Latest Breakthroughs in Quantum Computing 2024: Progress, Challenges and Reality

I’ve been reading quantum hardware papers on and off since 2019, mostly because I kept getting asked the same question by clients: “Should we be worried about this yet?” For five years my answer was some version of “not really, but keep watching.” Then December 2024 happened, and I had to rewrite my standard answer.

Here’s the honest framing. The latest breakthroughs in quantum computing 2024 delivered weren’t about qubit counts. They were about qubits that stop lying to you. That sounds boring until you realise error rates have been the single thing blocking this entire field for thirty years. In 2024, three separate teams, on three completely different hardware platforms, cracked pieces of it at the same time.

This article covers what actually shipped, what the numbers really mean when you strip out the press-release language, and where quantum computing still falls flat on its face. I’m not going to tell you the future is exciting. I’m going to tell you what changed.

Quantum Computing 2024 in Simple Terms

A regular bit is a 0 or a 1. A qubit can hold a weighted blend of both, and when you entangle a group of them, the machine explores an exponentially large space of possibilities in one go. That’s the sales pitch, and it’s true for a small number of problems.

The catch nobody puts on the slide: qubits are absurdly fragile. Heat, stray magnetic fields, vibration, even cosmic rays knock them out of their state. That’s noise, and it’s why we’ve spent the last decade in what the field calls the noisy intermediate-scale quantum (NISQ) era: machines with enough qubits to be interesting and too much noise to be useful.

The fix has always been quantum error correction: spread one unit of information across many physical qubits so the group can detect and repair errors, producing one reliable logical qubit. Think RAID for quantum states. The problem is the exchange rate. Historically, encoding one decent logical qubit cost you around a thousand physical ones, and adding more physical qubits made things worse, not better, because each new qubit brings its own noise.

That exchange rate is what shifted in 2024. Keep that one idea in your head and the rest of this article makes sense.

Breakthrough 1: Error Correction and More Reliable Qubits

If you only read one section, read this one. Everything else in quantum computing progress downstream depends on it.

Google’s Willow Chip: Making Qubits Less Noisy

On 9 December 2024, Google Quantum AI published results in Nature on Willow, a 105-qubit superconducting chip, and the headline was one word: below-threshold.

Here’s what that means in practice. The team encoded a logical qubit into a surface code grid and scaled it up from 3×3 to 5×5 to 7×7 physical qubits, and each time they scaled up, the error rate was cut in half. Errors going down as the system gets bigger. This is the “below threshold” milestone that’s been an open problem since Peter Shor introduced quantum error correction in 1995.

The unglamorous engineering underneath is what impressed me more than the headline. Willow keeps the tunability of the older Sycamore architecture while pushing average qubit lifetimes (T1) from roughly 20 microseconds to about 68 microseconds. Because the system is past the threshold, small hardware improvements get amplified exponentially: gate fidelities roughly twice as good as Sycamore produced encoded error rates around twenty times better.

Google also ran a random circuit sampling benchmark that Willow completed in minutes and that a classical supercomputer would need an incomprehensible span of time to match. I’d treat that number with a shrug. Random circuit sampling is a benchmark designed to be hard for classical machines and useless for anything else. The error correction result is the real news. The RCS number is the poster.

Topological Qubit Breakthrough

This one flew under the radar and I think it’s the most interesting result of the year.

In November 2024, researchers from Quantinuum, Harvard and Caltech demonstrated what they described as the first true topological qubit. The experiments ran on Quantinuum’s H2 ion-trap processor, which runs 56 fully connected qubits with gate fidelities above 99.8%, and the team encoded qubits into qutrits to build a lattice representing the Z₃ toric code, then studied parafermion and charge-conjugation defects.

Plain English: instead of correcting errors with software after they happen, topological quantum computing stores information in a geometric property of the system that noise can’t easily disturb. Braid the quasiparticles and the information rides along in the shape of the braid. This builds directly on the earlier Quantinuum, Harvard and Caltech work creating and braiding non-Abelian anyons on H2, and it offers an alternative route to fault tolerance that could reduce the need for resource-hungry magic state distillation.

Is it going to replace the surface code next year? No. It’s a physics demonstration on a machine with 56 qubits. But if the topological approach scales, the qubit overhead problem gets dramatically cheaper, and that changes the timeline for everything.

