Quantum Computing Hardware Approaches Compared
There's a lot of noise around quantum computing, and most of it comes from people who've never actually characterized a qubit. I've spent enough time adjacent to this field to have strong opinions about which hardware approaches are most promising, and I'll share them here — but I want to be upfront that nobody really knows which technology will win at scale. We're in the vacuum tube era of quantum computing, and the transistor hasn't been invented yet.
Superconducting Qubits: The Current Leader
IBM and Google have pushed superconducting transmon qubits further than any other approach. IBM's Eagle processor (127 qubits, 2021), Osprey (433 qubits, 2022), Condor (1,121 qubits, 2023), and Heron (133 qubits but much higher quality, 2024) represent a sustained investment that's hard to match. Google's Sycamore (53 qubits) famously demonstrated quantum supremacy in 2019, and their Willow chip (105 qubits, late 2024) showed below-threshold quantum error correction for the first time.
Superconducting qubits work by cooling circuits made of aluminum and niobium on silicon substrates to about 15 millikelvin — colder than outer space. At these temperatures, the circuit behaves as a quantum harmonic oscillator, and the Josephson junction (a thin insulating barrier between two superconductors) provides the nonlinearity needed to create a usable two-level quantum system.
The good news:
- Gate fidelities are now above 99.5% for single-qubit gates and above 99% for two-qubit gates on the best devices
- Gate speeds are fast — typically 20-50 nanoseconds for single-qubit gates and 50-300 ns for two-qubit gates
- The fabrication process uses standard semiconductor tools (electron beam lithography, thin film deposition, etching) on silicon wafers
- There's a large and growing ecosystem of control electronics, software tools, and trained engineers
The challenges are equally real:
- Coherence times (T1 and T2) are typically 100-300 microseconds, which limits circuit depth to roughly a few thousand gates before errors accumulate
- The dilution refrigerators required cost $500K-$1M each and have limited cooling power at the millikelvin stage (maybe 10-20 microwatts), constraining how many qubits you can operate
- Qubit connectivity is limited — most architectures use a 2D grid where each qubit connects to 2-4 neighbors, requiring SWAP operations to interact distant qubits
- Frequency crowding and crosstalk between qubits become harder to manage as system size grows
Trapped Ions: The Quality Alternative
IonQ and Quantinuum (formed from Honeywell Quantum Solutions) are the commercial leaders in trapped-ion quantum computing. The approach uses individual atoms (typically ytterbium-171 or barium-137) held in electromagnetic traps, with qubit states encoded in the atomic energy levels.
This connects to the ideas in Verification and Testing: How Billion-Transistor Chips Get V.
Trapped ions have some remarkable properties. Their coherence times can exceed minutes — orders of magnitude better than superconducting qubits. Gate fidelities above 99.9% for single-qubit and above 99.5% for two-qubit have been demonstrated. And every qubit is identical, because they're literally the same species of atom. Quantinuum's H2 processor achieved a quantum volume of 65,536 in 2024, which by that metric is the highest of any quantum computer.
I'd argue that trapped ions produce the highest quality qubits available today. But there's a problem: speed and scale. Two-qubit gates take 100-600 microseconds — a thousand times slower than superconducting gates. Moving ions around in a trap (shuttling) to bring them together for entangling gates adds overhead. Current systems have 20-56 qubits, and scaling to thousands is an open engineering challenge.
Quantinuum's approach uses a "race track" trap topology where ions are shuttled between different zones for computation, storage, and measurement. This allows all-to-all connectivity (any qubit can interact with any other), which is a big architectural advantage over superconducting grids. But the shuttling time eats into the coherence budget.
Photonic Quantum Computing
PsiQuantum and Xanadu represent two different photonic approaches. PsiQuantum is building a fusion-based quantum computer using photonic qubits generated and entangled through linear optical circuits. Their bet is that photonics can be manufactured in existing semiconductor fabs (they're partnered with GlobalFoundries) and that the ability to operate at room temperature (no dilution refrigerators for the photonic parts) gives a path to million-qubit scale.
