The Quantum Computing Mirage: What Three Years of Broken Promises Have Taught Me

I've lost count of how many quantum computing briefings I've sat through where executives project timelines on screens that quietly shift right every six months. The promises sound identical to what I heard in 2022, except the dates change. Quantum advantage was coming in 2024. Then 2025. Now it's 2026. Next year, I'll probably hear 2027.

Google's announcement about their Willow processor in late 2024 followed a script I could recite from memory. A hundred-plus qubits. Performance beyond classical supercomputers. A calculation verified as correct. The press release carefully avoids mentioning that the calculation serves no purpose beyond demonstrating the machine works. It's a benchmark divorced from any application someone would pay to run.

IBM's Nighthawk rollout last year hit similar notes. One hundred twenty qubits with better connectivity, allowing marginally more complex circuits. Expected availability late 2025, with scaling to thousands of qubits by 2028. I asked an IBM quantum researcher off the record what “thousands of qubits” actually enables. The answer, after some hedging: “Still not enough for most things people care about.”

The Error Correction Problem That Keeps Getting Worse

Here's what quantum computing vendors don't emphasize in announcements: physical qubits are catastrophically unreliable. Decoherence times — how long a qubit maintains its quantum state before environmental noise destroys it — are measured in microseconds to milliseconds, depending on implementation. That's not long. Classical bits maintain state indefinitely unless you actively flip them.

Building reliable logical qubits from unreliable physical ones requires quantum error correction codes. Surface codes, the most promising approach, need roughly 1,000 physical qubits to produce one logical qubit at useful fidelity. Some estimates run higher — 1,500 to 2,000 physical qubits per logical qubit, depending on target error rates.

Do the math. If you need 10,000 logical qubits for a useful computation — and many proposed applications require more — you’re looking at 10 million physical qubits. Current systems have a few hundred. IBM's roadmap targets thousands by 2028. Even if they hit that goal, they're three to four orders of magnitude short.

IBM published results last year showing 10× faster decoding of error correction codes. They're optimizing the classical computation required to interpret syndrome measurements and apply corrections. That matters for scalability, but it doesn't change the fundamental overhead ratio. You still need roughly 1,000 physical qubits per logical qubit. Decoding them 10× faster just means you waste less classical computing time managing the error correction.

The Qiskit framework improvements — 24 percent better accuracy on 100-qubit circuits through dynamic circuits and error mitigation — sound significant until you check baseline accuracy. Running a 100-qubit circuit on current hardware without extensive error mitigation gives you garbage results. Improving garbage by 24 percent still gives you mostly garbage. These aren't production systems. They're lab instruments requiring expert supervision.

What Google's Willow Actually Demonstrated

Google's Quantum Echoes algorithm on Willow performed a quantum simulation matching advanced nuclear magnetic resonance simulations of a 15-atom molecular system. The technical execution deserves respect. Simulating quantum systems using quantum hardware is conceptually clean — you’re modeling quantum behavior with a quantum device. Classical computers struggle with this because quantum state spaces grow exponentially with system size.

But a 15-atom simulation doesn't get you to drug discovery. Pharmaceutically interesting molecules — proteins, enzyme binding sites, antibody structures — contain hundreds to thousands of atoms. The computational complexity doesn't scale linearly. Simulating a system twice as large requires far more than twice the quantum resources. The gap between demonstrating a 15-atom system and simulating molecules relevant to drug development is enormous.

I spoke with a computational chemist at a major pharmaceutical company after Google's announcement. They use classical quantum chemistry software — Gaussian, ORCA, Q-Chem — to model molecular interactions daily. These tools run on HPC clusters and handle much larger systems than current quantum hardware can touch. Yes, they use approximations. Yes, certain calculations remain intractable. But classical methods keep improving, and quantum computers aren't close to being competitive for real pharmaceutical workflows.

