AI in quantum computing
Computers that think in waves, not switches
Quantum computers use the strange rules of atoms and light to tackle problems that would take normal computers longer than the age of the universe. AI is helping build them, and they may one day help build better AI. Here is everything you need to understand both, no physics degree required.
Start here
What is a quantum computer?
Every computer you have used stores information as bits. A quantum computer stores it as qubits, which follow the rules of the very small. That one change makes them better at a narrow but important set of problems.
01 · The bit
A light switch
A normal bit is always 0 or 1, off or on. Your phone juggles billions of them, one definite value at a time.
02 · The qubit
A dimmer with a direction
A qubit can be in a superposition: a blend of 0 and 1 with a direction (called phase). When you look at it, it snaps to 0 or 1, with odds set by the blend.
03 · The trick
Waves that cancel
A quantum program is choreography. It arranges qubits so the wrong answers cancel out like opposite waves, and the right answer gets louder. That is called interference.
“A quantum computer tries every answer at the same time.” Not quite. If it did, you would only ever read out one random answer. The real power comes from interference: steering the odds so the right answer is the one you are most likely to see. That only works for problems with the right hidden structure, which is why quantum computers will not replace your laptop.
Side by side
Classical vs quantum, in plain words
A quantum computer is not a faster laptop. It is a different kind of tool, like a telescope next to a pair of glasses.
| Classical computer | Quantum computer | |
|---|---|---|
| Basic unit | Bit: 0 or 1 | Qubit: a blend of 0 and 1 until measured |
| How bits connect | May be correlated; known values can be copied freely | Can be entangled; unknown states cannot be copied |
| How it computes | Step-by-step logic (AND, OR, NOT) | Rotates quantum amplitudes so they reinforce or cancel; probabilities come from the squared sizes of the amplitudes |
| Output | Deterministic or randomized, depending on the program | Measurement outcomes follow quantum probabilities; many algorithms run many times and read the pattern |
| Errors | Extremely rare | Common today; needs heavy error correction |
| Where it lives | Your pocket, at room temperature | Labs and data centers, often colder than outer space |
| Best at | Almost everything: email, games, video, AI chatbots | Simulating molecules and materials, certain math problems, some optimization |
Interactive · why qubits are a big deal
Add qubits and watch the numbers explode
A direct simulation of n qubits stores 2ⁿ complex numbers (amplitudes), here assumed to take 16 bytes each. Every extra qubit doubles that. Some special states and circuits can be simulated far more compactly. Drag the slider.
10 qubits is easy. Keep going.
The physics, made simple
Four ideas run the whole show
You do not need equations. These four ideas explain nearly everything a quantum computer does.
Idea 1
Superposition
Like a spinning coin that is not yet heads or tails. A qubit holds a blend of 0 and 1 until it is measured.
Idea 2
Entanglement
Two qubits share joint quantum correlations, even far apart. Depending on the state and how they are measured, their results can match, be opposite or relate in other ways that ordinary bits cannot fully copy. Einstein called it “spooky action at a distance.” It cannot send messages, but it lets qubits work as a team.
Idea 3
Interference
Quantum amplitudes, not the odds themselves, behave like waves. Line them up and they grow; put them out of step and they cancel. The size of the resulting amplitude sets the odds of each answer, and algorithms use this to boost right answers.
Idea 4
Measurement & decoherence
Looking forces a qubit to pick 0 or 1. Heat, vibration and stray signals “look” by accident, breaking the spell. That is the main reason qubits make errors.
Interactive · interference
Make two waves fight
Slide the second wave out of step. In step, the waves add up (a likely answer). Half a step apart, they cancel to nothing (an answer the algorithm has erased).
In step: the waves add up. This answer gets louder.
Quantum playground
Learn it by playing with it
Three small games that run real quantum math in your browser. Nothing to install, nothing to break.
Tilt the qubit, then measure
Point the arrow up for 0, down for 1, or anywhere between. Each measurement gives a single 0 or 1. Only after many shots do the odds appear.
Hit measure. At 90° it is a perfect coin flip.
Your 2-qubit circuit
Both qubits start at 0. Add gates, read left to right. Bars show the chance of each result; pink bars have a flipped wave (negative phase).
Gates for q0
Gates for q1
Link them
H makes a 50/50 blend · X flips 0↔1 · Z flips the wave · CNOT flips q1 only if q0 is 1
Challenges
Pick a challenge, or just experiment.
Find the hidden star card
One of 16 cards hides a star. Round 1: you search like a normal computer, flipping cards one at a time. Round 2: a quantum computer uses Grover’s algorithm, which needs only about √16 = 4 steps. On a million cards that is roughly 785 steps instead of up to a million.
Round 1 · classical search
Round 2 · Grover’s quantum search
All 16 cards start equally likely (6.25% each). Each step flips the star card’s wave, then reflects every wave around the average. Watch the star grow.
