Kurzweil Scorecard: He Trusted the Market to Steer. The World Reached for Export Controls.

Ray Kurzweil’s social predictions are usually the ones people skip. The hardware curves are testable; the politics feel like mood. But the dozen forecasts in this batch — clustered around chapter 8 of The Singularity Is Near, “The Promise and Peril of GNR,” plus his closing personal credo — turn out to be the most falsifiable of all, because the 2020s actually ran the experiment.

The result is strange. Kurzweil’s predictions about atoms — energy, abundance, the physical infrastructure of plenty — are aging well, in one case spectacularly. His predictions about institutions — that an open free market with light oversight would prove the safest way to steer genetics, nanotechnology, and robotics — are the ones cracking. The technologist read the technology correctly and misjudged the politics. That is the opposite of how futurists usually fail.

The bet at the center

Everything in this batch hangs on one load-bearing claim. Kurzweil argued that maintaining an open free-market system for incremental progress, with each step subject to market acceptance, would provide the most constructive environment for technology and strong AI to embody human values (ch. “Promise and Peril of GNR”). He still believed it two decades later. In The Singularity Is Nearer he restated the case almost verbatim: “each step toward superintelligence is subject to market acceptance… AI is emerging from a deeply integrated economic infrastructure, it will reflect our values because in an important sense it will be us.”

The corollary was a warning. Trying to control advanced technologies through secretive government programs, combined with inevitable underground development, would create an unstable environment in which dangerous applications dominate (ch. “Promise and Peril of GNR”). Openness safe; secrecy dangerous. That was the thesis.

The 2020s chose control. The United States built an export-control regime that splits the world into tiers, restricts H100/H200/Blackwell-class accelerators to adversary states, and added 65 Chinese entities to its restricted list in 2025 alone. When the policy wobbled — an August 2025 deal let Nvidia and AMD resume some China sales in exchange for handing Washington 15% of the revenue — it was a negotiation between a government and two firms, not a market clearing. Meanwhile the frontier closed: leading labs keep their best models behind APIs, and Meta delayed its largest open model while signaling future “superintelligence” systems may stay paywalled. The open-market path Kurzweil prescribed is the one policymakers and frontier labs walked away from. His thesis isn’t disproven — it’s untested, because nobody is running it.

The abundance machine showed up — for electrons

Now the part Kurzweil got right, and early. He predicted that emerging technologies would provide clean renewable energy generation and storage (ch. “Ich bin ein Singularitarian”), and in The Singularity Is Nearer he leaned hard on decentralized solar: “nano-based technology will also reduce manufacturing costs by facilitating 3D printing of solar cells… so photovoltaics can be created when and where they are needed.”

The mechanism detail (nano-printed films) hasn’t arrived. The outcome has, faster than almost anyone in 2005 priced in. Solar PV crossed roughly 2.8 terawatts of cumulative installed capacity in 2025, making it the largest power-generation technology on Earth by capacity. It supplied about 7% of global electricity in 2024, up from 5% a year earlier, and contributed more than a quarter of all global energy-supply growth in 2025 — the first time a modern renewable led absolute energy expansion.

The patent record shows the story shifting from generation to orchestration. US 12,494,653, Controlling a hybrid power plant, granted December 2025, claims a method that schedules a co-located wind park, a second renewable source, and a storage system against a live forecast of both power output and electricity price — dispatching stored energy to the moments it’s worth most. The interesting engineering is no longer “can we make the electron”; it’s “can we time it.” US 12,502,992, granted weeks later, claims a solar-plus-storage EV-charging carport — the decentralized, point-of-use generation Kurzweil described, minus the nanotech. The destination is his; the road is ordinary silicon and lithium.

