Kurzweil Scorecard: The 2024–2025 Nanotube Obituary, and the Silicon Transistor That Won Anyway

In the second half of 2024, Nantero ceased operations. The company had spent roughly twenty years and seventy-eight million dollars trying to turn carbon nanotubes into a universal memory. It licensed to Fujitsu, partnered with Imec, and as late as 2020 was still publishing patents on lower-voltage nanotube switches. Then it was gone.

Five quarters later, in Q4 2025, TSMC began volume production of its N2 node — the company’s first implementation of gate-all-around nanosheet transistors. The N2 process stacks horizontal silicon channels completely wrapped by their gates, a three-dimensional geometry squeezed into what the industry calls a 2-nanometer-class node.

These two events, months apart, form the cleanest possible verdict on a batch of Kurzweil’s 2005 nanotech predictions. He told us where to look: carbon nanotubes, three-dimensional molecular computing, sub-100nm features by the 2020s. The 100nm prediction came in early and stayed in. The nanotube prediction lost to the nanosheet.

The predictions

This batch collects twelve nanotech claims from The Singularity Is Near (2005). Seven are factual name-checks of 2003–2004 breakthroughs Kurzweil used to anchor his timeline — DNA hands at Ludwig Maximilians Munich, Feringa’s 58-atom solar-powered motor, Nantero’s ten-billion-junction wafer, IBM’s “constructive destruction” of defective nanotubes, the USC–NASA Ames 258-gigabits-per-square-inch self-organized memory. Those are easy to score: they happened.

The other five are forecast rails. “At an exponential shrinkage rate of about a factor of four per linear dimension per decade, key feature sizes for most electronic and many mechanical technologies will fall below 100 nanometers by the 2020s” (ch. “Nanotechnology: The Intersection of Information and the Physical World”). “Drexler-style molecular assemblers will be able to manufacture almost any atomically stable physical device” (same chapter), by 2025. “Nanotechnology will bring the economics of software, including rapid deflation, to hardware and physical products” (ch. “Deflation… a Bad Thing?”). And the bet that tied it all together: “Nanotubes remained the best bet for ushering in three-dimensional molecular computing” (ch. “Nanotubes Are Still the Best Bet”).

In The Singularity Is Nearer (2024), Kurzweil dialed back almost none of this. He writes that “we are on track for nanotechnology concepts using atom-by-atom placement to be implemented sometime in the 2030s,” and cites Drexler’s 2013 estimate that atomically precise manufacturing could produce most objects for roughly two dollars per kilogram. The 2005 bet is still on the table in the 2024 update, just slid into the 2030s.

Where we actually are

The nanotube winter. Nantero’s May 2003 demonstration of a wafer with ten billion nanotube junctions was one of the most cited datapoints in The Singularity Is Near. It was supposed to be the bridge between laboratory curiosity and a commercial nonvolatile memory that would eat DRAM’s lunch. Nantero’s late patents — including US 10,885,978 (“Nonvolatile nanotube switches with reduced switching voltages and currents,” granted January 2021) — describe stacked nanoscopic elements aimed at cutting the write voltage that had always been NRAM’s commercial headache. US 11,239,415 (February 2022) is a hybrid with carbon nanotubes as one electrode and graphene as the other, separated by an insulating layer. These are the patents of a company working hard on a solution to a problem the market was about to stop caring about. In our patent corpus, carbon-nanotube-transistor filings peaked around 2017 at 34 per year, drifted down through 2022 (six filings), and sat at five in 2025. Not a winter of funding — a winter of belief.

Silicon nanosheets ate the 3D computing story. Three-dimensional transistor geometry did arrive in the 2020s. It just arrived made of silicon. TSMC’s N2 process delivers roughly a 15 percent improvement at equal power versus N3E, or a 25–30 percent power reduction at equal performance. Gate-all-around nanosheet transistors are the most three-dimensional thing in mass-market electronics. They are not made of nanotubes. MIT’s RV16X-NANO — the 14,000-transistor carbon nanotube RISC-V chip Shulaker’s group published in 2019 — remains the most advanced logic device ever built from CNTs. Nearly seven years later, it has no commercial successor.

Feature sizes crossed below 100nm a decade earlier than predicted. Intel’s 90nm node shipped in 2003, two years before The Singularity Is Near was published. By the early 2020s the industry was into 5nm-class nodes, and now 2nm-class. Kurzweil’s factor-of-four-per-decade shrinkage roughly matches what happened; his calibration on the starting point was off by a decade.

Molecular assemblers by 2025: no. The Drexler vision — a general-purpose, atomically precise machine that builds almost anything — has not arrived. The field’s most persistent institutional home, CBN Nano Technologies, holds foundational patents including US 8,276,211 and US 8,171,568 (“Positional diamondoid mechanosynthesis,” both granted 2012), plus a cluster of more recent grants around tip-based synthesis and feedstock preparation. They are reagents and recipes, not machines. A 2024 arXiv review of APM’s trajectory notes that the field “remains in its early stages, with applications largely confined to specialized fields.”

DNA nanotechnology did quietly deliver something. The 2005 “DNA hand” from Ludwig Maximilians Munich turned out to be a step on a road Kurzweil didn’t name. DNA origami — Rothemund’s 2006 technique, post-dating The Singularity Is Near — has produced the most biologically functional nanorobots yet built. A 2019 paper from a Chinese team described a DNA origami nanorobot carrying thrombin into tumor vasculature, triggered to unfold by nucleolin expression. Recent reviews describe origami nanorobots selectively opening in acidic endosomes and achieving greater than 80 percent siRNA-mediated oncogene knockdown in triple-negative breast cancer cells. The 2024–2025 literature describes clinical translation as “feasible but imminent.” The mechanism Kurzweil expected was diamondoid mechanosynthesis. What is actually maturing is programmable DNA.

Feringa’s motor won a Nobel; practical applications are still slow. The 58-atom solar motor cited in 2005 became the 2016 Nobel Prize in Chemistry. A 2024 paper from Feringa’s own group (Nature Chemistry, April 2024) reports a formylated overcrowded-alkene motor with substantially improved photon efficiency — the original design converted only about 2 percent of absorbed photons into rotary motion. A 2025 JACS paper describes near-infrared-driven motors powered by upconversion nanoparticles. The research is real. The commercial nanomachine is not.

“Software economics extended to hardware” — wrong mechanism. Hardware deflation has accelerated during the 2020s, but via completely different paths: Chinese contract manufacturing, 3D printing, algorithmic product design, and generative-AI-assisted CAD. Physical consumer electronics have compressed costs the way software did, but the causal mechanism is a supply-chain and software story, not a nanotechnology one.

The scorecard

What Kurzweil nailed, and what he missed

In AI, Kurzweil’s pattern was: right about direction, right about rough timing, wrong about mechanism. In nanotech, he is right about the trend (things keep getting smaller and more three-dimensional) and right about some of the underlying science (Feringa’s motor really did matter), but wrong about the winning substrate. He bet heavily on carbon. The industry bet on silicon nanosheets wrapped in high-k dielectrics. The industry won.

“Nanotubes are still the best bet” is the name of an entire chapter. And the three predictions here that come out as “wrong mechanism” share a family resemblance: Kurzweil expected the material economy to be changed by a discrete nanotech breakthrough — a Drexler-style assembler, a nanotube fabric, a bloodstream-capable molecular machine — when instead the material economy is being changed incrementally, from many unrelated directions at once. Cost compression in physical goods is real. It is not coming from the atomically precise manufacturing route.

He also underestimated how aggressively the semiconductor industry would push silicon into three dimensions. If you assume silicon is a spent force, you reach for nanotubes. The industry did not assume that. The honest lesson: the 2030s claims Kurzweil is still making in The Singularity Is Nearer about AI-guided atomically precise manufacturing are forward extrapolations from a 2005 substrate bet that has now empirically lost. The DNA origami line — not in his 2005 list, added belatedly in the 2024 book — is where the biologically functional nanomachine work actually lives. That should update the priors.

Method note

This post is part of an ongoing project scoring the ~1,100 testable predictions in The Singularity Is Near (2005) against present-day evidence. For each prediction, we searched a local corpus of 9.3 million US patents and 357 million scientific papers to establish trend shape, then pulled the actual claims and descriptions of the most relevant individual patents (such as Nantero’s US 10,885,978 and CBN Nano’s US 8,276,211) and read the key 2024–2025 follow-up literature, including the Feringa group’s efficiency work in Nature Chemistry. Web sources were used to verify current commercial status — Nantero’s 2024 closure, TSMC’s Q4 2025 N2 production ramp, and the MIT CNT computer’s post-2019 trajectory. Direct quotes from The Singularity Is Nearer are drawn from the text as published. Where evidence was ambiguous, we scored conservatively.