Efficient Error Correction

Meanwhile IBM took a completely different swing at the same problem: don’t make better qubits, make better codes.

In a March 2024 Nature cover paper, IBM introduced a family of low-density parity-check codes and showed that 12 logical qubits could be preserved for nearly a million syndrome cycles using 288 physical qubits total, assuming a 0.1% physical error rate, where the surface code would have needed close to 3,000 physical qubits for the same performance. The “gross” code packs those 12 logical qubits into 144 data qubits plus 144 syndrome ancillas.

Roughly a tenfold reduction in overhead. That’s not a marginal optimisation. That’s the difference between a machine you can build this decade and one you can’t. The catch, and IBM is upfront about it, is that qLDPC codes need long-range connections between distant qubits, which is a genuinely hard chip design problem on a flat superconducting layout.

I’ll take a position here: the gross code was more consequential than Willow for the actual timeline to a fault-tolerant quantum computer. It just didn’t get a chip name and a launch video.

Error-Corrected Logical Qubits

Then there’s the Microsoft and Quantinuum line of work, which is where quantum error correction stopped being a lab curiosity and started running actual workloads.

In September 2024, Microsoft applied its qubit-virtualization system to Quantinuum’s 56-physical-qubit H2 machine to create and entangle 12 logical qubits, the largest number of entangled logical qubits on record at the time, showing a 22× circuit error rate improvement over the corresponding physical qubits. That was a 3× advance over the four logical qubits the two companies announced back in April. The underlying results showed logical error-rate improvements ranging from 11× to 800× over physical baselines. In a Bell-state preparation, the logical circuit error rate dropped from roughly 0.8% to 0.001%.

Microsoft then did the same trick on Atom Computing’s neutral-atom hardware, entangling 24 logical qubits in a GHZ state and running computation with 28 logical qubits, with Atom Computing hitting 99.6% two-qubit gate fidelity.

Here’s the table I keep coming back to when someone asks me to summarise the year:

Result	Team	Platform	What it proved
Willow, below-threshold surface code	Google Quantum AI	105-qubit superconducting	Error rate halves as the code grows
Gross code (qLDPC)	IBM	Simulation / superconducting roadmap	12 logical qubits from 288 physical, ~10× cheaper
12 entangled logical qubits	Microsoft + Quantinuum	56-qubit H2 trapped-ion	22× circuit error improvement, ran real chemistry
24 to 28 logical qubits	Microsoft + Atom Computing	Neutral-atom (1,200+ physical)	Logical qubits work across hardware types
First topological qubit	Quantinuum, Harvard, Caltech	H2 trapped-ion, qutrits	Alternative path to fault tolerance

Four different modalities. One shared direction of travel. That’s the pattern that made the latest breakthroughs in quantum computing 2024 produced feel different from previous years. It wasn’t one lab getting lucky.

Breakthrough 2: Algorithms and Applications Move Closer to Reality

Hardware progress means nothing if there’s nothing worth running. 2024 was the first year I saw applications work that wasn’t purely aspirational.

Chemistry and Materials

Microsoft and Quantinuum used two logical qubits, integrated with an AI model and cloud HPC, to estimate the ground state energy of the active space of an important catalytic intermediate. It was the first end-to-end chemistry simulation combining reliable logical quantum computation with classical high-performance computing and AI. They got results within chemical accuracy.

Note the architecture there, because it’s the thing people miss. The quantum processor wasn’t doing the whole job. It handled the specific piece classical methods struggle with, and classical infrastructure did everything else. Every credible near-term quantum application looks like this: a hybrid pipeline where the QPU is one specialised component in a mostly classical stack, orchestrated much the same way you’d orchestrate any other accelerator in a well-run MLOps setup.

Pasqal ran a related line of work with neutral-atom processors, studying how water molecules arrange themselves inside protein pockets, a detail that strongly affects drug binding and is genuinely painful to model classically.

AI and Machine Learning

The quantum machine learning field has a credibility problem, and it earned it. Most papers show a tiny circuit matching a classical baseline on a toy dataset and call it promise.

A few 2024 results were better than that. Terra Quantum built a hybrid quantum neural network for classifying liver images using only a handful of qubits inside a federated learning setup, so patient data never left each hospital. The quantum part is small. The privacy architecture is the actual product. High-dimensional feature mapping in quantum neural networks shares core mathematical principles with classical high-dimensional vector representations, as explored in our deep-dive on Image Search Techniques.

My honest read: if you’re building AI systems today, quantum machine learning is not on your roadmap, and anyone telling you otherwise is selling something. The gains available right now are in classical work: predictive analytics, better retrieval pipelines, disciplined AI data engineering. Come back to quantum ML when logical qubit counts are in the hundreds.

Physics, Engineering and Simulation

Simulating quantum systems with a quantum system is the one application where the advantage argument is airtight, which was Feynman’s original point back in 1981. In 2024 that showed up in lattice gauge theory experiments, quantum field theory simulations, and early work on plasma physics relevant to fusion energy.

Engineering is further behind but moving. Rolls-Royce worked with quantum software teams on circuits for computational fluid dynamics in jet engine models, designed and validated in simulation, not run at useful scale on real hardware. Worth knowing the difference when you read the headlines.

Improved Quantum Algorithms

Quieter progress here, mostly in error mitigation techniques, variational quantum eigensolver enhancements, and better compilation. Riverlane does a lot of the unglamorous work on the decoding layer: real-time decoders that identify errors fast enough to correct them mid-computation. Without that layer, below-threshold hardware is just an interesting graph.

Better Developer Tools

You can write and run quantum circuits today from a laptop. Qiskit, Cirq, PennyLane, Q#. Azure Quantum Elements, IBM Quantum Platform, Amazon Braket and Google Cloud all put real hardware behind an API. Quantinuum’s InQuanto chemistry package was integrated with Azure Quantum Elements in 2024.

If you’re curious, spend an afternoon with a two-qubit Bell state in Qiskit. It costs nothing and it’ll immediately teach you why measurement collapses everything and why debugging this stuff is so miserable. You can’t inspect intermediate states without destroying them. Once you’ve felt that, the error correction results above land much harder.

Breakthrough 3: Industry-Scale Chips, Cloud Access and Investment

Stronger Processors and Cloud Platforms

The qubit count race cooled off in 2024, and that’s a healthy sign. McKinsey’s Quantum Technology Monitor put it well: 2024 marked a shift from growing qubits to stabilizing qubits, and that shift signals to mission-critical industries that the technology could become a reliable part of their infrastructure.

Cloud platforms are how anyone outside a national lab actually touches this hardware. You queue a job, it runs on a machine in Colorado or California, results come back. No dilution refrigerator required.

Funding and Market Growth

McKinsey estimated quantum computing revenue at around $4 billion in 2024, projecting growth to as much as $72 billion by 2035, with quantum computing, communication and sensing together reaching up to $97 billion. In 2024, roughly a third of investment in quantum technology came from governments and public institutions, a share that collapsed to 3% by 2025 as private capital took over.

That government-to-private handoff tells you more than any qubit number. Private money moves when the technical risk drops enough to price.

Quantum Processors with Scalable Architectures

The architecture conversation shifted from “one bigger chip” to modularity, linking multiple processors with high-fidelity couplers so they behave as one machine. IBM’s roadmap leans on this because their qLDPC codes need long-range connections that a single flat chip can’t provide. Trapped-ion systems solve connectivity differently: H2’s 56 qubits are fully connected, so any qubit can interact with any other. Different trade-offs, no consensus winner.

New Wiring and Qubit Density Designs

Here’s the mundane problem that kills timelines: wiring. Every superconducting qubit needs control lines running into a cryostat held near absolute zero, and every wire carries heat. IBM’s 1,121-qubit Condor packed over a mile of cryogenic wiring into one fridge. You cannot scale to a million qubits that way.

Neutral atoms sidestep this neatly. Atom traps cost a few thousand dollars each, atoms are identical by physics so there’s no device variability problem, and the array can be reconfigured mid-circuit so connectivity is set by software rather than lithography. Pasqal demonstrated a 506-atom defect-free register in 2024.

Main Quantum Computing Challenges After 2024

Now the part most articles skip. These are the real quantum computing problems, and none of them got solved last year.

