Quantum Computing and Data Science: What May Change, and What Already Needs Attention

Quantum computing attracts two kinds of writing: breathless predictions and blanket dismissal. Neither helps someone who works with data or financial models and needs to know what, if anything, to plan for. This post separates three questions: what the hardware can do today, which data and finance workloads researchers think it could eventually help, and where the one concrete risk, to encryption, already needs attention.

I am not a quantum researcher. I have tried to stick to claims I can point to a primary source for, and to say plainly where the evidence is thin.

Where the hardware stands

In 2018 John Preskill coined the term NISQ, for noisy intermediate-scale quantum devices. His assessment was that machines with 50 to 100 qubits might do some tasks beyond classical computers, but that noise in the gates limits how large a circuit can run reliably, and that such a device “will not change the world right away” (Preskill, 2018).

The way out of noise is error correction, where many physical qubits act together as one more reliable logical qubit. A 2024 paper from Google Quantum AI reported a milestone: surface-code memories operating below the error threshold, where adding qubits suppresses logical errors instead of increasing them. Their largest was a 101-qubit distance-7 code, with a logical error of 0.143% per correction cycle (Google Quantum AI, 2024). That is real progress. It is also a memory, a demonstration that the approach works, not a machine running useful calculations.

For scale, one published resource estimate puts breaking 2048-bit RSA at under a million noisy qubits running for under a week, assuming a 0.1% gate error rate (Gidney, 2025). Set that against a 101-qubit experiment and the remaining gap is several orders of magnitude, though the two numbers describe different tasks and the estimates keep falling. In 2019 the same line of work had estimated 20 million qubits for an eight-hour run (Gidney and Ekerå, 2019).

Optimisation: promising on paper, unproven in practice

Portfolio construction and scheduling problems are often framed as optimisation, so they come up in every finance-and-quantum pitch. A survey of the field covers derivative pricing, risk modelling and portfolio optimisation, and also discusses how feasible these algorithms are on near-term devices (Herman et al., 2022). A review of quantum optimisation separates provably exact methods from heuristic ones, and stresses the need for clear benchmarks against strong classical optimisers (Abbas et al., 2023).

That second point deserves emphasis. Classical solvers for portfolio and combinatorial problems are mature and fast. A quantum method has to beat the best of them on realistic instances, not a weak baseline, and for heuristic approaches I am not aware of an established advantage on realistic problems.

Monte Carlo: the clearest theoretical case

Monte Carlo simulation underpins option pricing and many risk measures. Ashley Montanaro described a quantum algorithm that estimates the expected output of a randomised process with a near-quadratic speedup over the best possible classical approach (Montanaro, 2015). Here is what that means, using illustrative arithmetic only.

Illustrative scaling only. It ignores constants, the cost of preparing the model on a quantum device, and error-correction overhead, which can shrink the practical gain considerably on early fault-tolerant machines. A quadratic speedup is valuable, but it is not exponential, and it only arrives with hardware that does not yet exist.

Machine learning: be sceptical of exponential claims

Claims of quantum speedups for machine learning depend heavily on how data is loaded and accessed. A useful cautionary example is Ewin Tang’s 2018 result, which gave a classical algorithm for recommendation systems that is only polynomially slower than a quantum one previously believed to offer an exponential speedup (Tang, 2018). Her algorithm assumes a data structure supporting certain sampling operations, which shows how much the assumptions about data access matter. For everyday analytics on tabular data, there is currently no quantum shortcut to plan around.

The risk that is not hypothetical: harvest now, decrypt later

Peter Shor showed in the 1990s that a large enough quantum computer could factor integers and compute discrete logarithms efficiently (Shor). That threatens public-key schemes such as RSA and elliptic-curve cryptography. Grover’s search algorithm offers only a quadratic speedup (Grover), which is why standards work centres on public-key cryptography.

The urgent part is timing. NIST’s transition report describes the “harvest now, decrypt later” threat: adversaries collect encrypted data today in order to read it once quantum technology matures, and sensitive data often keeps its value for years (NIST IR 8547 draft). For records with long confidentiality lives, the clock is already running, whatever the date of the first capable machine.

What NIST has standardised

In August 2024 NIST published its first three post-quantum standards and encouraged administrators to begin transitioning (NIST announcement):

• FIPS 203: ML-KEM, a key-encapsulation mechanism for establishing shared secrets.
• FIPS 204: ML-DSA, a lattice-based digital signature scheme.
• FIPS 205: SLH-DSA, a stateless hash-based signature scheme.

NIST later selected HQC as an additional key-encapsulation mechanism in March 2025, and Falcon is also in ongoing standardisation (NIST PQC project). The November 2024 draft IR 8547 proposed that quantum-vulnerable RSA and elliptic-curve schemes at 112-bit security be deprecated after 2030 and all of them disallowed after 2035. It is a draft, so check the current status before relying on those dates.

What I would actually do

• Inventory where public-key cryptography is used: TLS endpoints, VPNs, file-transfer jobs, signed documents and vendor integrations.
• Rank data by confidentiality lifetime. Long-lived sensitive records matter most for harvest-now risk.
• Ask suppliers for post-quantum roadmaps and favour products that let algorithms be swapped without a rebuild.
• Do not re-plan analytics around quantum speedups until benchmarks against strong classical baselines exist.

The sober summary is that quantum computing is unlikely to change your modelling stack soon, and is already a reason to review how long your encrypted data needs to stay secret.