Abundance for atoms didn’t become abundance for people

Kurzweil’s larger claim was that genetics, nanotechnology, and robotics would create extraordinary wealth, overcoming poverty by converting inexpensive raw materials and information into any product (ch. “Promise and Peril of GNR”). Here the scorecard turns. More than a billion people did leave extreme poverty since 1990 — Kurzweil cites the trend approvingly in Nearer — but that engine was globalization and economic growth, not GNR, and it has stalled. The World Bank counted roughly 808 million people in extreme poverty in 2025 (about one in ten), and a March 2026 revision put the 2024 figure near 847 million after the poverty line was raised from $2.15 to $3 a day. More than three-quarters of the remaining extreme poor now live in sub-Saharan Africa or conflict-affected states — places the abundance machine hasn’t reached. The prediction was right that technology would create wealth and wrong that it would dissolve poverty on a 2030s timeline through GNR.

The defensive budget materialized — for the wrong letter of GNR

Kurzweil argued governments should spend tens of billions of dollars per year, under 1% of GDP, to address the existential threat of GNR (ch. “Promise and Peril of GNR”). Split by letter, the bet half-landed. For the “G” — biodefense — the money is real: the FY2025 request put roughly $3.8 billion into the U.S. preparedness office alone, atop a proposed $20 billion mandatory pool across health agencies after COVID. For the AI risk now dominating the conversation, it’s a rounding error: the U.S. AI Safety Institute was slated for about $50 million and has faced cuts; the better-funded UK institute runs near £66 million a year. Tens of billions for biology; tens of millions for the machines. The defensive posture exists, pointed mostly at the threat of 2005 rather than that of 2025.

The awareness arrived; the city didn’t dissolve

Two more landed cleanly. Kurzweil predicted that public awareness of GNR dangers and benefits would need to be raised to create political support for defensive funding without triggering antitechnology mandates (ch. “Promise and Peril of GNR”). It happened — a few years later than “by the 2010s,” but unmistakably. Academic output on AI governance and regulation went from 61 papers in 2018 to over 3,500 in 2025; the EU AI Act passed in 2024; an international network of safety institutes now exists. And he predicted technology is unlikely to be derailed despite cultural rebellion (ch. “Ich bin ein Singularitarian”); in Nearer he was still swatting at “increasingly strident Luddite voices that advocate broad relinquishment.” No relinquishment came. Both on track.

The miss in this group is the city. Kurzweil predicted that as distributed technologies develop, the need to aggregate people in cities will diminish and people will spread out (ch. “Promise and Peril of GNR”). For about eighteen months in 2020–21 it looked prophetic. Then gravity won. U.S. office vacancy sits near 19%, but 77% of new job postings in early 2026 are fully on-site, hybrid is the durable compromise, and the decentralizing technologies he named never displaced the agglomeration economics of cities. Right that the tools would enable dispersal; wrong that people would take them up.

The scorecard

What Kurzweil missed (and what he nailed)

The pattern is sharp once you sort it. On the physics of abundance, Kurzweil was directionally right and sometimes early — solar’s rise outran his own timeline, and the patent frontier has already moved from making clean energy to scheduling it. On the institutions that would govern that abundance, he was wrong in an instructive way. He assumed the safest path was openness, and that markets would encode human values into the technology. The 2020s decided the opposite: that the safest path ran through export controls, entity lists, and a handful of closed labs. His thesis wasn’t refuted by evidence; it was abandoned by choice.

That’s the bias worth naming. Kurzweil’s error here isn’t optimism about timelines — his energy call was, if anything, conservative. It’s that he modeled institutions as if they would behave like markets: incremental, decentralized, self-correcting. Institutions facing a perceived existential technology behaved like states instead — they centralized, classified, and restricted. A futurist who can extrapolate a cost curve across forty years still could not predict that the people in charge would decide his preferred steering mechanism was too dangerous to use.

Method note

This scorecard pairs Kurzweil’s 2005 predictions, and his 2024 restatements, against current evidence. The technology landscape comes from a corpus of roughly 9.3 million patents and several hundred million research papers, queried for deployment trends and read at the level of individual filings; named patents were pulled and read directly. Macro figures on poverty, energy deployment, AI-safety funding, export-control policy, and workplace trends come from public reporting accessed this week — the World Bank, the IEA, government budget documents, and primary news coverage. Verdicts are judgment calls, and the “too early to call” rows mean exactly that: the clock hasn’t run out.