Scaling Up to Large Systems

The 2024 results ran on 56 to 105 physical qubits producing a dozen or two logical ones. Useful cryptographic or chemical workloads need thousands of logical qubits, which even at gross-code efficiency means hundreds of thousands to millions of physical qubits. That’s four orders of magnitude. Below-threshold error correction says the path exists. It doesn’t say the path is short.

Noise and Engineering Complexity

Every added qubit needs control electronics, calibration and cooling. Calibration drifts, so machines get recalibrated constantly, and performance today isn’t performance last Tuesday. This is the challenge of quantum computing that never makes headlines because “we spent six months on calibration protocols” isn’t a press release. Rigorous system validation demands rigorous quality assurance frameworks, analogous to the structured testing pipelines discussed in Software Testing Basics.

Limits of Algorithms and Verification

Quantum computers are not faster computers. They’re faster at a narrow class of problems: the ones sitting in the complexity class BQP but not in classical polynomial time. Sorting, databases, web serving, running your ML models: no speedup, ever.

Verification is the sharper issue. If a quantum computer solves something no classical computing system can check, how do you know the answer’s right? Several claimed advantage results have been quietly matched by improved classical algorithms afterwards. Healthy science, awkward marketing.

Security and Encryption

On 13 August 2024, NIST finalised its first three post-quantum encryption standards: FIPS 203 (ML-KEM), FIPS 204 (ML-DSA) and FIPS 205 (SLH-DSA), concluding an eight-year standardisation process that began in 2016. The NSA’s CNSA 2.0 sets 2030 as the mandatory migration deadline for national security systems.

The threat model is “harvest now, decrypt later”: encrypted traffic captured today, stored, and opened once a big enough machine exists. If your data needs to stay private past 2035, this is a today problem, and it belongs in the same risk register as your AI governance work rather than in a someday folder.

Skills, Cost and Access

Cost is brutal. Cryogenic infrastructure, laser systems, specialist staff. Almost nobody should buy a quantum computer; rent time through a cloud platform. And the skills gap is real. You need quantum information theory plus practical software engineering, and that intersection is thin. If you’re a developer weighing whether to invest a year in this, my advice is blunt: learn the concepts well enough to evaluate vendor claims, and put your serious hours into classical AI software development, which will pay off considerably sooner.

Emerging Real-World Use Cases From 2024

Drug Discovery and Health

Protein simulation and drug binding are the clearest early targets, because molecular behaviour is quantum mechanical and classical approximations get expensive fast. The Microsoft and Quantinuum catalytic intermediate calculation and Pasqal’s protein-cavity water work are both narrow, both real. Neither designed a drug.

Materials, Energy and Climate

Batteries, catalysts, superconductors, nitrogen fixation. Same logic: small, hard, quantum-mechanical problems where a modest quantum advantage translates into enormous industrial value. Chemistry and materials science will get there first, and it won’t be close.

Finance, Logistics and Optimization

The most oversold category. Portfolio optimisation and routing demos are everywhere; verified advantage over good classical solvers is essentially nowhere. Classical optimisation has had fifty years of tuning, and it keeps winning. Treat any finance quantum advantage claim as unproven until someone publishes the classical baseline they beat.

AI and Data Analytics

Long-term, quantum methods might help with specific kernels inside larger AI systems. Short-term, the interesting movement is the other direction: AI is improving quantum computing. Google used machine learning in Willow’s error decoding, and Microsoft’s chemistry result had an AI model in the loop. That’s the relationship that’s actually working today, and it’s the kind of hybrid orchestration problem that teams at SoftbrixAI deal with in classical AI agent development right now, minus the dilution refrigerator.

What To Expect After the Breakthroughs of 2024

If I had to bet, and I’d bet with limited confidence: the next few years are about logical qubit counts, not physical ones. Watch that number. It’s the only metric that tracks real quantum computing progress.

Expect the qubit-count press releases to keep coming and to mean progressively less. Expect at least one loudly announced quantum advantage claim to be knocked down by a smarter classical algorithm, which is the field working correctly. Expect chemistry and materials to produce the first genuinely commercial result, probably in a narrow simulation that saves a lab months rather than something you’d recognise as a product.

What should you actually do?

