Quantum AI is not an IT priority yet: what it means and what companies should actually do in 2026

Combining two technologies that receive enormous amounts of attention —artificial intelligence and quantum computing— can sound inevitable.

The phrase Quantum AI suggests the next major leap: AI models trained on quantum computers, agents capable of solving problems beyond the reach of GPUs, and a new generation of infrastructure that makes today’s data centers obsolete.

But that picture mixes genuine research progress with expectations that remain far from enterprise production.

An article circulated by MSN, based on Gartner research, makes exactly that distinction. Gartner predicts that enterprise AI workloads at scale will not run on quantum hardware through at least 2028, while classical accelerated computing will continue to dominate production benchmarks during that period.

The conclusion is not that quantum computing is irrelevant.

The more useful conclusion for a CIO is that Quantum AI and the AI generating business value today operate on different timelines, economics, and maturity levels.

What Gartner is actually saying

In August 2026, Gartner published an unusually clear prediction: it does not expect any enterprise AI workload at scale to depend on quantum hardware before 2028.

Its argument rests on an important technical point: there is still no peer-reviewed result demonstrating quantum advantage on a production AI workload in performance, cost, or capability.

That is far more specific than saying “quantum computers are useless.”

In fact, 2026 has produced important progress in quantum computing. IBM and the University of Chicago, for example, announced quantum advantage experiments involving logical circuits that are difficult to simulate classically. IBM and collaborators have also reported results in quantum-material simulation.

The distinction matters:

demonstrating quantum advantage on a scientific or benchmarking problem is not the same as demonstrating quantum advantage for training, inference, or operating an enterprise AI workload.

Both statements can be true at the same time:

quantum computers can reach milestones beyond some classical methods;
enterprise AI can still run more effectively and economically on CPUs, GPUs, and TPUs for years.

The problem starts with the definition of “Quantum AI”

Part of the confusion comes from several different categories being grouped under the same label.

1. Classical AI

This is the AI already running in production:

transformers;
multimodal models;
deep learning;
reinforcement learning;
recommendation systems;
LLM-based agents.

All of it runs on classical infrastructure: CPUs, GPUs, TPUs, and specialized accelerators.

2. Quantum-inspired AI

A quantum computer is not necessarily involved here.

These are classical algorithms that borrow ideas from quantum techniques —for example, some forms of annealing, tensor networks, optimization, or sampling— while still running on conventional hardware.

That matters because a company can experiment with quantum-inspired methods today without buying or renting a QPU.

3. Hybrid quantum-classical systems

In these workflows, one part of the problem is sent to a quantum processor while the rest remains on classical infrastructure.

A simplified flow looks like this:

classical application
        ↓
CPU/GPU preprocessing
        ↓
small quantum circuit
        ↓
QPU
        ↓
probabilistic measurements
        ↓
classical postprocessing
        ↓
application

This approach is real and can be used in research and pilot projects.

But it is still very different from running a complete enterprise model inside a quantum computer.

4. Quantum-native AI

This is the stronger version of the promise: an AI technique that requires quantum hardware and delivers a measurable advantage over the best classical alternatives.

That is the scenario for which Gartner says production evidence does not yet exist.

Four obstacles before production Quantum AI

Gartner highlights four areas that still need substantial progress.

Hardware

Physical qubits are fragile.

Noise, decoherence, and gate errors limit how many operations can be executed reliably and how deep circuits can become.

Simply increasing the number of physical qubits does not solve the problem by itself.

Error correction

Large, reliable algorithms require logical qubits protected by many physical qubits.

That is why the number that matters most is not always the one in a vendor headline.

A system with many physical qubits but high error rates may be less useful than another system with fewer qubits but better logical qubits and more reliable circuits.

Middleware

A quantum computer does not operate in isolation.

Software is required to:

compile circuits;
map them to available hardware;
calibrate devices;
manage execution queues;
orchestrate classical and quantum execution;
mitigate errors;
observe and reproduce experiments.

That tooling layer is still evolving rapidly.

Algorithms

Even if the hardware improves, another question remains:

Which AI algorithm can exploit it and outperform an optimized classical implementation?

The correct comparison is not against a generic CPU from a decade ago.

It is against modern GPU clusters, highly optimized compilers, specialized kernels, quantization, sparsity, smaller models, and distributed architectures that are improving at the same time.

Physical qubits vs. logical qubits: the distinction that changes the conversation

The industry often announces processors by their qubit count.

But for fault-tolerant applications, the increasingly important metric is the logical qubit.

A logical qubit uses redundancy and error correction to represent information much more reliably than an individual physical qubit.

In simplified form:

many physical qubits
        ↓
error correction
        ↓
fewer reliable logical qubits
        ↓
useful large-scale algorithm

That is why Gartner recommends tracking metrics such as useful logical qubits and error rates instead of focusing only on headlines about processors with more physical qubits.

What about recent “quantum advantage” announcements?

It is also important to avoid the opposite mistake: assuming that because Quantum AI is not production-ready, nothing important is happening in quantum computing.

