Over 1,000 logical qubits by 2031. That’s the target Xanadu Quantum Technologies has put on the board, and it’s the kind of number that sounds either enormous or trivial depending on how much you know about quantum computing. So let’s start there, because that single figure explains why quantum hardware keeps turning up in conversations about AI agents and machine learning security.
Logical qubits are not the qubits you’ve heard about
When a company announces a chip with hundreds or thousands of qubits, those are usually physical qubits — fragile, error-prone, and individually not much use. A logical qubit is a stable unit stitched together from many physical ones, with error correction layered on top. It’s the difference between a pile of wobbly bricks and one brick you can actually build with.
Which is why 1,000 logical qubits by 2031 is an ambitious line in the sand rather than a modest one. It’s a claim about reliability at scale, not just component count. And reliability at scale is exactly what turns a physics experiment into infrastructure that other systems can depend on.
Where the plumbing work is happening
The less glamorous side of this is integration, and that’s where a lot of the recent movement sits. Work on integrated quantum technologies centers on quantum computing and photonics, with real progress in getting qubits onto chips alongside the rest of the hardware stack. Xanadu has also announced collaborations that point squarely at the plumbing:
- Backline, launched with AMD, aimed at streamlining how CPUs, GPUs, and FPGAs connect with quantum technology
- A collaboration with ASML to advance lithography for photonics
Neither of those is a headline that makes anyone’s pulse race. Both matter more than the qubit counts. Lithography is how you manufacture photonic chips at volume instead of hand-building them. And CPU-GPU-FPGA coordination is the boring glue that decides whether a quantum processor is a lab curiosity or something a normal software team can call from normal code.
If you’ve spent time around AI agents, this pattern should feel familiar. Large language models existed for years before anyone could reliably plug one into a calendar, a database, and a payment system. The models got attention. The connective tissue determined what actually shipped.
IEEE is putting quantum on the same list as multi-agent AI
IEEE’s Future Directions work on technology megatrends predicts quantum technologies will play a major role in what comes next. What caught my attention is the company quantum keeps on that list. Alongside it sit neural interfaces, financial system transformation, multi-agent AI ecosystems, autonomous research systems, and high-efficiency bioreactors.
Read that grouping again and notice something: several of those trends are about systems acting on their own. Multi-agent ecosystems are software agents negotiating with other software agents. Autonomous research systems are machines running their own experiments. Financial system transformation is money moving through automated pipelines. These are not trends about better answers to human questions. They’re trends about machines doing things without a person in the loop for every step.
Which brings us to trustworthy machine learning
The IEEE Secure and Trustworthy Machine Learning conference exists because of that shift. When a model just suggests things to a human, mistakes are absorbed by the human. When agents take actions, every weakness in the model becomes a weakness in the world — a bad output becomes a bad transaction, a bad lab order, a bad message sent on your behalf.
I want to be careful here, because the specifics of what’s being submitted to that conference from this line of research aren’t public in anything I’ve seen. What I can point at is the shape of the overlap. Security researchers care about the cryptography that protects models, data, and the channels agents talk over. Quantum computing has a well-known relationship with cryptography. Quantum hardware also depends heavily on error correction, which is another way of saying it’s a field obsessed with getting trustworthy behavior out of unreliable parts. That’s not a bad description of what trustworthy machine learning is trying to do either.
What this means if you’re not a physicist
You don’t need to track qubit counts. You do benefit from noticing three things.
First, timelines are getting specific. A 2031 target is close enough that companies are planning around it instead of gesturing vaguely at the future.
Second, the interesting work is at the seams. Manufacturing partnerships and hardware integration tooling tell you more about maturity than any performance claim.
Third, the agent systems you’re starting to use will eventually run on infrastructure that includes hardware nobody’s explained to you yet. The security questions being raised now, at venues built for exactly that purpose, are the ones that decide whether that’s fine or a problem. Worth caring about before it’s your problem.
đź•’ Published: