Kurzweil Scorecard: The Sun Won. The Fuel Cells Lost.

In Chapter 8 of The Singularity Is Near, Kurzweil staked his energy
forecast on a paired bet: nanoengineered solar panels and microscopic
fuel cells, both manufactured like silicon chips, would together replace
the centralized fossil grid by sometime in the 2030s. Twenty-one years
later, half of that bet is winning faster than even he expected. The
other half died in a Boston warehouse in 2013.

The predictions

The batch contains four linked claims from the GNR (genetics,
nanotechnology, robotics) chapter:

1. “By 2005 at least one company was pioneering microscopic fuel cells
   using MEMS technology, manufactured like electronic chips and offering
   energy-to-size ratios significantly exceeding conventional technology.”
2. “Nanoengineered fuel cells and solar cells will provide clean
   energy.”
3. “Nanoengineered solar panels and microscopic fuel-cell technologies
   will ultimately power devices, cars, and homes in a distributed,
   renewable, and clean fashion, and these decentralized energy systems
   will not be subject to disaster or disruption.”
4. “Future technologies will enable humanity to reverse remaining
   environmental destruction.”
   (all from The Singularity Is Near, ch. “Promise and Peril of GNR”)

In The Singularity Is Nearer (2024), Kurzweil restates the solar
side with new confidence: “From 3.6 percent in 2021, it would take only
about 4.8 doublings to reach 100 percent, which would put us at 2032
to meet all of our energy needs from solar alone.” The fuel-cell
companion prediction quietly disappears from the 2024 book.

Where we actually are

Solar arrived ahead of schedule — but not from atomically precise
nanotech.

Ember’s 2026 Global Electricity Review puts solar at 8.7% of world
electricity in 2025, generating 2,778 TWh. The world added a record
647 GW of new solar capacity in 2025 alone, and renewables together
edged past coal (33.8% vs. 33.0%) for the first time since 1919.
Solar is doubling roughly every 2.3 years as a share of the mix —
slightly faster than the 28-month doubling Kurzweil cited.

The efficiency story is the more interesting one for the prediction.
Kurzweil’s bet was that nanoengineering would crack the silicon
ceiling. It is — but through perovskite tandem stacks, not through
the diamondoid mechanosynthesis Drexler imagined. A 2023 paper in
Science titled “Interface passivation for 31.25%-efficient
perovskite/silicon tandem solar cells” (10.1126/science.adg0091)
described a 1.17 cm² device that broke the 31% certified efficiency
barrier — the first device to clearly exceed silicon’s 29% theoretical
limit using nanoscale interface engineering with a buckminsterfullerene
contact. By late 2025, multiple Chinese labs had pushed all-perovskite
tandems past 30.1% certified efficiency using dipolar passivation,
and a Nanjing-led group reported 31.9% with low-lead chemistry.

The patent record tracks the science. US 12,598,856, granted April
2026 to a perovskite-silicon tandem developer, claims a stress-relief
geometry where a step surface or trench in the top subcell is
“at least partially filled with another material such as an insulator
support or electrically conductive support to transfer stress away
from the absorber layer.” That sentence is the entire commercial
problem in miniature: perovskites are efficient but mechanically
brittle, so the patent landscape since 2024 reads as a parade of
mechanical-stability patches around the chemistry. Top assignees in
perovskite solar since 2022 include CubicPV, Swift Solar, Panasonic,
Contemporary Amperex (CATL), and the Alliance for Sustainable Energy.
Ahead of schedule on the destination. Wrong mechanism on the path.

MEMS micro fuel cells: the spectacular dead end.

Kurzweil’s 2005 claim that at least one company was building
microscopic fuel cells “manufactured like electronic chips” was true
when written. Lilliputian Systems, an MIT spinout founded in 2002,
raised over $150 million for a butane-fed silicon-MEMS solid-oxide
fuel cell aimed at cell-phone charging. Intel’s foundries fabbed the
MEMS dies. Brookstone agreed to retail the consumer product, branded
Nectar. Mechanical Technology Inc. spent roughly $60 million on a
parallel methanol-MEMS effort.

The patent corpus traces the rise and fall: US grants for “MEMS &
fuel cell” peaked at three per year in 2006 and 2012, and zero have
issued since 2019. Lilliputian missed launches in 2012 and 2013,
sold off its IP by late 2013, and dissolved. MTI shelved its program
in 2011. The technology worked. The market killed it. Lithium-ion
cell costs fell roughly 80% per megawatt-hour between 2012 and 2020,
which collapsed the energy-density advantage that had been the
entire commercial premise. Kurzweil’s literal claim — that MEMS fuel
cells had reached working prototypes by 2005 — verifies as historical
fact. His implied
claim — that this lineage would scale into homes, cars, and devices —
was overtaken by a different battery chemistry he didn’t bet on.

Distributed, renewable, not subject to disaster: arriving — through
software, not chips.

The decentralized-energy half of the prediction is being built. US
virtual power plant capacity hit 37.5 GW in 2025 (Department of
Energy and Wood Mackenzie figures), up 14% year over year, with the
DOE targeting 80–160 GW by 2030. The patent landscape is consistent
with that: filings on virtual-power-plant aggregation accelerated
sharply from one or two grants per year before 2020 to ten in 2025
and four already in early 2026. US 12,579,588, granted March 2026,
claims a method for “generating nodal power networks” and
“determining virtual power plant hubs along the transmission route,”
with “market depth data for a geolocation exchange” — the post-2020
VPP patents are increasingly indistinguishable from electricity-market
microstructure patents, which is itself a tell about where the field
is heading.

But the substrate is rooftop solar plus lithium-ion batteries plus
load-flexible EVs and water heaters, coordinated by software. Not
microscopic fuel cells. Kurzweil got the topology right and the
hardware wrong.

The “not subject to disaster or disruption” half is more fragile than
it sounds. Texas in February 2021 lost roughly 4.5 million customers
during a freeze that knocked out gas, wind, and solar simultaneously,
demonstrating that distribution alone doesn’t immunize against
correlated weather. The 2025 hurricane season again exposed
distribution-network vulnerabilities along the Gulf Coast. Distributed
generation helps; it is not the magic shield Kurzweil sketched.

Reverse environmental destruction: the embarrassing column.

Direct air capture is the closest concrete proxy for Kurzweil’s
“reverse” claim, and the numbers are unforgiving. Climeworks’ Mammoth
plant in Iceland — the world’s largest, online May 2024 — has a
nameplate of 36,000 tons of CO₂ per year. Global DAC capacity rose
from roughly 59 ktCO₂/yr in 2024 to a projected 569 ktCO₂/yr in 2025,
mostly via 1PointFive’s Stratos plant in Texas. Two flagship
North American projects (Carbon Engineering’s Dreamcatcher and
CarbonCapture Inc.’s Project Bison) were canceled in 2024, removing
roughly 6 Mt/yr of planned capacity.

The patent landscape and literature tell a similar story of energetic
beginnings. DAC patent grants jumped from two in 2020 to 28 in 2025,
with 11 already granted in early 2026. Top assignees include 8 Rivers
Capital, Carbon Capture Inc., Air Company, AirMyne, Climeworks, and
Arizona State University. US 12,616,934, granted May 2026, describes
a multi-stage absorption-desorption reactor using a regenerable base
solution — a respectable engineering refinement, but the parent
technology has been around for decades.

The literature is louder still: a 2022 review in Energy Conversion
and Management on DAC progress has 222 citations; a 2023 paper in
Nature Communications on electrochemical DAC using neutral red as
a redox-active material (10.1038/s41467-023-35866-w) has 139.
But annual capture in 2025 is on the order of 0.0006 gigatons. The
IPCC’s 1.5°C-aligned scenarios call for cumulative removals of 100–1,000
gigatons by 2100. We are roughly six orders of magnitude short of
“reversing environmental destruction” on Kurzweil’s 2030s timeline.

The scorecard

What Kurzweil missed (and what he nailed)

The pattern across these four predictions is consistent. Kurzweil
read the deflation curves correctly — solar’s price collapse, the
exponential improvement in materials science, the inevitability of
decentralization. He read the substrate incorrectly. The cheap
distributed energy he predicted is arriving on time. It is being
built with crystalline silicon, polycrystalline perovskite films,
spinel-cathode lithium chemistry, and grid-aggregation software —
none of which correspond to the diamondoid-nanotech roadmap he and
Drexler championed in 2005. The fuel cells he paired with solar
turned out to be a 2000s-era footgun: the right answer to the wrong
question.

The reverse-environmental-destruction claim is in worse shape, and
not because the technology is implausible. DAC works; it just
doesn’t scale at gigaton rates on a 2030s budget. The combination of
unfavorable thermodynamics (CO₂ is dilute), unfavorable economics
(removal costs $400–$1,000/ton vs. carbon prices well below that),
and unfavorable politics (project cancellations in 2024) makes the
prediction plainly behind schedule. Kurzweil’s optimism here is the
optimism of someone forecasting from chip economics into a domain
where chip economics don’t apply.

The scorecard is not a verdict on Kurzweil. It is a verdict on
forecasting itself: trends in price-performance are easier to
extrapolate than trends in which technology will win. Get the
deflation right; get the substrate wrong. The destination is closer
than people think. The path is almost never the one the futurist
named.

Method note

For each prediction, we counted patent grants by year over the past
two decades using the full-text index across our 9.3 million-document
US patent collection, then read the most recent grants in detail
to confirm what is actually being claimed. We pulled top-cited
papers from a 357-million-record scientific literature corpus and
cross-checked the numbers against published industry data from Ember,
the IEA, the US Department of Energy, and trade press for DAC and
VPP capacity. Quotes from Kurzweil are taken from The Singularity
Is Near (2005) and The Singularity Is Nearer (2024).