The classical computing community responds to each quantum achievement by optimizing their algorithms. After Google's 2019 quantum supremacy claim with Sycamore, researchers developed classical algorithms that narrowed the performance gap substantially. The same pattern plays out repeatedly. Quantum researchers announce a hard problem for classical computers. Classical algorithm researchers respond with better approaches. The quantum advantage evaporates or becomes marginal.

Where the Qubit Scaling Curve Actually Goes

Superconducting qubits — the technology IBM and Google primarily use — require dilution refrigerators that cool systems to millikelvin temperatures. Operating temperatures are around 15 millikelvin, colder than outer space. The refrigeration systems are expensive, power-hungry, and physically large. You can't just stack them arbitrarily to add more qubits.

IBM's 300 mm wafer fabrication for quantum processors is genuine engineering progress. Producing superconducting qubit chips on semiconductor manufacturing infrastructure means you can potentially scale production using existing fab technology. The problem isn't making qubits. It's connecting them, controlling them, and reading them out without introducing noise.

Each qubit needs control lines — microwave pulses to manipulate quantum state. It needs readout circuitry to measure state. In superconducting systems, this means physical wiring running from room temperature down to millikelvin temperatures. Current systems use coaxial cables, which is manageable for hundreds of qubits but doesn't scale to millions. Some researchers are developing multiplexing schemes to reduce wiring requirements, but this introduces new error sources.

Connectivity matters critically. Most quantum algorithms assume all-to-all qubit connectivity — any qubit can interact directly with any other. Physical systems don't provide this. Superconducting qubit chips typically arrange qubits in 2D grids where each qubit couples to nearest neighbors. Running an algorithm that requires distant qubits to interact means executing SWAP operations to move quantum state around the chip. Each SWAP introduces errors and consumes gate-depth budget.

IBM's Nighthawk improved connectivity with 218 couplers for 120 qubits, allowing slightly more flexible circuit topologies. This enables 30 percent more complex circuits, which sounds good until you realize the baseline is “extremely limited circuits.” Going from very constrained to somewhat less constrained doesn't fundamentally change what computations are feasible.

The Application Gap Nobody Wants to Quantify

Optimization problems appear frequently in quantum computing pitches. Portfolio optimization for finance. Logistics routing. Manufacturing scheduling. Supply chain management. Quantum annealing and variational quantum algorithms supposedly handle these better than classical approaches.

Except classical optimization has also improved dramatically. Mixed-integer programming solvers are far more capable than a decade ago. Machine learning techniques like reinforcement learning tackle optimization problems quantum algorithms target. Companies deploy these classical methods in production today and get useful results.

D-Wave has sold quantum annealers commercially since 2011. Their systems now have thousands of qubits. After over a decade, clear examples of quantum annealing outperforming classical optimization on real business problems remain scarce. Most published comparisons show quantum annealing competitive with or slightly better than some classical algorithms, but not decisively superior to state-of-the-art classical methods.

The pattern suggests quantum computing might find niche applications where it's marginally better than classical approaches for specific problem instances. That's not worthless, but it's far from the revolutionary impact implied by vendor marketing. Marginal improvements don't justify the infrastructure cost and operational complexity of quantum systems.

Cryptanalysis through Shor's algorithm represents the clearest threat quantum computing poses. It can factor large numbers exponentially faster than classical algorithms, breaking RSA and other public-key cryptosystems. This danger is mathematically proven, not speculative. A fault-tolerant quantum computer with a few thousand logical qubits running Shor's algorithm would compromise much of current encryption infrastructure.

The timeline matters. Building fault-tolerant systems with thousands of logical qubits requires millions of physical qubits with error correction. Generous estimates put this 15 to 20 years out. Conservative estimates say 30+ years, or possibly never. NIST finalized post-quantum cryptographic standards in 2024. Migration to quantum-resistant algorithms is underway. We're hardening systems against a threat that won't materialize for decades, if ever.