Xanadu uses squeezed-state photonic qubits in a continuous-variable framework, which is mathematically different from the discrete qubit model. Their Borealis system demonstrated quantum advantage for a specific sampling problem in 2022.
Related reading: 3D NAND Flash: Layer Stacking, String Architecture, and the .
Photonic approaches have inherent advantages: photons don't interact with each other, which means low decoherence, and they travel at the speed of light, enabling fast operations and natural networking capability. The challenge is that photon loss is a persistent problem, and creating the deterministic entanglement needed for computation is difficult with photons that prefer not to interact.
PsiQuantum has raised over $700 million and is building what they claim will be a million-qubit error-corrected system. I'm cautiously skeptical — the technical milestones they need to hit are substantial — but if they succeed, the manufacturing scalability argument is compelling.
Neutral Atoms: The Dark Horse
This is the approach I find most exciting right now. Companies like QuEra Computing (spun out of Harvard and MIT) and Pasqal (based in France) use arrays of individual atoms trapped in optical tweezers — focused laser beams that hold atoms in precise positions.
Neutral-atom systems have some remarkable recent results:
- QuEra demonstrated a 48-logical-qubit error-corrected system in late 2023, using 280 physical qubits — this was a significant milestone for the entire field
- Atom arrays can be reconfigured dynamically, moving atoms with optical tweezers to create different connectivity patterns for different computations
- Rydberg interactions (exciting atoms to high energy states) provide strong, controllable entanglement between neighboring atoms
- Systems with 1,000+ atoms have been demonstrated in academic labs, and the path to 10,000+ seems clearer than for other platforms
The downsides? Two-qubit gate fidelities are currently around 99.5% — good but not matching the best trapped-ion results. Mid-circuit measurement and feed-forward (needed for many error correction schemes) is still being developed. And the optical systems for controlling hundreds of individual atoms are complex.
For a related perspective, see CXL Memory Expansion: Disaggregated Memory Pools and Compute.
Topological Qubits: Microsoft's Bet
Microsoft has spent over a decade pursuing topological qubits based on Majorana zero modes in semiconductor-superconductor nanowires. The theory is beautiful: topological qubits would be inherently protected from local noise, dramatically reducing the overhead for error correction.
After a retracted paper in 2021 and years of skepticism, Microsoft announced in 2024-2025 that they'd demonstrated the key building blocks: topological gaps in indium arsenide nanowires proximitized with aluminum, with evidence of Majorana signatures. They've integrated these into a small qubit device.
I'll be honest — I'm not yet convinced they've solved the fundamental physics questions, and they're years behind other platforms in terms of working qubit count. But if topological qubits work as theorized, the error correction overhead drops so dramatically that you'd need far fewer physical qubits to build a useful machine. It's a high-risk, high-reward bet that only a company with Microsoft's resources and patience could sustain.
My Assessment
If I had to rank these approaches by likelihood of delivering the first genuinely useful (not just demonstrative) quantum computer:
- Superconducting qubits — the ecosystem is deepest, the engineering is most mature, and both IBM and Google are executing steadily. Error correction is being demonstrated.
- Neutral atoms — the scaling path looks promising, and the recent error correction results from QuEra are genuinely impressive. I think this is the technology most likely to surprise people in the next 3-5 years.
- Trapped ions — incredible qubit quality, but the speed and scaling challenges are real. Quantinuum's work is excellent, but I worry about the path to thousands of qubits.
- Photonics — if PsiQuantum delivers, it could leapfrog everyone. Big if.
- Topological — the potential is huge but the timeline is the longest.
The honest answer is that we probably need error-corrected machines with thousands of logical qubits to solve problems that classical computers truly can't. We're at maybe 10-50 noisy logical qubits today. The gap is still large, but it's shrinking faster than many expected five years ago. I'd estimate 8-15 years before quantum computers are solving real commercial problems that classical machines can't touch — but I've been wrong before, and breakthroughs in error correction could compress that timeline.