A 100-year story
From a wild idea to working machines
Quantum computing started as a thought experiment. Filter by era to follow how physics became engineering.
- 1900
Energy comes in packets
Max Planck proposes that energy is released in tiny chunks, or “quanta.” Quantum physics is born.
- 1925–27
The rules get written
Heisenberg, Schrödinger and others build quantum mechanics: particles described by waves of probability.
- 1935
Spooky action and a famous cat
Einstein and colleagues question entanglement. Schrödinger imagines a cat that is both alive and dead to show how odd superposition is.
- 1964
Bell’s test
John Bell shows entanglement can be tested in a lab. Experiments by Clauser, Aspect and Zeilinger later prove it is real, earning the 2022 Nobel Prize.
- 1981
Feynman’s challenge
Richard Feynman argues that to simulate nature, which is quantum, you need a computer that is quantum too.
- 1985
A universal quantum computer
David Deutsch describes a quantum computer that could, in principle, run any quantum program.
- 1994
Shor’s algorithm
Peter Shor shows a quantum computer could factor huge numbers fast, which would break much of today’s internet encryption. The field explodes.
- 1995–96
Error correction and Grover
Shor invents quantum error correction. Lov Grover finds a quantum shortcut for searching. NIST builds the first quantum logic gate with trapped ions.
- 1999
The superconducting qubit
Researchers at NEC in Japan control a qubit made from a tiny superconducting circuit, the ancestor of Google’s and IBM’s chips.
- 2001
15 = 3 × 5
IBM and Stanford run Shor’s algorithm on 7 qubits to factor the number 15. Tiny, but a real proof of concept.
- 2016
Quantum in the cloud
IBM lets anyone run programs on a real 5-qubit machine over the internet. Students start writing quantum code.
- 2019
“Beyond classical”
Google’s 53-qubit Sycamore finishes a sampling task in 200 seconds that it estimated would take a supercomputer 10,000 years. Rivals later found faster classical methods, but the race was on.
- 2024
Willow goes below threshold
Google shows that adding more qubits to an error-correcting code now cuts errors instead of adding them. NIST publishes the first post-quantum encryption standards.
- 2025
A Nobel for quantum effects in circuits
John Clarke, Michel Devoret and John Martinis win the Nobel Prize in Physics for showing quantum effects in electric circuits, the basis of superconducting qubits. Google reports a verifiable quantum advantage, and Caltech traps 6,100 atom qubits.
- 2026
Logical qubits at work
IBM and the University of Chicago run a verified computation on 70 error-protected qubits. Google adds neutral-atom machines to its plans and warns that encryption upgrades should start now.
The qubit zoo
Six ways to build a qubit
Nobody knows yet which design will win. Each one is a different bet on how to keep fragile quantum states alive. Pick one to compare.
Leading companies and machines
Who is building what
Qubit counts alone can mislead. Accuracy, connections between qubits and error correction matter just as much.
| Company | Qubit type | Flagship machine | Recent milestone |
|---|---|---|---|
| Google Quantum AI | Superconducting, now also neutral atoms | Willow, 105 qubits | Below-threshold error correction (2024); verifiable advantage with Quantum Echoes (2025) |
| IBM | Superconducting | Nighthawk, 120 qubits | Verified advantage on 70 logical qubits with UChicago (July 2026); targets a fault-tolerant “Starling” in 2029 |
| Quantinuum | Trapped ions | Helios, 98 qubits | 99.92% two-qubit accuracy; dozens of error-protected logical qubits; filed to go public in 2026 |
| IonQ | Trapped ions | Tempo | Record 99.99% two-qubit accuracy (2025); bought Oxford Ionics |
| Microsoft | Topological (experimental) | Majorana 1 | Announced a new qubit design in 2025; many physicists still want more proof |
| Amazon (AWS) | Superconducting “cat” qubits | Ocelot | Chip built to cut error-correction overhead by up to 90% (2025) |
| QuEra · Atom Computing · Pasqal | Neutral atoms | Atom arrays | Hundreds to thousands of atoms held by lasers; partners with Google, Microsoft and others |
| PsiQuantum · Xanadu | Photons (light) | Photonic chips | Betting on standard chip factories to reach a million qubits |
| NVIDIA | Not a qubit maker | CUDA-Q, NVQLink | Links AI supercomputers to quantum chips from 17+ builders |
Fall 2026 check-in
Where we are right now
Real progress, real limits. Quantum computers now beat supercomputers on carefully chosen science tasks, but none yet does everyday useful work better than classical machines.
Biggest arrays
6,100
atom qubits trapped by Caltech in 2025. Most commercial chips use about 100 to 1,000.
Error-protected qubits
70
logical qubits used in IBM and UChicago’s verified advantage run in July 2026.
Speed on a science task
13,000×
faster than the Frontier supercomputer for Google’s Quantum Echoes algorithm (2025).
Still needed
~1M
physical qubits in some estimates for breaking today’s RSA encryption on superconducting chips. Other theoretical designs propose tens of thousands under demanding assumptions, and drug-design needs depend on the calculation.
Google’s six-step roadmap · how far along are we?
Beyond classical
Beat a supercomputer on one task.
Error correction prototype
More qubits, fewer errors (Willow).
Long-lived logical qubit
About 1 error in a million steps.
Logical gate
Reliable operations between protected qubits.
Engineering scale-up
Around 100 logical qubits.
Large error-corrected computer
About a million physical qubits.
Better together
How AI and quantum help each other
AI and quantum computing are not rivals. Right now AI is mostly helping quantum computers work. Later, quantum computers may hand AI something it badly needs: accurate data about how molecules and materials behave.
Direction 1 · happening now
AI → quantum
- Fixing errors in real time. Google DeepMind’s AlphaQubit learned to spot qubit errors better than hand-built methods (2024). AlphaQubit 2 is fast enough to keep up with a live chip. arXiv
- Shrinking quantum programs. AlphaTensor-Quantum, from DeepMind, finds shorter circuits by cutting the most expensive gates. Nature MI
- Tuning the hardware. Machine learning calibrates hundreds of qubits and finds the quietest settings far faster than people can.
- Hybrid supercomputers. NVIDIA’s NVQLink wires GPUs straight to quantum chips so AI can run beside the qubits. NVIDIA
Direction 2 · coming next
Quantum → AI
- New training data. AI models for chemistry learn from simulations. Quantum computers can produce data about molecules that classical computers cannot calculate.
- Checking AI’s guesses. An AI might propose a new battery material; a quantum computer could test how its electrons really behave.
- Quantum machine learning. Researchers are testing quantum versions of AI models. So far there is no proof they beat normal AI on real tasks, so treat big claims carefully.
- Faster sampling and optimization. Some AI steps involve sampling from huge probability spaces, a natural job for qubits. Still early research.
Interactive · be the decoder
Can you fix the qubits faster than the noise?
Error correction stores one bit of information across several qubits. Noise flips some of them. Your job, like AlphaQubit’s, is to guess what the original was. Tip: go with the majority.
Seven qubits, one secret bit. Make your call.
Why “below threshold” matters
Spotlight · Google Quantum AI
What Google’s team has cracked
Google has been at the front of the race since 2014. These are scientific firsts, the hard problems the whole field had to solve before quantum computers can do useful everyday work.
Oct 2019Sycamore
Beat a supercomputer for the first time
A 53-qubit chip sampled random circuits in about 200 seconds, a task Google estimated would take 10,000 years on a supercomputer. Better classical methods later narrowed the gap.
Why it mattersProved a programmable quantum chip could outrun classical machines at something.
Nov 2024AlphaQubit
AI learned to read quantum errors
Google DeepMind and Google Quantum AI trained a neural network on Sycamore data. It made 6% fewer mistakes than the most accurate (but very slow) existing method, and 30% fewer than a fast one built to scale. Google
Why it mattersThe clearest example yet of AI directly improving a quantum computer.
Dec 2024Willow
Solved a 30-year-old error-correction problem
Google’s 105-qubit Willow chip grew its protected qubit from a 3×3 grid to 5×5 to 7×7, and the error rate halved each time. That is “below threshold,” a goal set when error correction was invented in 1995. Willow also ran a benchmark in under five minutes that would take a top supercomputer about 10²⁵ years. Google
Why it mattersShows that bigger quantum computers can be more reliable, not less.
May 2025RSA estimate
Cut the qubits needed to break RSA by about 20×
Google researcher Craig Gidney estimated that 2048-bit RSA encryption could be broken with under a million noisy qubits in less than a week, down from about 20 million in 2019. arXiv
Why it mattersA wake-up call to move the internet to quantum-safe encryption.
Oct 2025Quantum Echoes
The first verifiable quantum advantage
Willow ran an “echo” algorithm, scrambling qubits forward, nudging one, then running time backward, 13,000 times faster than the Frontier supercomputer. Another quantum computer can check the result. That benchmark result was published on the cover of Nature. In a separate preprint with UC Berkeley, the team used related methods to refine models of two real molecules, a task still within reach of classical computers. Google Research
Why it mattersA benchmark advantage that another quantum computer can check, plus an early, still classically reachable step toward chemistry uses.
Oct 2025Nobel Prize
Physics Nobel for the ideas behind its chips
Michel Devoret, chief scientist for quantum hardware at Google Quantum AI, and John Martinis, who led Google’s hardware team until 2020, shared the prize with John Clarke for 1980s experiments on quantum effects in circuits. Nobel Prize
Why it mattersThe physics inside Google’s and IBM’s chips is now officially Nobel-level science.
