Most quantum-computing commentary fixates on the wrong thing: whether a future machine will break the cryptography behind Bitcoin. It's the flashiest scenario and the least useful lens. The more important question — and the one that actually shapes where value gets created — is what quantum builds, and where it fits in the compute stack we're now pouring trillions of dollars into for AI.
What Quantum Is Actually Good At
A quantum computer is not a faster version of the machine on your desk. It is a fundamentally different device that is exceptional at a narrow but profound class of problems — and useless for most others. The trap is imagining it as a universal speed-up. The opportunity is understanding the specific problems where it changes what is possible, not just what is fast.
- Simulating nature. The flagship use case. Molecules and materials are themselves quantum systems, and classical computers choke on simulating them past a handful of atoms. A capable quantum computer could model chemical reactions, catalysts, and materials directly — the kind of work behind better batteries, room-temperature superconductors, fertilizers, and drugs.
- Optimization. Routing, scheduling, portfolio construction, grid balancing, logistics — combinatorial problems where the search space explodes. Quantum (and quantum-inspired) methods can explore those spaces in ways classical heuristics cannot.
- Sampling and certain ML subroutines. Early and more speculative, but there are structured problems in machine learning where quantum sampling could offer real advantage.
Notice what unites these: they are all about creating value — new materials, new molecules, better decisions — not destroying it. The drug-discovery and materials-science applications alone justify the field. That is the story worth telling, and it is the one the headlines bury.
A Clear-Eyed Word on the Signature Risk
It would be incomplete not to mention the risk, so here it is in one paragraph. A sufficiently powerful quantum computer could eventually break the public-key signatures that secure wallets and much of the internet (this is the part that's real; the 'quantum kills mining' claim is not). But it is bounded and mitigable: the machines required remain far beyond today's hardware, and the defense already exists — NIST finalized post-quantum cryptography standards in 2024, and the migration is underway. This is an engineering transition, not an apocalypse. The one habit worth adopting now is 'harvest-now-decrypt-later' hygiene for anything that must stay secret for a decade. With that said — back to the part that matters more.
The Underrated Question: Quantum and the AI Data Center
Here is the angle almost no one connects. We are in the middle of the largest build-out of compute in history, and it is all for AI. So the natural question is: does quantum disrupt the GPU data center that powers the AI boom?
The honest near-term answer is no — not by replacing it. There is no known quantum speed-up for the core of AI inference, which is dense linear algebra. Training and serving large models will stay on GPUs and TPUs for the foreseeable future. Anyone telling you a quantum chip will run your LLM is selling something.
But add an orchestration layer — a control plane that classifies an incoming workload and routes it to the right silicon — and the picture changes. The quantum processor becomes a new accelerator tier inside the data center, invoked for the narrow problems it is uniquely good at, accessed the same way you call a GPU today. Quantum-as-a-service already exists on the major clouds; the missing piece is the software that makes a QPU callable from an ordinary application, without a developer ever writing a quantum circuit. That layer is the bridge — and it is where the most investable opportunity sits.
The lever that makes this consequential is energy. AI's binding constraint is no longer chips — it is power. Data centers are bumping against the grid. For the specific structured problems quantum suits, a QPU can reach an answer that would cost astronomical GPU-hours to brute-force classically. It does not compete with the GPU for inference; it offloads a class of work that is ruinously expensive to run the classical way. On the margin, that eases the single tightest constraint in AI infrastructure.
Follow that thread and a new infrastructure category appears: quantum data centers and the cryogenic, photonic, and networking sub-markets they require; QPU-as-a-service; error-mitigation and compilation middleware; and the routing layer that ties classical and quantum compute into one fabric. None of this substitutes for the GPU build-out. All of it is additive — a new layer in the stack rather than a replacement of the existing one.
This is also where the hardware itself is heading. The GPU began life as a graphics card and became the accelerator that now defines the AI server; the logical next step is the quantum-accelerated server — a node that pairs classical CPUs and GPUs with a quantum processing unit in the same system, so the orchestration layer can hand the right workload to quantum silicon without it sitting in an exotic lab apart from the rest of the stack. Bringing quantum out of the isolated laboratory and into the server is one of the hardest and most valuable problems in the field. It is why we backed Sygaldry, the quantum-computing company founded by Chad Rigetti, a pioneer of the industry. Building the quantum silicon that can eventually live alongside classical compute — rather than in a category of its own — is exactly the kind of atoms-meets-bits bet that compounds as AI's appetite for compute keeps growing.
How We Think About It
Quantum is the purest example of the kind of company we look for in deep tech: an 'atoms-meets-bits' advantage where the moat is physical and scientific. You cannot prompt your way to a quantum processor. In a world where AI is commoditizing generic software, that durability is exactly what makes a category resilient. Both sides of the quantum story are investable — the security migration the signature risk forces, and the accelerator-and-orchestration layer that will plug quantum into the AI data center — and both pass the test we apply to everything: when far more capable AI arrives, does this get more valuable, or less? For quantum, the answer is clearly more.