In July 2026, IBM and the University of Chicago announced an experiment involving 70 logical qubits designed to execute circuits that are difficult to simulate classically while statistically validating the fidelity of the result.

IBM also published work with Qedma and Algorithmiq involving quantum systems and materials in regimes that challenge classical methods.

These are meaningful advances.

But the workload is the key difference.

An experiment that demonstrates advantage in sampling, quantum dynamics, or material simulation does not automatically prove that training a transformer, serving an LLM, or running an agent system is better on a QPU.

That claim needs its own evidence.

The real IT problem is opportunity cost

Organizations do not choose technologies in a vacuum.

Every dollar spent on an experiment competes with other priorities:

AI infrastructure;
data quality and governance;
observability;
cybersecurity;
automation;
agents;
model evaluation;
reliability;
cloud modernization.

Gartner argues that mixing the Quantum budget with the AI budget makes it harder to measure the return on either one.

Generative AI and agentic AI can already be evaluated through metrics such as cycle-time reduction, automation, accuracy, or cost per operation.

Quantum AI still looks much more like an R&D investment and a bet on future optionality.

That is why managing the two budgets separately makes sense.

So what is QCaaS?

A company that wants to learn quantum computing does not need to build a quantum computer.

QCaaS — Quantum Computing as a Service — provides access to quantum hardware through the cloud.

Conceptually it resembles renting a GPU, although the operating model is different.

Python / SDK
    ↓
quantum circuit
    ↓
cloud service
    ↓
execution queue
    ↓
QPU or simulator
    ↓
measurements
    ↓
result returned to the application

Current services include:

Amazon Braket, which provides simulators and access to multiple kinds of quantum hardware through AWS;
IBM Quantum Platform, which offers remote access to QPUs and the Qiskit ecosystem;
Azure Quantum, which provides hardware from multiple providers, simulation, and resource-estimation tools.

QCaaS is useful for:

education;
research;
prototyping;
benchmarking;
exploring hybrid algorithms;
understanding the real constraints of the hardware.

What an organization should not do is turn a QCaaS experiment into a business-ROI claim before comparing it against a serious classical baseline.

The most concrete quantum priority for IT is probably not Quantum AI

Paradoxically, one of the quantum-related tasks that should already be appearing on security roadmaps does not require a quantum computer at all.

It is post-quantum cryptography (PQC).

NIST has already published three principal standards:

ML-KEM for key establishment;
ML-DSA for digital signatures;
SLH-DSA for hash-based signatures.

NIST recommends that organizations begin migration now because discovering where vulnerable cryptography is used, updating protocols, replacing certificates, and coordinating dependencies can take years.

This creates a very useful distinction:

Quantum AI
Can quantum computing accelerate enterprise AI?
→ still primarily R&D

PQC
How do we protect today's systems against future quantum attacks?
→ engineering work can start now

For many IT teams, PQC is far more actionable today than Quantum AI.

A reasonable strategy for 2026

Instead of choosing between “ignore quantum” and “bet on quantum,” an organization can split the problem across different time horizons.

Area	Maturity in 2026	Reasonable action
GenAI and agentic AI	Production	Invest, measure ROI, and improve reliability
Post-quantum cryptography	Migration underway	Inventory cryptography and prepare transition
Quantum-inspired methods	Usable on classical hardware	Test when there is a concrete problem
Quantum-classical / QCaaS	Experimental	Run small pilots with R&D budgets
Quantum-native AI at scale	Research	Track progress, but do not design the enterprise stack around it

And every quantum pilot should begin with something innovation programs sometimes lack: exit criteria.

Before running the experiment, define:

the best classical baseline;
the metric that must improve;
the maximum pilot budget;
the result that would justify continuing;
the result that should stop the project.

That prevents what Gartner effectively warns about as vendor theater: pilots that continue because the technology is fascinating rather than because they are approaching a viable business case.

Which signals are actually worth watching?

Knowing when to change this posture requires more than counting quantum-computing announcements.

The most meaningful signals would include:

more useful logical qubits;
lower logical error rates;
deeper fault-tolerant circuits;
reproducible benchmarks against the best classical methods;
algorithms with advantage on commercially relevant workloads;
competitive total cost;
tooling and operations mature enough for production.

When several of those variables change at the same time, the conversation will begin to move beyond pure R&D.

Conclusion

Quantum computing is advancing, and some recent experiments are genuinely important.

But quantum progress and enterprise Quantum AI are not synonyms.

In 2026, a production AI architecture is still built around GPUs, CPUs, TPUs, cloud infrastructure, data, observability, evaluation, and classical software. Quantum can belong in a laboratory, a research program, or a QCaaS pilot, but there is not yet enough evidence to make it a dependency of the enterprise AI platform.

The right strategy is not to ignore quantum computing.

It is to place each technology on the correct time horizon:

classical AI and agentic AI for producing value now;
PQC for preparing future security;
quantum-inspired techniques for experiments on current infrastructure;
QCaaS for learning within clear boundaries;
quantum-native AI as a research bet that must prove its advantage before receiving production budget.

That is less spectacular than promising an immediate revolution.

It is also a much more responsible way to build technology.