Dec 2025 · rev. Mar 2026AlphaQubit 2
AI error correction that keeps pace
The December 2025 preprint showed the new decoder correcting surface-code errors in under a microsecond per cycle for larger codes. A March 2026 revision added a first: decoding a type of “color code” in real time at that scale. arXiv
Why it mattersReal machines need decoders that are both smart and fast. This is both.
Mar 2026Two new moves
Neutral atoms and a crypto warning
Google added neutral-atom quantum computers alongside its superconducting chips, led by physicist Adam Kaufman in Boulder, Colorado. The same month it estimated that the encryption used by cryptocurrencies could fall to under 500,000 physical qubits, and urged everyone to start upgrading now. The Quantum Insider
Why it mattersTwo paths to a useful machine, and a clearer deadline for the rest of us.
Honest note: none of these has yet cured a disease or built a better battery. They are the foundations those future results depend on.
Why anyone cares
Big problems quantum + AI could help solve
Nature runs on quantum rules, so the biggest wins are in understanding molecules and materials. Labels show how close each one is.
Medicine
Designing new drugs
Quantum computers may improve selected chemistry calculations about how a drug molecule fits a protein. AI narrows the options; classical methods, lab experiments and clinical testing remain essential.
Early experimentsFood
Cleaner fertilizer
Making fertilizer uses roughly 1 to 2% of the world’s energy. Bacteria do it at room temperature with an enzyme too complex for today’s computers to model.
A top targetEnergy
Better batteries
Find materials that store more energy and charge faster, for electric cars and the power grid.
Early experimentsClimate
Capturing carbon
Design catalysts and materials that pull CO₂ from the air or turn it into fuel.
Years awayMaterials
Room-temperature superconductors
Understand materials that carry electricity with zero loss, which could transform power lines and MRI machines.
Years awaySecurity
Safer encryption
A big enough quantum computer could break today’s internet locks. The fix, quantum-safe encryption, already exists and is rolling out now.
Act nowWhat to expect
The road to useful quantum computers
These are companies’ own targets. Roadmaps slip, so read them as goals, not promises.
- 2026–27
The “quantum advantage” era
More verified wins on science problems in chemistry and physics. Hybrid systems pair quantum chips with AI supercomputers.
- 2029
First fault-tolerant machines
IBM plans Starling: 200 logical qubits running 100 million operations. Quantinuum targets Apollo. Google aims for a commercially useful machine by the end of the decade.
- 2030–35
Encryption switch-over
NIST’s draft transition plan (still a draft) proposes deprecating some quantum-vulnerable algorithms in its standards after 2030 and removing them by 2035. Those dates apply to particular standards, not as one universal deadline, and upgrades are not automatic: organizations need to plan and test their migrations early.
- 2033+
Large-scale quantum computing
IBM’s Blue Jay targets 2,000 logical qubits and a billion operations. Machines like this could tackle real drug and materials design.
What it will do
Specialist super-tool
Accessed through the cloud, working beside AI and normal supercomputers on chemistry, materials, physics and certain math problems.
What it won’t do
Replace your laptop or phone
No quantum phones, no quantum video games, and no instant answers to every problem. Most computing will stay classical, forever.
Plain-language glossary
Words you will keep hearing
Type to filter. Every definition is one or two sentences, no math.
Check yourself
Myth or fact?
Six quick calls. Good for a class warm-up or dinner-table debate.
Score: 0 / 6
Sources for this page
- Google: Meet Willow, our state-of-the-art quantum chip (Dec 2024)
- Google Research: A verifiable quantum advantage (Oct 2025)
- Google: AlphaQubit tackles one of quantum computing’s biggest challenges (Nov 2024)
- A scalable and real-time neural decoder for topological quantum codes (AlphaQubit 2)
- Nature Machine Intelligence: Quantum circuit optimization with AlphaTensor
- Gidney: How to factor 2048-bit RSA integers with less than a million noisy qubits (2025)
- The Quantum Insider: Google adds neutral atom systems (Mar 2026)
- The Quantum Insider: Google on quantum attacks on cryptocurrency encryption (Mar 2026)
- Nobel Prize in Physics 2025 press release
- IBM and UChicago demonstrate quantum advantage on logical circuits (July 2026)
- IBM quantum roadmap
- Quantinuum Helios
- IonQ 99.99% two-qubit gate performance (Oct 2025)
- Caltech 6,100-qubit array (Sept 2025)
- Microsoft Majorana 1 · IEEE Spectrum on the debate
- NVIDIA NVQLink
Put curiosity to work
Read it. Try it. Question it.
Explore new AI research, or take a paper-first investigation into your classroom.