom-up assembly, colloids, lipids, and nanocrystals, and not roaming swarms of nanobots assembling objects on demand atom-by-atom. As Smalley insisted, nature’s and chemistry’s preferred manufacturing tricks are not mechanical grasping but environment-sensitive self-organization. Working with this reality means we now have OLED screens (from quantum-dot research) and mRNA vaccines (from Lipid Nanoparticles research). The image that sold nanotech to the public (top-down atom manipulation) was categorically different from the science that delivered value (bottom-up chemistry).
The lesson here is that a metaphor can be wrong in form and right in direction. And the Drexler–Smalley debates forced a separation between spectacle and science. This separability enabled nanotech to succeed while its original metaphor failed.
4. The Quiet Revolution of Nanotech
If the story ended with the hype fading, nanotechnology would be a cautionary tale and nothing else. Instead, in a twist of technological fate, the failure of nanotech’s metaphor proved liberating because it was separable from the productive science. And nanotechnology became so successful that it became hard to see. The laboratory reality could diverge from the public spectacle without requiring consumer products to keep selling the spectacle. Additionally, when the ‘nano’ label becomes awkward, you can often stop saying it while continuing to ship the underlying capability. This was the quiet revolution.
This quiet revolution can be told through three wins: quantum dots in displays, lipid nanoparticles in vaccines, and semiconductor scaling in computation. Each win is “wet” in the sense that it depends on chemistry, materials, and nanoscale physics rather than miniature machine tools. And each win also shows how nanotechnology’s successes could be rebranded into product categories that do not advertise their nanoscale origins.
Quantum dots, first, are a perfect emblem of the post-metaphor era because their core behavior is not mechanical at all: it is quantum. They are typically 1-10 nanometers in size, a size range where quantum effects allow color to be tuned by size rather than by pigment chemistry. In 2023, Bawendi, Brus, and Ekimov received the Nobel Prize in Chemistry “for the discovery and synthesis of quantum dots,” an explicit recognition that controlling size, growth, and surface chemistry had become a reproducible technology rather than a laboratory curiosity.
Quantum dots then became commercial without remaining culturally “nano.” QLED TVs use “a quantum-dot layer” to produce “more accurate and vibrant colors,” an engineering description that omits the word “nano” even though the dots are nanocrystals. Sony markets similar QD-enhanced products as “Triluminos,” another illustration that nanotech’s wins were absorbed into brand identities that do not require the nanotech label to function.
Second, lipid nanoparticles: the delivery technology that makes mRNA vaccines possible and that almost no one has heard of. As this review from 2016 (i.e, pre-COVID) states, lipid nanoparticles (LNPs) are “among the most frequently used vectors for in vivo RNA delivery,” an unassuming phrase that situates the mRNA vaccine platform within a longer trajectory of nano-enabled delivery science. The unprecedented speed at which the mRNA COVID-19 vaccine was developed would have been impossible without a mature, modular delivery technology that could be adapted quickly.
And again, the metaphorical contrast is explicit in public communication. A Reuters fact check states that the Pfizer-BioNTech and Moderna vaccines “do not use nanorobotic technology,” which is a precisely worded public rebuttal of the dry, nanobot imaginary. The vaccine story is nano, but it is nano as chemistry and delivery, not nano as tiny robots patrolling bloodstreams. In another review on mRNA vaccines, this time from late 2021: “Lipid nanoparticles are going into billions of arms,” a line which is both a literal statement of adoption and a metaphor for how a mass public intervention based on nanotechnology can quietly diffuse without being popularly recognized as “nanotechnology.”4
Third, semiconductor scaling is the most ironic win, because it is the nanoscale success story that now powers the AI boom that has replaced “nanotechnology” as the default future-word. The “wet” part of this story is that at these nanometer scales, matter behaves with the same physics that makes wet chemistry powerful: surfaces dominate, and quantum tunneling emerges. And fabrication becomes an ultra-synchronized dance of materials rather than a simple scaling of mechanical patterning.
These examples show why nanotechnology could fade as a label and thrive as a substrate. The technologies that mattered most were never dependent on the dry metaphor for their technical success, and they did not need the nanotech label for their commercial success.