What the Talent Shortage Reveals

Quantum computing requires expertise at the intersection of physics, mathematics, and computer science. You need to understand quantum mechanics deeply enough to reason about qubit behavior. You need mathematical sophistication to work with Hilbert spaces and unitary transformations. You need programming skills to implement algorithms.

Universities produce maybe 500 to 1,000 graduates per year with serious quantum computing training globally — a generous estimate. Industry demand far exceeds supply. Companies hire PhD physicists and teach them software engineering, or hire software engineers and teach them quantum mechanics. Neither path is efficient.

If quantum computing were about to become mainstream technology, we'd see massive university program expansion training quantum developers. We'd see bootcamps and online courses producing job-ready quantum programmers. We'd see salary surveys for quantum software engineers reflecting high demand. Some of this exists, but at modest scale. The education pipeline doesn't reflect an imminent technology revolution.

Compare this to classical machine learning. When deep learning took off around 2012, universities couldn't expand programs fast enough. Online courses proliferated. Bootcamps emerged. Within five years, you could hire ML engineers without requiring PhD-level expertise. The ecosystem grew organically in response to genuine demand.

Quantum computing shows no comparable acceleration. It's a specialized research field attracting bright people, but the broader engineering community isn't rushing to acquire quantum skills because there are no jobs requiring them. Quantum computing remains primarily academic research and corporate R&D. The transition to an engineering discipline keeps receding.

Why IBM and Google Keep Building Anyway

Both companies have invested hundreds of millions, probably billions, in quantum computing programs. IBM operates multiple quantum data centers. Google has a dedicated quantum AI lab. Neither can walk away without admitting those investments were premature.

Cloud quantum computing as a service generates minimal revenue. Running experiments on IBM Quantum or Google's quantum processors is interesting for researchers but not a sustainable business model. These offerings exist primarily for ecosystem development — getting people familiar with quantum programming so there's demand when useful hardware eventually arrives.

The corporate motivation isn't immediate ROI. It's positioning for potential future markets and hedging against competitors achieving breakthroughs. If quantum computing does eventually work, being years behind IBM or Google would be catastrophic for competitors. So Microsoft, Amazon, and others maintain quantum programs despite unclear commercial timelines.

Research publications and patent portfolios from quantum programs justify continued investment to boards and shareholders. The programs employ world-class physicists doing legitimate research. Even if practical quantum computers remain decades away, the research produces scientific value and attracts talent.

This creates sustained funding without requiring near-term applications. Classical tech companies are rich enough to run multi-decade research programs. Quantum computing can continue indefinitely at current burn rates without demonstrating utility, similar to corporate research labs of previous eras.

What Changed Between 2019 and 2025

Google's 2019 quantum supremacy announcement generated enormous attention. Nature published the paper. Media coverage went worldwide. The narrative was: quantum computing has arrived; classical computers are obsolete for certain problems.

Within months, IBM published analysis showing classical supercomputers could complete Google's benchmark calculation much faster than Google estimated. Other researchers developed improved classical algorithms. The quantum advantage shrank dramatically under scrutiny.

Since then, quantum announcements have been more cautious. Terms like “quantum advantage” get defined more carefully. Claims get hedged more thoroughly. Vendors emphasize specific benchmarks rather than general computational superiority. The field learned that overpromising invites backlash.

Qubit counts increased steadily. Error rates decreased incrementally. Coherence times improved gradually. None of these advances translated into new applications. The gap between better qubits and useful quantum computers didn't narrow meaningfully.

Classical computing also improved. GPUs got faster. AI accelerators proliferated. Neuromorphic chips emerged. For many problems quantum computing supposedly targets, classical hardware advances provided better near-term solutions. Quantum advantage became a moving target as classical capabilities increased.

The Post-Quantum Cryptography Migration Actually Matters

While quantum computers struggle to do anything useful, cryptographic infrastructure is undergoing massive changes to defend against them. NIST selected lattice-based and hash-based algorithms resistant to both quantum and classical attacks. Government agencies mandate transition timelines. Financial institutions are planning migrations.

