How quantum computing could transform IT: what changes, and when

What a quantum computer does differently

A classical bit is 0 or 1. A qubit can be in a superposition of both, and several qubits can be entangled, so that their state can only be described together. A quantum algorithm uses entanglement and interference to steer that combined state toward the answer. Measuring it returns one ordinary result, so a program runs many times, each run called a shot, and the output is a count of results rather than a single number.

Describing a quantum system on a classical computer gets exponentially harder as the system grows. Microsoft's documentation gives the example of particles that can sit in 40 locations: storing their quantum state takes 2⁴⁰ numbers, more than 130 GB, and every added location doubles that. A machine built from quantum parts represents such states natively. That is why simulating nature, meaning molecules, materials and physics, is the application with the firmest theory behind it.

Two limits shape everything else in this guide:

• Speedups exist only for specific algorithms. A quantum computer is not a faster general-purpose processor. Each useful speedup comes from an algorithm built for one kind of problem.
• Today's qubits are noisy. In 2018 John Preskill named the current stage NISQ, noisy intermediate-scale quantum: machines whose gate errors limit how large a circuit they can run reliably. The fix is error correction, which combines many physical qubits into one logical qubit; Google describes each logical qubit as made of hundreds of physical qubits. A machine that runs long algorithms on logical qubits is called a fault-tolerant quantum computer. In December 2024 Google showed its Willow chip halving the logical error rate each time the code grew, which it called the most convincing prototype of a scalable logical qubit to date.

Which problems quantum computers are expected to help with

A useful filter comes from a 2023 analysis by Hoefler, Haener and Troyer in Communications of the ACM. A future error-corrected quantum computer will run each operation far more slowly than a classical chip, so they argue a problem needs two properties before quantum hardware can win. It needs a speedup better than quadratic, because a Grover-style square-root saving is swallowed by the overhead, and it needs small inputs, because loading classical data into qubits is slow. Their summary: quantum computers will be practical for big compute on small data, not for big data. Their shortlist puts chemistry and materials science first, along with breaking public-key cryptography with Shor's algorithm.

Chemistry and materials. Molecules are quantum systems, so this is the natural fit. The evidence, though, is more cautious than the marketing. A 2023 study in Nature Communications examined ground-state energy estimation, the most common task in quantum chemistry, and concluded that evidence for an exponential quantum advantage across chemical space has yet to be found, while polynomial speedups may still make quantum computers useful there. In October 2025 Google ran a proof-of-principle experiment on two molecules of 15 and 28 atoms whose results matched traditional NMR measurements: a direction for drug and materials research, separate from the benchmark on which Google claimed its speed advantage.

Optimization. This is the area where the evidence is least settled. Google's quantum team wrote in November 2025 that whether quantum machines can succeed on optimization problems where classical ones fall short remains largely open. Its new algorithm, Decoded Quantum Interferometry, published in Nature, offers a speedup for specific structured problems on a sufficiently large error-corrected machine. IBM noted in August 2026 that optimization was not a focus of its July advantage demonstrations, and it tracks candidates in an open benchmarking library, QOBLIB.

Machine learning and simulation of fluids. Hoefler and colleagues list many current machine learning training approaches, Grover-based drug design, turbulent fluid dynamics, and weather and climate simulation among the areas unlikely to see a quantum advantage with current algorithms in the foreseeable future.

Quantum advantage claims, and how they held up

Quantum advantage means a quantum computer completing a task beyond what classical computing can do alone, with a result that can be validated, which is how IBM defined it in July 2026. Every major claim so far has been on a scientific benchmark, and classical algorithms have kept improving in response:

The pattern matters more than any single row. Benchmarks are chosen to be hard for classical computers, classical researchers then look for shortcuts, and the frontier moves, as Google itself expected when it announced Willow. IBM's own post makes the same point: announcing an advantage opens a result to scrutiny rather than closing the case. None of these results is a business workload. For an IT leader, an advantage headline is a milestone for the field, not a reason to buy anything.

What quantum computers will not replace

Most of an IT estate moves and stores data: web and API traffic, transactions, databases, analytics, file storage, and AI training and inference. That is exactly what quantum computers are worst at. Hoefler and colleagues compared one classical chip of today with an optimistic future quantum computer of 10,000 error-corrected logical qubits. The quantum machine's input and output bandwidth came out at about 1 Gbit/s against 10,000 Gbit/s for the classical chip, and its 16-bit floating-point throughput at 10.5 thousand operations per second against 195 trillion. Their conclusion is that any problem limited by access to classical data, such as searching a database, will be solved faster by classical computers.

Three consequences for planning:

• It is an accelerator, not a server. In hybrid quantum computing, a classical computer and a quantum computer work together on one problem: the classical side prepares jobs, sends them to the quantum processor and processes the results. Some qubit technologies must be kept near absolute zero, so most organizations reach the hardware through the cloud; IBM also offers a dedicated on-premises system, priced by quote.
• Nothing in a typical estate moves to it. The workloads that fit are research computations in chemistry, materials and physics, not the systems IT runs today, so there is no capacity, licence or data center decision to make for quantum today.
• Classical methods keep improving. The tensor-network and Monte Carlo techniques that answered the advantage claims above run on ordinary clusters and GPUs, and they set the bar any quantum pilot has to clear.

Where the hardware is: vendor roadmaps as of September 2026

The dates below are the vendors' own targets, taken from their pages. IBM, for example, states that its roadmap represents only goals and objectives and is subject to change. Treat them as intentions, and weigh each milestone by whether the vendor met its previous ones.