Another advantage of nanotechnology here was the tolerance engineered into the funding structure. That tolerance mattered because hype has costs. The highest estimate of cumulative NNI funding totals a little over $45 billion since 2001. The annual budget request for FY 2025 was just over $2.2 billion. The funding is at a scale large enough to matter yet small enough to be carried as a long-term public program. Nanotech wasn’t a venture-scale wager demanding a short-term multiple.
It is worth noting, though, that the commercialization of nano’s biggest successes — lipid nanoparticles, semiconductor lithography, display technologies — was not purely a government affair. Private capital entered aggressively once the products had separated from the nanobot metaphor. And a pandemic ‘helped’ mature mRNA vaccine technology. The government money bought time for the metaphor to fade without killing the science beneath it. Then venture capital made the most of the mature science.
Eventually, nanotechnology stopped being a story about tiny machines and became a set of techniques for exploiting surface area, self-assembly, and quantum confinement to build materials and devices that work better, often without announcing why.
5. Do We Really Have to Talk About AI?
Yes. The temptation, in the age of AI, is to treat nanotechnology as a quaint prequel: an earlier wave of hype that fizzled because it never delivered its central promise. The evidence supports a more nuanced and more useful reading: nanotechnology did deliver, but it delivered through mechanisms that were not the mechanisms the public was taught to imagine, and it delivered in forms that could comfortably shed the label “nanotechnology.” This story offers three fundamental lessons in Separability, Rebrandability, and Capital Patience.
The promise of the dry metaphor was directionally right but descriptively wrong about the form it would take. Nanobots and assemblers were the stuff of nano dreams. Reality turned out to be made of size-tunable band gaps, self-assembled lipid structures, and lithographic control constrained by quantum mechanics rather than tiny nanobotic claws placing atoms one by one. This separability of the top-down picture from the bottom-up actualization is the first lesson. It allowed the laboratory reality to diverge from the public spectacle without sabotaging commercialization.
The second lesson is rebrandability. Thanks to the science and spectacle being separable, mRNA vaccines, QLED displays, and semiconductor lithography are marketed under domain-specific labels rather than “nanotechnology.” Companies that once wore “nano” on their sleeve now often present themselves as ordinary materials, semiconductor, or biopharma firms.5
This transition was anchored by the NNI’s capital patience. ~$45B in cumulative federal funding over a quarter of a century created a distributed portfolio that could tolerate long translation timelines. These patient billions allowed incremental advances to accumulate in the background until they were pervasive. Private capital entered aggressively once the underlying science had successfully decoupled from the dry nanobot metaphor, allowing the wins to be absorbed into the market without the baggage of the original brand.
How does AI fare on the metrics of Separability, Rebrandability, and Capital Patience? From my industry-outsider, AI-user, Substack-reader vantage point, AI appears to sit in a tighter vise made of metaphor, brand, and capital.
AI’s metaphor is of a ‘digital polymath’ who reasons, plans, and has intent. Unlike nanotech’s dry metaphor, AI’s anthropomorphic metaphor wasn’t retroactively attached — it was the founding act itself. The Agentic/Humanoid metaphor we are being sold now was conceived in the 1956 Dartmouth conference proposal, written by McCarthy, Minsky, Rochester, and Shannon, i.e., by the founding fathers of AI themselves. In fact, it was metaphors all the way down: mind-as-computer, thinking-as-computation, intelligence-as-symbol-manipulation. The very name “artificial intelligence” was McCarthy’s coinage for the conference, and it encoded the humanoid metaphor directly into the field’s identity.
If ChatGPT is the “IBM in atoms” moment for AI (i.e., real demonstration of real technology that happens to perfectly validate the wrong—or at least incomplete—picture of what the field actually does at scale), then the thing that makes the metaphor feel real (conversational fluency) is the same thing being sold as a product. For AI, ChatGPT is both the iconic demo and the productive technology.
The passing of the Turing Test — that long-anticipated “watershed” moment — by LLMs came, and this spectacular success did not vindicate the metaphor. Because the Turing Test was but a behavioral benchmark (can you fool someone?), and LLMs satisfied it through statistical pattern completion without anything resembling understanding. The industry doubled down. The thinking went: turns out conversational fluency doesn’t require intelligence, but real intelligence means reasoning and planning. The same thing had already happened with chess, long considered the pinnacle of intelligence. Then Deep Blue beat Kasparov in 1997, and today no human can beat a modern chess program. But we don’t think of these programs as “intelligent.”
