Quantum Computing 2026-2126: What a Reasonable Century of Progress Could Look Like

Making the Middle Concrete

Quantum computing is usually discussed in one of two ways: as a near-term breakthrough that is always a few years away, or as a distant technology that will eventually become powerful without much explanation of what has to happen between today's machines and that future.

I wanted to make the middle concrete.

This project is a 100-year scenario forecast for quantum computing from 2026 to 2126. It starts from the experimental state of the field today and follows what I think a physically reasonable sequence of transitions could look like: noisy processors, reliable error correction, early fault-tolerant systems, commercially useful quantum acceleration, modular quantum-HPC, and eventually distributed quantum infrastructure.

The important word is scenario. Everything after the present-day evidence is a forecast, not a claim that the future has already been determined.

Why I Wrote This

The idea came after reading AI 2027, the forecasting scenario produced by the AI Futures Project.

What interested me most about AI 2027 was not whether every date would turn out to be correct. It was the decision to stop talking about the future in vague terms and instead ask, repeatedly, what happens next. The scenario forces concrete assumptions into the open so that they can later be evaluated, challenged, or falsified.

What would the same style of long-range thinking look like for quantum computing?

Quantum-computing roadmaps and forecasts already exist. Hardware companies publish roadmaps, researchers estimate algorithmic resources, governments plan post-quantum security transitions, and the field has decades of theoretical work describing what scalable quantum computation requires.

But I did not want to take the rapid progress we are currently experiencing in AI and simply project that feeling onto quantum hardware.

AI can improve through unusually fast feedback loops. Better models can write code, improve software, assist researchers, optimize data pipelines, use larger compute clusters, and increasingly contribute to the work required to build the next generation of AI systems. That can create a strong sense of algorithmic compounding.

Quantum computing is different. A quantum processor has to obey the physics of a real device. Qubits decohere. Gates have finite error. Control signals create crosstalk. Superconducting systems require extreme cryogenics. Neutral atoms require large-scale optical control. Trapped ions trade extraordinary fidelity for difficult transport and laser scaling. Photonic architectures live or die by optical-loss budgets. Error correction can turn one useful logical qubit into hundreds or thousands of physical components, and the decoder has to keep up in real time.

If quantum computing continues to improve, but we force the timeline to respect physical error correction, manufacturing, control, networking, energy, and industrial deployment, what does a reasonable century of progress look like?

The result is deliberately slower and more phase-based than an AI-style exponential extrapolation. That does not mean quantum progress has to be slow. A breakthrough in error correction, physical fidelity, modular interconnects, control, or algorithms could pull important milestones forward by years. It means that the forecast should expose those dependencies rather than assuming a smooth exponential curve.

AI 2027 is the inspiration for the forecasting format, not the source of the quantum predictions in this paper. This is my separate scenario synthesis, and there is no affiliation with or endorsement from the AI 2027 authors or the AI Futures Project.

What Matters More Than Physical Qubit Count

The easiest number to market in quantum computing is physical qubit count. It is also becoming one of the least useful numbers by itself.

A machine with a million physical qubits is not automatically more useful than a smaller machine if its error rates, logical clock speed, decoder, or error-correction architecture prevent it from running deep algorithms.

For a useful fault-tolerant system, the full stack has to work:

1. Physical qubits and native gates
2. Repeated syndrome extraction
3. Real-time decoding
4. Logical qubits with errors below the physical layer
5. Universal logical operations, including expensive non-Clifford resources
6. Enough reliable logical depth to execute a problem that matters

This is why the paper tracks not only qubits but also physical-to-logical overhead, logical error, non-Clifford throughput, decoder latency, modular interconnects, and the cost per completed algorithmic result.

Why I Did Not Make the Curve Exponential

It is tempting to look at the last few years of AI progress and expect every frontier technology to follow the same shape. I do not think that is a good default assumption.

AI development is heavily driven by software, algorithms, compute scaling, data, and increasingly automated research workflows. Quantum computing has all of those layers too, but its progress is additionally gated by physical fidelity.

A useful quantum computer must repeatedly remove entropy faster than errors accumulate. That means the system has to improve across multiple coupled constraints:

• Physical gate error
• Leakage and correlated errors
• Coherence and measurement fidelity
• Decoder latency
• Error-correction overhead
• Non-Clifford resource production
• Cryogenic or optical control
• Fabrication yield
• Inter-module entanglement
• Networking and cost

Some of these can improve very quickly. Some cannot. The result may look less like one exponential curve and more like a series of technological phase transitions separated by bottlenecks. That is the structure of this forecast.

Security Probably Moves Before the Economic Revolution

One of the strangest things about quantum computing is that its security impact begins before the machine is powerful enough to break cryptography. The reason is harvest-now-decrypt-later.

