Ch 1 · The bug in classical physics — Quantum Essay

From Coffee Cup to Willow Chip: A Quantum Physics Essay

Around 1900, physics had a bug. The math at the time said a warm cup of coffee should kill you with X-rays. Stand next to a fireplace and you'd be incinerated. That's literally what the equations predicted. They called it the ultraviolet catastrophe.

When your math says infinity and reality says no, the math is wrong.

The rule in science is simple. When your math says infinity and reality says no, the math is wrong. So physics had to figure out what was actually happening.

ρ(ν, T) = (8π ν² / c³) k T

INFINITY!

As ν rises, the formula goes to infinity.

Reality didn't agree. Physics had a bug. Planck would spend the rest of his life trying to explain it away.

A German physicist named Max Planck found the answer, and he hated it. He proposed that energy doesn't flow smoothly. It comes in tiny discrete chunks. Plug in one new constant, called h, and the infinities vanish. The math suddenly matches reality.

Planck spent the rest of his life trying to explain his own discovery away. The universe is actually built that way.

QuantizeReset wave

PLANCK'S CONSTANTh = 6.626 × 10⁻³⁴ J·s

E = h f

Continuous. Smooth. The classical view of how energy flows. It predicts infinity.

Five years later, Einstein looked at Planck's idea and said, no, this is not a trick. Light itself comes in particles. He called them photons. You can count them one at a time. He won his Nobel Prize for that work, not for relativity.

He won his Nobel Prize for this work. Not for relativity.

That's when the quantum stopped being a bookkeeping shortcut and became a thing.

METAL SURFACEBELOW THRESHOLD · No electrons ejected

FREQUENCYf = 30

KEmax = h f − φ

Below threshold. Brightness no longer matters. Each individual photon lacks enough energy. Counted, not poured.

A Danish physicist named Niels Bohr asked the next obvious question. If light comes in chunks, why not matter? He worked out that electrons can only sit on specific energy rungs around an atom. They jump from rung to rung instantly. They never exist in between.

That one idea is why chemistry works. It's why the periodic table isn't random.

That one idea is why chemistry works. It's why the periodic table isn't random. It's why every transistor in your data center behaves the way it does.

n=1n=2n=3n=4n=5n=6

3 → 24 → 25 → 26 → 2

En = −13.6 / n² eV · λ = h c / ΔE

Δ E = -1.89 eV. Photon emitted at 656 nm (Hα · red). The hydrogen Balmer series is exactly these four lines.

Then a French aristocrat named de Broglie ran the math the other direction. If light, which everyone knew was a wave, also acts like a particle, then a particle should also act like a wave. He was right. Electrons have a wavelength. Shoot them at a crystal and they diffract like ripples in a pond.

After that, the universe stopped being either-or. Everything is both.

After that, the universe stopped being either-or. Everything is both.

SOURCESLITSSCREEN

Wave (don't look)Particle (look at slits)

λ = h / p

Wave. The beam interferes with itself. Bright bands appear where peaks reinforce, dark bands where they cancel. Each electron goes through both slits.

Now the strange part. A physicist named Schrödinger worked out the equation that describes how these wave-particle objects behave. The answer was unsettling. A particle doesn't actually have a position until you measure it. Before you look, it's in a probability cloud of every possible position at the same time.

That isn't a metaphor. That's the math.

That's superposition. The famous cat in the box, alive and dead simultaneously until you open the lid. That isn't a metaphor. That's the math. And that math runs your MRI, your laser, your flash storage, every semiconductor in your environment.

−3σμ+3σ

MeasureMeasure × 10ResetTrials0

P(x) = |ψ(x)|² · Gaussian · σ = 1

Click Measure to collapse the cloud. The particle will appear at one position.

Then Einstein, ironically, surfaced one more weird thing and spent thirty years trying to disprove it. Two particles can be linked together so that measuring one instantly determines the state of the other, regardless of distance. He called it spooky action at a distance. He hated it.

Without entanglement, there is no quantum computing.

A physicist named John Bell put it on trial in 1964 with a clean experiment. Spooky action won. Einstein was wrong. Particles really can be entangled.

That entanglement is the resource. Without it, there is no quantum computing.

ALICE?|0⟩ + |1⟩click to measure

entangled · ∞ km

BOB?|0⟩ + |1⟩click to measure

Trials0Correlated— / —Reset pair

|ψ⟩ = ( |00⟩ + |11⟩ ) / √2

Pure Bell state. Both qubits are simultaneously |0⟩ and |1⟩, perfectly correlated.

In 1981, Richard Feynman, who was the smartest physicist of his generation and knew it, said something that should have been obvious but nobody had said out loud. He said: nature isn't classical. So if we actually want to simulate nature, stop trying to do it on classical computers. Build the computer out of the same physics.

Three hundred qubits hold more simultaneous states than there are atoms in the observable universe.

That's the qubit.

A normal bit is a light switch. On or off. Zero or one. A qubit is that switch in superposition, holding zero AND one at the same time. Two qubits hold four states simultaneously. Three hold eight. Ten hold a thousand and twenty-four states all at once.

Three hundred qubits hold more simultaneous states than there are atoms in the observable universe. That's not an exaggeration, that's the math. Two to the three hundredth power exceeds ten to the eightieth, and ten to the eightieth is the atom count of everything we can see.

states held simultaneously1.13 × 10¹⁵with 50 qubits

1 qubit50 qubits300 qubits

1one bit
10⁹CPU transistors
10¹⁸grains of sand
10²²stars
10⁵⁰atoms on Earth
10⁸⁰atoms in universe

More than the transistors in every CPU on Earth combined.

