Moore's Law Is Slowing — and the Curve It Rode On Doesn't Care
June 21, 2026 · Ray Kurzweil, The Singularity Is Near~5 min read
Pick a year in the last six decades and someone is announcing that Moore's law is finally dead. This year they have the best evidence yet. Intel's 18A is in volume and 14A is on the bench, the labels have dropped past "nanometer" into the angstrom range, and each shrink now costs more and buys less than the one before. The obituary writes itself. But there's a quieter fact from the same month that breaks the obituary's logic: in 2026, quantum error correction stopped being a physics curiosity and became an engineering discipline — and if you've read Ray Kurzweil, you know exactly what that timing means.
The thing that's dying was never the thing that mattered
Here's the move almost everyone gets wrong. They watch transistors stop shrinking and conclude the exponential is over. Kurzweil's whole argument in The Singularity Is Near is that this confuses a runner with the race. Moore's law — cramming more transistors onto silicon — was never the source of the exponential. It was the fifth paradigm to carry it. Before silicon there were four others: electromechanical calculators counting the 1890 census, relays cracking Enigma, vacuum tubes calling the 1952 election for Eisenhower, then discrete transistors. Each one rode a smooth S-curve up, flattened against a physical wall, and got quietly replaced. Computing power per dollar kept doubling straight across all five handoffs, indifferent to which technology happened to be carrying it. So when people point at struggling 14A yields and say "the exponential is stalling," they're pointing at one runner's cramping legs and calling it the end of the race.
Watch where the next runner already is
The part that makes the pattern uncanny is the timing of the handoffs. The next paradigm doesn't show up after the old one collapses, panting and late. It's already running in some unglamorous niche while the incumbent is still winning. Kurzweil's favorite example: around 1960, engineers were still shrinking vacuum tubes for the last drops of performance — and the transistor that would replace them was already out in the world, humming inside cheap portable radios, beneath everyone's notice. The replacement always arrives early and disguised as a toy. Which is why the most interesting tech news of June 2026 isn't a faster GPU. It's that quantum error correction crossed a threshold this year: vendors are reporting logical qubit counts near 90–100 — QuEra around 96, Quantinuum around 94 with error rates under 0.01% — with decoding now happening in well under a microsecond. Those aren't products you can buy. They're a transistor in a portable radio: the next runner, already on the track, while everyone stares at silicon's cramping legs.
Computing price-performance is a cascade of S-curves, not one line: relays → vacuum tubes → transistors → integrated circuits → Moore's law, each rising, hitting a wall, and handing off to a successor already running in a niche. As Intel's angstrom nodes (18A→14A) flatten in 2026, quantum error correction crosses into engineering reality (QuEra ~96 / Quantinuum ~94 logical qubits, sub-microsecond decoding). Framework: Ray Kurzweil, The Singularity Is Near. Figures per 2024–2026 reporting; QEC counts vary by vendor. A reflection, not an investment forecast.
Why your gut keeps getting this wrong
If the pattern is this clean, why does the "Moore's law is dead" headline land every single year? Because human intuition runs on a linear model and the world runs on an exponential one, and the two diverge most violently right at the bends. Inside one paradigm — the flat top of an S-curve — progress really does feel like it's stalling, because locally it is. Your eyes are honest; they're just zoomed in too far to see the next curve lifting underneath. Kurzweil's correction is almost arithmetic: stop extrapolating the single technology in front of you, and extrapolate the staircase of paradigms instead. The first view says silicon is running out of room, which is true. The second view says the exponential has survived this exact moment four times already, which is also true — and far more predictive of what 2030 looks like.
The honest caveat (so this isn't just cheerleading)
None of this is a promise that quantum computers will be on your desk by Tuesday, and Kurzweil himself is careful here in a way his fans often aren't. The macro trend is what's predictable; the specific runner is not. The S-curve cascade says a successor will keep the exponential alive — it does not say which one, or that today's loudest candidate is it. Logical qubits in the 90s are real and genuinely new, but they are not yet a machine that beats your laptop at anything you'd pay for, and plenty of niche transistors-in-radios have fizzled before scaling. The disciplined read isn't "quantum saves Moore's law." It's narrower and more durable: the exponential has never depended on any one technology surviving, so the death of any one technology has never been evidence against it.
