Kurzweil Scorecard: The Chemistry Was Real. The Nanobots Weren’t.

Buried in the endnotes of The Singularity Is Near are two citations that most readers skip. One is a 2000 paper on shining light through zeolite crystals to oxidize toluene. The other is a turn-of-the-century roadmap for catalysis science, paired with advances in a mesoporous silica called MCM-41. Dry stuff. But Kurzweil didn’t cite them idly. They were ammunition in the most important scientific fight of the book — his rebuttal to the chemist who told him molecular assemblers were physically impossible.

He won the citation. He lost the mechanism. And the gap between those two outcomes is the most honest thing this batch of predictions can teach us about forecasting.

The predictions

Both items come from (The Singularity Is Near, ch. “Response to Critics”), the chapter where Kurzweil answers the people who said his timelines were fantasy. The headline antagonist was Richard Smalley, the Nobel laureate who coined the “fat fingers / sticky fingers” objection: you cannot build a molecular assembler that places individual atoms, Smalley argued, because the manipulator arm would itself be made of atoms too fat and too sticky to do the job.

Kurzweil’s counter was that precise, designed chemistry at the molecular scale was already routine. To prove it, he reached for catalysis. His footnotes assert, first, that the photooxidation of toluene and p-xylene in cation-exchanged zeolites had been demonstrated by 2000, and second, that a roadmap for catalysis science in the twenty-first century, along with advances in the mesoporous molecular sieve MCM-41, had been articulated by 2001 and earlier. The implied syllogism: if chemists can already steer reactions atom by atom inside an engineered crystal cage, Smalley’s impossibility proof collapses, and the road to molecular manufacturing is open.

In The Singularity Is Nearer (2024), Kurzweil keeps the destination intact. He writes that “a molecular assembler might be a tabletop-size unit capable of manufacturing virtually any physical product for which it has the requisite atoms,” with the incremental cost of any object falling to “only pennies per pound … essentially just the cost of the atomic precursor materials.” The 2024 book pushes the assembler era into the late 2030s and early 2040s. The vision didn’t shrink. It slid.

Where we actually are

Start with the easy part: the footnotes are accurate.

The Panov paper exists exactly as cited — “Photooxidation of Toluene and p-Xylene in Cation-Exchanged Zeolites X, Y, ZSM-5, and Beta,” published in 2000, now sitting at 40 citations. It is real, modest, and exactly what Kurzweil said it was. The catalysis roadmap and the MCM-41 work are equally real. The paper Kurzweil leaned on, “Advances in Mesoporous Molecular Sieve MCM-41” (1996), has drawn 578 citations, and MCM-41 went on to anchor a research program that has never stopped. Searching three decades of the chemistry literature, mentions of MCM-41 mesoporous sieves climbed from a single paper in 1991 to a steady 100 to 260 every year through 2025. As a factual matter, Kurzweil cited live, load-bearing science. Verdict on the citations themselves: verified.

Then the story forks.

The molecular-precision chemistry Kurzweil pointed at didn’t stall — it became one of the most productive corners of modern science. But it walked away from his assembler entirely. The clearest evidence is single-atom catalysis, a field that didn’t exist as a named discipline when The Singularity Is Near went to print. The concept was formalized in 2011; by the end of 2024 the field had produced more than 8,000 papers. In our literature holdings, annual single-atom catalysis output rose from 4 papers in 2010 to 366 in 2025 — a near-vertical curve. This is, quite literally, chemistry at the single-atom scale: one metal atom anchored to a support, acting as the entire active site. It is the molecular precision Kurzweil promised. It just isn’t building anything. It’s catalyzing reactions.

The patents make the divergence concrete. US 12,544,749, granted in February 2026, describes preparing a single-atom catalyst for the oxidative coupling of methane using chemical vapor deposition — heating a tungsten hexacarbonyl precursor through a tightly specified temperature ramp (150–250 °C at fractions of a degree per minute) to deposit metal in a single-atom or single-molecule state. The stated payoff is that “the use of metal raw materials can be minimized” while reactivity is maximized. That is atomic economy in the service of cheaper chemical plants — not a desktop fabricator.