That brings us to last year. December 2024. Google announced a quantum chip called Willow. One hundred and five qubits.

Quantum computing stopped being a science demonstration and became an engineering problem. Engineering problems get solved.

They ran a benchmark calculation. It finished in under five minutes. The same calculation on the world's fastest classical supercomputer would take ten septillion years. Write it out. Ten followed by twenty-five zeros. That's roughly a quadrillion times longer than the universe has existed.

Now here's where it gets philosophical. The lead of Google Quantum AI, a man named Hartmut Neven, wrote in the official launch blog that the calculation could not have happened here. There isn't enough universe for it. He invoked the multiverse interpretation of quantum mechanics, first proposed by David Deutsch, and said the calculation appears to have been performed across parallel universes, with the answers interfering back into ours.

I want to be clear. That's not a science fiction writer talking. That's the head of the program, writing in the official launch announcement.

But the multiverse line is the headline. The actual news from Willow is more practical and more important.

For thirty years, the entire problem with quantum computing was that quantum systems get noisier as you add qubits. Errors compound. That's why we couldn't build them at scale. Willow is the first chip in history where adding more qubits makes the system MORE accurate, not less. They proved it by scaling from a three by three grid to a five by five to a seven by seven. Every time they made it bigger, the error rate dropped exponentially.

That means quantum computing stopped being a science demonstration and became an engineering problem. Engineering problems get solved.

physical qubits17 qubits

logical error per cycle10⁻²10⁻³d=3d=4d=5d=6d=7logical error per cyclesurface code distanceeach step of 2 in d roughly halves the error0.60% per cycle

Code distance

d = 3d = 5d = 7

d=3: 0.60% • d=5: 0.280% • d=7: 0.140%

Each step of 2 in code distance halves the logical error rate. Below threshold, scaling helps. Above threshold, scaling hurts. Willow lives below.

At which point a thoughtful person asks the obvious question. If quantum computers are this powerful, and AI is the technology eating the world, why aren't they merged together? Why isn't there a quantum AI?

The data loading step eats every speedup the quantum computer was supposed to give you, before you've even started.

The answer is more interesting than the question.

They solve different problems. AI workloads are billions of parallel matrix multiplications on enormous piles of correlated data. GPUs were literally built for that. Nvidia got rich because their hardware happened to be a perfect fit for what AI needs.

Quantum computers don't beat GPUs at that. Quantum's real advantages are narrow and specific. Factoring large numbers, which breaks RSA encryption. Unstructured search, with a modest speedup. And simulating quantum physical systems like molecules and materials. That's the actual sweet spot.

Neural network training is not in that sweet spot. There is no known quantum matrix multiplication that beats a GPU.

There are three concrete bottlenecks.

First, scale. Willow has one hundred and five physical qubits. After error correction, that becomes one logical qubit. Useful AI on quantum would need millions of logical qubits. We are five to seven orders of magnitude away from that. Not five times. Five orders of magnitude.

Second, and this is the killer, the data loading problem. To run AI on a quantum computer, you have to load classical training data into quantum states. That loading step takes time proportional to the size of the data. It eats every speedup the quantum computer was supposed to give you before you've even started. There's a researcher named Scott Aaronson who has been hammering this point for fifteen years and he keeps being right.

Third, dequantization. Every couple of years a researcher claims a quantum machine learning speedup. Within months, a clever classical algorithm matches it. An undergraduate named Ewin Tang did this to several major claims and torched them. The translation is that a lot of so-called quantum AI papers turn out to be solvable on a laptop once someone gets motivated.

So where IS the overlap happening today? It's happening, but it's narrow. Classical AI is being used to RUN quantum hardware. Google uses machine learning to calibrate qubits and decode errors in real time on Willow. There are hybrid algorithms where a classical optimizer drives a small quantum circuit for chemistry and optimization problems. And AI is being used to design better quantum algorithms.

The real first intersection won't be training your chatbot. It will be quantum-accelerated drug discovery and materials science, where the underlying problem is already quantum. Probably ten years out.

Matrix multiplicationFactoring 2048-bit RSAUnstructured searchQuantum chemistry simulation

GPU · classical~10¹⁵ FLOPsnative

QUANTUMno advantagedata-loading kills it

GPU wins. Neural network training is matrix multiplication at scale. GPUs were built for it. There is no known quantum matrix multiply that beats a GPU.

Which brings us to why we're actually having this conversation.

Your encryption has a clock on it now.

BlueAlly is not selling you a quantum computer. Nobody is, yet. But your encryption has a clock on it now. Google's own team estimates that breaking modern RSA encryption is roughly ten years out. The intelligence community is already operating under the assumption that adversaries are doing harvest-now-decrypt-later attacks. They're scooping up encrypted traffic today, sitting on it, and waiting for the quantum hardware that decrypts it.

NIST has already published the post-quantum cryptography standards. The migration is real, it's underway, and it isn't optional.

The conversation worth having isn't about quantum computers. It's about what in your environment is on the harvest-now-decrypt-later list. What encrypted data, if it were decrypted in 2035, would still hurt you? Customer financial records. Healthcare data. M&A communications. Anything with regulatory tails. Anything with a long secrecy half-life.

That's the inventory worth doing. And that's the conversation that turns the physics lecture into something actionable.

HealthcareFinancial servicesDefense & intelligencePharmaceuticalTechnology & SaaSLegalEnergy & utilitiesRetail & e-commerce

Secrecy half-life50+ years

Likely decryption window2032 to 2037

Data at risk

Patient records (HIPAA)
Clinical trial data
Genomic + biomarker data
Insurance and billing

From Coffee Cup to Willow Chip: A Quantum Physics Essay

Data at risk

Recommended posture