• If you handle long-lived sensitive data: start the post-quantum crypto migration. The standards are finalised, the tooling exists, and the 2030 deadlines are closer than they look.
• If you’re in pharma, chemicals or materials: get one person watching this seriously and run a pilot with a cloud QPU. Not because it’ll work yet, but because the learning curve is long.
• If you’re in finance or logistics: wait. Fix your classical optimisation first.
• If you’re a developer: spend a weekend on Qiskit for the mental model, then go back to work that ships. Better AI consulting and stronger RAG systems will move your numbers this quarter. Quantum won’t. For evaluating retrieval infrastructure, speed, and pricing of real-time web search APIs, check out our evaluation of the Best API Search Company’s Homepage in 2026.

The latest breakthroughs in quantum computing 2024 delivered moved the field from “maybe this is impossible” to “this is an engineering problem.” That’s a genuinely big shift. It is not the same as “this works now,” and the gap between those two statements is where most of the hype lives.

FAQ: Latest Breakthroughs in Quantum Computing 2024

Is quantum computing real in practice?

Real, yes. Practical, not yet. You can log into a cloud platform right now and run a circuit on a real machine in Colorado. What you can’t do is solve a commercially valuable problem faster than a classical computer, with results you can verify. The 2024 error correction work removed the biggest theoretical doubt about whether useful machines are buildable, but “buildable” and “built” are separated by a decade of engineering.

What changed most in 2024?

Error correction crossed from theory into demonstrated fact. Google showed error rates falling as the code grows, IBM cut the qubit overhead roughly tenfold, and Microsoft ran real chemistry on logical qubits across two different hardware types. Before 2024, adding qubits usually made things worse. That relationship inverted, and it inverted on multiple platforms in the same year, which is why it’s credible.

Are quantum computers close to breaking common encryption?

No. Breaking RSA-2048 with Shor’s algorithm needs somewhere in the range of hundreds of thousands to millions of high-quality physical qubits, running for hours. The best 2024 machines had roughly a hundred. That said, “harvest now, decrypt later” makes it a present-tense problem for anything that must stay secret into the 2040s, which is exactly why NIST finalised the post-quantum standards in August 2024.

What are the main challenges now?

Scale, wiring, calibration and verification. Getting from a hundred physical qubits to a million is an engineering slog involving control electronics, cryogenics and heat. Calibration drifts constantly. And verifying answers from a machine that outperforms every classical checker is an unsolved problem in its own right. Cost and the skills shortage make everything slower.

What is a logical qubit?

One reliable qubit built from many unreliable ones. The physical qubits carry redundancy so the system can spot and fix errors without measuring the encoded information directly, because measuring it would destroy it. Historically it took around a thousand physical qubits per logical qubit; IBM’s 2024 gross code got 12 logical qubits out of 288 physical. That exchange rate is the number to watch.

Which areas are likely to benefit first?

Chemistry and materials science, comfortably. Molecules are quantum systems, so simulating them on a quantum machine is the natural fit rather than a forced application: catalyst design, battery chemistry, drug binding. Fundamental physics simulation is close behind. Finance and logistics get the most marketing attention and have the weakest evidence, because classical optimisation is very, very good.

What were the biggest quantum computing breakthroughs in 2024?

Four stand out: Google’s 105-qubit Willow chip demonstrating below-threshold surface code error correction; IBM’s gross code cutting error correction overhead by roughly ten times; Microsoft and Quantinuum entangling 12 logical qubits on the 56-qubit H2 system and running an end-to-end chemistry simulation; and the topological qubit built from qutrits by Quantinuum, Harvard and Caltech. Different platforms, same message: reliability finally started scaling.

Can quantum computers break encryption today?

No, and not for years. The machines that exist have far too few qubits and far too much noise. The realistic risk is retrospective: data stolen today and decrypted later. If that applies to you, migrate to ML-KEM and ML-DSA now rather than waiting for a machine that can do the damage.

---

References

• Quantum error correction below the surface code threshold (Google Research)
• High-threshold and low-overhead fault-tolerant quantum memory (Nature)
• Microsoft and Quantinuum create 12 logical qubits (Microsoft Azure Quantum Blog)
• NIST releases first three finalized post-quantum encryption standards
• McKinsey Quantum Technology Monitor
• First-ever topological qubit demonstration (The Quantum Insider)