The same is true of how catalysts are now designed. US 12,494,270, granted December 2025, lays out a method for designing a ternary PtFeCu catalyst by building a database of catalytic activity using a machine-learning neural-network potential, then running Monte Carlo simulations to find the thermodynamically stable nanoparticle. The claims specify exact atomic compositions — Pt₀.₇₈Fe₀.₀₉Cu₀.₁₃ — chosen by computing the surface separation energy as oxygen adsorbs onto each candidate alloy. US 12,102,992 does something parallel for ethylene oligomerization, using density-functional-theory transition-state models to predict which new ligand will deliver the right activity and selectivity before anyone makes it in a flask. Atom-by-atom design, validated computationally, is here. It arrived through quantum chemistry and machine learning, not through nanobots laying down diamondoid.

That computational turn now has industrial muscle behind it. Meta’s Fundamental AI Research lab and Carnegie Mellon built the Open Catalyst Project around datasets — OC20 and OC22 — comprising 1.3 million molecular relaxations distilled from over 260 million DFT calculations, explicitly to train models that discover catalysts for hydrogen production and renewable-energy storage. Microsoft’s Graphormer won the associated challenge. The goal of all this compute is not Kurzweil’s universal assembler; it’s better catalysts for clean fuels and fertilizer.

And the descendants of MCM-41 found their killer application somewhere Kurzweil never flagged: carbon capture. The single most-cited paper our search surfaced in the MCM-41 lineage isn’t about manufacturing at all — it’s a 2002 study of a “polyethylenimine-modified mesoporous molecular sieve of MCM-41 type as a high-capacity” CO₂ adsorbent, the “molecular basket” concept, at over 1,000 citations. Its sibling on CO₂ capture sits at 761. The torch then passed to metal-organic frameworks, the engineered-porosity materials that eclipsed zeolites for many uses; the 2013 review “The Chemistry and Applications of Metal-Organic Frameworks” has accumulated more than 16,000 citations, making it one of the most-cited chemistry papers of the era. MOFs are now a real market: the carbon-capture MOF segment was valued at roughly $1.4 billion in 2024, with companies like Svante and Nuada running pilots at 1–30 tonnes of CO₂ per day.

The engineered molecular cages Kurzweil cited to win an argument about manufacturing turned out to be tools for pulling carbon out of the air.

The scorecard

What Kurzweil missed (and what he nailed)

The pattern here is subtle, and it recurs across this scorecard project. Kurzweil’s facts were sound. His direction was sound — molecular-scale precision did become central, and arguably more central than even he forecast, given how thoroughly single-atom catalysis and AI-driven design now dominate the field. What he got wrong was the teleology. He read catalysis as an on-ramp to a specific destination, the universal molecular assembler. The science took the precision and drove it somewhere more useful and more boring: cleaner methane conversion, oxygen-reduction catalysts for fuel cells, CO₂ scrubbers. Smalley’s narrow point — that you can’t make a general-purpose atom-placing robot arm — has held up for twenty years. Kurzweil’s broader instinct — that we’d gain exquisite control over molecular events — also held up. They were arguing past each other, and reality split the difference.

There’s a coda worth noting. Gabor Somorjai, whose catalysis work underlies the roadmap Kurzweil cited, died in 2025 after six decades building the molecular understanding of how surfaces drive reactions. His legacy isn’t a tabletop assembler. It’s the predictive, atom-level picture of catalysis that lets a machine-learning model now propose a Pt₀.₇₈Fe₀.₀₉Cu₀.₁₃ nanoparticle and be right. That is a quieter achievement than nanobots. It is also a real one, shipping in granted patents this year.

The lesson for forecasters: when a futurist cites a paper, check two things separately. Is the citation true? And is the inference the citation is carrying true? Kurzweil’s footnotes pass the first test cleanly. They fail the second — not because the science fizzled, but because it succeeded at something else entirely.

Method note

This scorecard draws on a corpus of roughly 9.3 million patent records and several hundred million scientific papers, queried by full-text search for catalysis, zeolite, mesoporous-sieve, and single-atom-catalyst terms, with publication-year trends used to establish each field’s trajectory. The most relevant recent patents were read in full — title, abstract, and claims — to confirm what is actually being built rather than merely counted. Citation counts and market figures were verified against the underlying records and current public reporting accessed this week. Prediction text is drawn from a structured extraction of The Singularity Is Near; the 2024 restatements are quoted from The Singularity Is Nearer.