This represents real cost and real risk. Changing cryptographic algorithms throughout complex systems takes years and breaks things. Compatibility issues emerge. Performance degrades because post-quantum algorithms are typically slower. Hardware acceleration for new algorithms doesn't exist yet on most platforms.

The threat model is “harvest now, decrypt later” — adversaries capture encrypted data today, store it, and decrypt it once quantum computers arrive. For data with decades-long confidentiality requirements, this matters. You need to transition now even if quantum computers remain 20 years away.

So organizations spend millions upgrading cryptography to defend against machines that don't exist and might not exist for decades. It's insurance — arguably sensible insurance — but the situation is bizarre. Defensive measures against quantum computers have more immediate impact than quantum computers themselves.

Where Technical Depth Actually Lies

Understanding quantum computing requires getting past surface-level explanations. Qubits aren't just “bits that can be 0 and 1 simultaneously.” That's a cartoon. Qubits are quantum two-level systems described by complex probability amplitudes in a Hilbert space. Quantum gates are unitary transformations operating on those state vectors. Measurement collapses superposition to definite states probabilistically according to the Born rule.

Gate fidelity specifications matter. A two-qubit gate with 99.9 percent fidelity sounds good until you realize complex algorithms require thousands or millions of gates. Errors accumulate. At 99.9 percent fidelity, after 1,000 gates you've got roughly a 37 percent chance the computation is correct, assuming independent errors. Error correction overhead is why you need so many physical qubits.

Coherence times set fundamental limits. T1 relaxation time measures how quickly excited states decay to the ground state. T2 dephasing time measures how quickly relative phases between superposition components degrade. Superconducting qubits typically achieve T1 around 100 microseconds and T2 around 50 microseconds. Every operation must complete before coherence is lost. This constrains circuit depth severely.

Connectivity topology determines what algorithms you can run efficiently. Heavy-hex lattice connectivity, which IBM uses, provides each qubit with three connections arranged hexagonally. This is better than simple grid connectivity but still far from all-to-all. Mapping logical circuit topology to physical connectivity requires SWAP insertion, expanding circuit depth and introducing errors.

These technical details separate informed commentary from promotional fluff. When a vendor announces a new processor, specifications that matter include qubit count, gate fidelities (single-qubit and two-qubit), coherence times, connectivity, readout fidelity, and crosstalk. Press releases often omit these or bury them in supplementary materials.

The Next Five Years, Realistically

IBM will probably hit their thousands-of-qubits target by 2028. Google will build larger processors. Coherence times will improve incrementally. Error rates will decrease gradually. None of this will enable practical applications that justify the investment.

Research will continue producing interesting results. Quantum simulation experiments will handle slightly larger systems. Error correction demonstrations will show better performance. Benchmarks will be achieved proving quantum computers can do specific artificial tasks faster than classical systems.

Companies maintaining quantum programs will emphasize long-term positioning and research value. Funding will continue because shutting programs down means admitting massive sunk costs. Cloud quantum services will persist as loss leaders supporting ecosystem development.

Classical computing will advance simultaneously. AI chips will get faster. Novel architectures will emerge. For most problems, classical solutions will remain superior. The opportunity cost of quantum computing—what else could have been done with those resources — will grow harder to justify.

Post-quantum cryptography migration will accelerate, driven by compliance requirements and institutional caution. This might end up being quantum computing's primary impact: forcing cryptographic upgrades through the threat of future capabilities rather than demonstrating actual capabilities.

Universities will continue training quantum researchers. Talented physicists and mathematicians will work on legitimate problems. Scientific progress will occur — just not at the pace or with the impact industry roadmaps suggest.

The honest assessment most researchers would give privately: we're making progress, the physics works, scaling remains extremely hard, practical applications are distant, predictions about timelines are guesses, and anyone promising a quantum revolution by 2026 is either ignorant or lying.

That's not what you'll read in press releases. But that's the reality.