Each time the goalpost moves, each time the metaphor survives by retreating to higher ground. This chase will continue to the ultimate promised land of Artificial General Intelligence. In other words, Artificial Intelligence is the metaphor. When the technology meets each benchmark through allegedly non-intelligent means (statistical correlation, gradient descent, stochastic parroting – take your pick), the metaphor doesn’t die. Instead, there’s a double-down effect. The field has never had an identity separate from the claim that machines can think the way humans do.
This does not bode well for separability and rebrandability of AI should the need arise.
How about capital patience? The capital situation of AI is also qualitatively and quantitatively different. The money betting on AI is mostly private capital, and it is at least an order of magnitude higher than nanotech’s cumulative government investment. However, the private capital here is more heterogeneous—the biggest bettors are established tech giants—and hence more resilient. If (not when — I’m not a naysayer) the AGI metaphor of the AI industry collapses, if the public stops believing in the grand narrative, then a violent repricing event will be inevitable, and it will be interesting to see what the day after will look like. Nanotechnology doesn’t offer AI the comfort that hype always ends harmlessly, but it does point to a more precise possibility that a metaphor can fail while the technology succeeds.
The analogy between nanotech and AI is tempting because both fields produce an intoxicating collision of visionary narrative and genuine, technically grounded progress. AI is accelerating human progress much like nanotechnology. AI is already providing value in the form of pervasive, incremental productivity: tickets resolved, data cleaned, drafts written, ideas connected, hypotheses tested, theorems proven — to name just a few. The question is whether the value it produces can satisfy the expectations its metaphor created. Nanotech never had to answer that question. AI does.
Thanks to Smrithi Sunil and Grant Mulligan for thoughtful comments on an earlier draft of this essay. Thanks also to Mike Riggs, Steven Adler, and Alexander Kustov for telling me, separately, that the story should primarily be about nanotech, not AI: proof that great minds do think alike. Their advice made this piece so much better. Finally, thanks to Alexander Kustov (again) and Andrew Burleson for the feedback on the near-final version.
And special thanks to Abby ShalekBriski for patiently listening to my too-long-for-a-polite-conversation rant at the Progress Conference about the misleading promises of nanotechnology, and for suggesting that maybe I should write about it instead of accosting people at the Progress Conference.
Sometimes the enterprise eventually collapses even if the storefront advertisement turns out to be honest: we did put men on the moon, and then haven’t been back for five decades.
by the Japanese scientist Norio Taniguchi of Tokyo University of Science in 1974.
This does not rule out Drexler’s vision, just that it hasn’t come true yet. Here’s an interesting write-up that explains this in more detail. I still foster Drexlerian dreams.
The only people who rightly identify mRNA vaccines as nanotechnology are the fear-mongering anti-vaccers who use ‘nanobots’ and ‘nanotech’ interchangeably while talking about covid vaccines.
This also happened in academia. Because the “nano” label meant higher chances of getting research funding in the wake of NNI announcement, some fields underwent rapid rebranding: microfluidics became nanofluidics, microelectromechanical systems (MEMS) became NEMS, researchers started talking about nanomedicine, nanobiology, nanoelectronics and so on in their funding applications. Nanotech programs started cropping up. When I joined my PhD program in 2018, it was called the Nanoscale Science PhD program ever since its inception in 2009. When I graduated in 2024, the program had become Chemistry & Nanoscale Science PhD program. The nanotechnology program of my undergrad has since been discontinued, and the Department of Nanotechnology (which was established in 2009) where I got my undergrad has now become Department of Metallurgical and Materials Engineering.
I think this is a very charitable view of current state of AI. Unfortunately, it has a stronger vibe of power loom invention and the whole luddites/enclosures/slavery/child labor economical change attached to it.
On the other side: Ben Feringa did (sort of) achieved wet nanotechnology at the end, but everybody's attentions was elsewhere by then
Code is very different from nano. It adheres to no scientific or engineering constraints. Code is essentially UFOlogy on steroids. As both words and code "create" the illusions hand in hand ("it's 'thinkng'") it mimics agency, thought, behavior, etc all from actions unconstrained by air, gravity, flow. Code's deception is rather colossal because anyone can and will be fooled by it, lacking an underlying knowledge base of what symbols are and what metaphors are.