What this means for you
So the next time a headline buries Moore's law, do two things. First, notice that they're eulogizing a runner, not the race; the right question is never "is this technology slowing" but "where's the next one already running." Second, resist the equal-and-opposite error — the next runner is real, but which toy on the track becomes the baton is genuinely unknowable, so bet on the staircase, not on a single step. The people who keep getting surprised by technology aren't short on information. They're zoomed in on one S-curve, mistaking its flat top for the edge of the world, while a hum they're not listening for rises somewhere off to the side.
A paradigm flattening out is not the exponential dying. It's the moment to stop watching the runner who's tiring and find the one already on the track — usually disguised as a toy nobody's buying yet.
Moore's law can slow all it likes. The curve it rode in on has changed horses four times already, and never once asked permission.
Source: framework from Ray Kurzweil, The Singularity Is Near (the law of accelerating returns runs as a cascade of S-curves; Moore's law is the fifth paradigm, and the next paradigm is already in a niche before the old one's wall). Real-world basis: June 2026 reporting on Intel's 18A/14A angstrom-era nodes and rising scaling cost, alongside 2024–2026 quantum-error-correction milestones (QuEra ~96 and Quantinuum ~94 logical qubits, sub-microsecond decoding) described as QEC's shift from physics curiosity to engineering discipline; exact qubit counts vary by vendor and report. A reflection, not an investment forecast.
这规律既然这么干净,为什么"摩尔定律已死"的标题年年都能击中人?因为人的直觉跑的是线性模型,而世界跑的是指数模型,两者偏差最剧烈的地方,恰恰在每一道弯。困在一个范式内部——也就是某条 S 曲线被压平的顶端——进步确实让人觉得在停滞,因为局部地看,它真就在停。你的眼睛没撒谎,只是放大得太近,看不见下面那条新曲线正在抬头。库兹韦尔的纠正近乎一道算术:别去外推你眼前这一门技术,去外推那一级级范式搭成的台阶。第一种看法说硅没地方可缩了,这是真的。第二种看法说这条指数早已四次熬过了眼下这一模一样的时刻,这也是真的——而且对 2030 年长什么样,预测力强得多。
所以下次再有标题给摩尔定律盖棺,做两件事。第一,看清他们悼念的是一名选手,不是这场赛跑;该问的从来不是"这门技术是不是变慢了",而是"下一门已经在哪儿跑了"。第二,别犯那个等而反之的错——下一棒是真的,但跑道上哪枚玩具会接到棒,是真的没法预知,所以押那级台阶,别押某一格台阶。那些总被技术打个措手不及的人,缺的不是信息。是他们把视线放大在一条 S 曲线上,把它压平的顶端错当成了世界的边缘,而某个他们没在留神去听的嗡嗡声,正从旁边某处升起。
ほぼ全員がここで踏み外す。トランジスタが縮まなくなるのを見て、指数曲線は終わったと結論づける。カーツワイルが『シンギュラリティは近い』で一貫して説くのは、まさにそれが「走る者」を「この競走そのもの」と取り違えている、ということだ。ムーアの法則——シリコンにより多くのトランジスタを詰め込むこと——は、その指数の源だったためしがない。それは指数を運んだ第五走者にすぎない。シリコン以前に四走者がいた。1890年の国勢調査を数えた電気機械式計算機、エニグマを破った継電器、1952年にアイゼンハワー当選を言い当てた真空管、そして個別トランジスタ。どれも滑らかな S 字曲線を駆け上がり、物理の壁にぶつかって平らになり、静かに交代させられた。一ドルあたりの計算力は、この五度の手渡しを貫いて倍々に伸び続けた——いま担いでいるのがどの技術かには、まるで無頓着に。だから 14A の難航する歩留まりを指して「指数が止まる」と言う者は、ある一走者の攣った脚を指して、それを競走全体の終わりと呼んでいるのだ。