Two caveats before you plan around any of these:

• Microsoft's topological approach carries a caveat from its own paper. The Nature paper published alongside Majorana 1 states that its measurement alone does not distinguish the topological states Microsoft needs from ordinary states that can mimic them.
• There is an independent check. DARPA's Quantum Benchmarking Initiative is testing whether any approach can reach utility scale by 2033, meaning a machine whose computational value exceeds its cost. It verifies vendors' designs rather than forecasting, which makes it a useful neutral yardstick.

How to experiment with quantum computing today

You do not need hardware of your own. A typical experiment is a loop: code on a laptop or in a notebook builds a small circuit, submits it as a task of many shots, receives counts back, and uses a classical optimizer to adjust the next circuit. Amazon Braket runs a managed version of this loop as Hybrid Jobs, with PennyLane, NVIDIA CUDA-Q or its own SDK.

Three cloud services give access to real quantum processors and simulators, billed per use:

Costs climb quickly once error mitigation is on. On Braket, a single IonQ Forte task with error mitigation needs at least 2,500 shots and costs $200.30. On Azure, the minimum charge for one IonQ Forte program with error mitigation on, the default, is about $168.

A sensible first project:

1. Pick a candidate that fits: a small-data, compute-heavy kernel such as a molecule, a material or a small combinatorial core, not a data-heavy workload.
2. Establish the best classical baseline first, with the same care you give the quantum side.
3. Prototype on a simulator. The Braket SDK includes a free local simulator, and Microsoft's Quantum Development Kit is free, open source and needs no Azure account.
4. Run a few small jobs on hardware with a hard budget and a fixed number of shots.
5. Compare honestly with the baseline and record the cost per useful result.
6. Estimate the full-scale version. The Azure Quantum resource estimator calculates the physical resources an algorithm would need on future fault-tolerant hardware, which tells you how far away your use case really is.

Run experiments in a separate sandbox account or subscription with budgets, alerts and least-privilege access, so a loop of hardware tasks cannot run up a bill unnoticed. If your cloud accounts lack that structure, our cloud transformation service builds a governed landing zone, with SSO, least-privilege roles, tagging, budgets and guardrails from the first day.

The one action for every IT team now: post-quantum cryptography

Quantum computing's first effect on most IT teams is defensive. This section is the summary; the full picture, with every deadline, what already runs post-quantum and a step-by-step migration plan, is in our guide to quantum computing in cybersecurity. Shor's algorithm factors integers exponentially faster than the best known classical method, which would let a large fault-tolerant quantum computer break RSA and elliptic-curve cryptography: the public-key algorithms behind TLS, VPNs, SSH, code signing and email encryption.

The estimates of what that takes keep falling. In 2019 Craig Gidney and Martin Ekerå estimated 20 million noisy qubits running for eight hours to factor a 2048-bit RSA key; in a 2025 preprint Gidney cut that to under a million noisy qubits running for under a week. In March 2026 Google estimated that 256-bit elliptic-curve keys could be broken with fewer than 500,000 physical qubits in a few minutes, about 20 times fewer physical qubits than before. IBM's, Google's and Quantinuum's flagship processors have 98 to 120 physical qubits, and neutral-atom arrays of about 10,000 qubits have yet to run deep circuits. NIST describes a machine that can break today's public-key cryptography as possibly years or decades away.

The problem is timing. In a harvest now, decrypt later attack, adversaries collect encrypted data now to decrypt it once the hardware exists, so anything that must stay confidential for years is already exposed. NIST's draft transition plan frames it as simple arithmetic: if data must stay secret for X years and migration takes Y years, you must start before X plus Y exceeds the years until such a machine exists. Canada's Cyber Centre tells federal departments to make systems that carry confidential data over public networks a high priority for migration.

The standards are ready. In August 2024 NIST published FIPS 203, 204 and 205: ML-KEM for key establishment, ML-DSA for digital signatures, and SLH-DSA as a hash-based signature backup, and it urged administrators to start integrating them immediately. In March 2025 NIST selected HQC as a backup to ML-KEM, with a final standard expected in 2027. As of its August 2026 project update, the HQC and Falcon-based signature standards were still in progress.

Quantum key distribution (QKD), sometimes presented as an alternative, does not replace this work. It needs specialized hardware and does not authenticate the parties, and the UK NCSC will not support it for government or military use, naming post-quantum cryptography as the best mitigation.

The deadlines are set too:

What to do now, in order (each step is worked through in the migration plan):

1. Inventory your cryptography: where RSA and elliptic-curve keys protect TLS endpoints, VPNs, SSH, code signing, certificates and PKI, databases, hardware security modules and vendor products.
2. Rank systems by how long their data must stay secret and by exposure to recording, starting with confidential traffic over public networks.
3. Ask vendors for their PQC roadmaps and write post-quantum support into procurement.
4. Turn on hybrid key exchange where your stack supports it. OpenSSL 3.5, released in April 2025, supports ML-KEM, ML-DSA and SLH-DSA and offers the hybrid X25519MLKEM768 key share by default.
5. Build crypto agility: automate how keys and certificates are replaced, the same discipline our guide to renewing an SSL certificate covers, so that changing an algorithm becomes a configuration change rather than a project.

The internet-facing part of the inventory is the easiest place to begin. Our security scanning service discovers your domains, subdomains, IP ranges and certificates and checks exposed services and their TLS configuration on a schedule.

A realistic quantum timeline for IT teams

Standards bodies and governments set the firm dates; hardware dates are vendor goals. This table uses only the sourced milestones above.

Plan against the first set of dates and track the second. A vendor milestone that slips moves your quantum pilot; it does not move your cryptography deadline.