An adversary can store encrypted information today and attempt to decrypt it years later. If the information remains sensitive long enough, future quantum capability changes the value of collecting it now. That is why the transition to post-quantum cryptography is already underway.

The paper follows the migration from NIST-standardized post-quantum algorithms, through legacy-system replacement, toward a later world where post-quantum cryptography is ordinary infrastructure and specialized quantum networks are used for things classical cryptography alone cannot provide.

The paper's central CRQC scenario is roughly 2038-2045, with an intentionally wide uncertainty range. That is not a government prediction or a claim of certainty. It is a scenario estimate tied to logical throughput, error correction, and algorithmic-resource assumptions.

The Industrial Prize Is Probably Not Generic QML

The most compelling long-run quantum applications are not necessarily the ones that receive the easiest headlines. I am skeptical of the idea that quantum machine learning simply replaces GPUs for ordinary classical datasets.

The deeper opportunity is where the underlying problem is quantum mechanical or where structured quantum algorithms produce a real end-to-end advantage. That puts strongly correlated chemistry and materials near the center of the forecast.

• Catalyst active sites
• FeMoco and nitrogen fixation
• Battery electrolyte and interface chemistry
• Polymer reaction pathways
• Semiconductor defects
• Superconducting materials
• Carbon-capture chemistry
• Selected rare-event finance and Monte Carlo workloads
• Selected structured nonlinear differential-equation problems
• Quantum-native data and sensing

The QPU does not replace the whole simulation pipeline. It solves the electronically hard part and hands results back to classical multiscale models. This hybrid loop is more important to the long-term forecast than any single quantum algorithm.

What Could Make This Happen Sooner?

The base case is intentionally not the fastest plausible scenario. Several developments could compress the first half of the timeline:

• Practical high-rate qLDPC or other lower-overhead error-correcting codes
• Physical gate errors falling enough to reduce required code distance
• Faster real-time decoders
• Large improvements in magic-state production
• Fault-tolerant modular interconnects
• AI-assisted calibration, code discovery, compiler optimization, materials development, and quantum-algorithm research
• A hardware modality achieving unexpectedly good manufacturing scale and fidelity
• Major algorithmic resource reductions for important chemistry or cryptographic workloads

This is why I treat the five eras as scenario boundaries rather than appointments on a calendar. The first commercially important FTQC systems could arrive earlier than the base case if several bottlenecks break at once.

What Could Make It Take Much Longer?

The opposite case is equally important. Quantum progress can stall if:

• Correlated errors create hard logical error floors
• Fabrication yield does not scale
• Cryogenic or optical control becomes prohibitively expensive
• Logical clock rates remain too slow
• Error correction consumes too much hardware
• Modular links cannot reach adequate entanglement rate and fidelity
• Classical algorithms improve faster than expected
• Useful quantum workloads remain too narrow to justify infrastructure costs

Does the cost per verified quantum algorithmic result fall faster than the classical alternatives improve?

A quantum computer does not have to become cheap in an absolute sense. If one calculation creates billions of dollars of value in chemistry, materials, security, or another domain, an expensive machine can still make economic sense. But raw qubit-count headlines are not enough.

What the Forecast Is Actually Claiming

This forecast is not claiming that quantum computers solve everything, that these dates are certain, or that the ranges represent scientific consensus. It is a technology scenario that asks which transitions would have to occur for quantum computing to become useful infrastructure over a century.

The most interesting endpoint is a change in how we discover useful matter. Today a significant part of materials and chemical progress comes from incomplete models, approximation, intuition, screening, and physical experimentation.

A mature quantum-classical scientific stack could reduce the amount of uncertainty that comes from not being able to accurately calculate the relevant quantum mechanics. That means some scarcity shifts from not knowing which material can do something toward knowing what should work and needing the energy, feedstock, fabrication, logistics, and institutions to make it.

The deepest long-run effect in this scenario is not the elimination of scarcity. It is that ignorance becomes a smaller component of material scarcity.

What Is in the Full Paper

This post is the readable companion to the full IEEE-style research forecast. The paper contains:

• The complete five-era quantitative scenario
• The logical-qubit and physical-to-logical overhead forecast
• Surface-code and qLDPC discussion
• Hardware tradeoffs across superconducting, neutral-atom, trapped-ion, and photonic systems
• NIST post-quantum cryptography migration and harvest-now-decrypt-later analysis
• The CRQC scenario and FeMoco resource estimates
• Finance and nonlinear-differential-equation discussion
• Quantum-network infrastructure
• Cross-era bottlenecks and falsification tests
• Deep-tech capital and industrial implications
• Thirty references and an explicit AI-use disclosure

Sources and Context

The full empirical bibliography is contained in the downloadable paper. The scenario-format inspiration discussed in this article is AI 2027 by the AI Futures Project.

AI 2027 is cited as inspiration for making future assumptions concrete. The quantum-computing timeline is my separate scenario synthesis and does not come from AI 2027.