## Does Quantum Computing Need to Be Physically Co-Located with a Supercomputer?

For most algorithms running today, the answer is no — but the moment you need quantum error correction, it becomes non-negotiable. That is the headline finding of a performance model published on July 21, 2026, by algorithm researchers from [Amazon Web Services (Quantum)](https://quantumintel.tech/companies/amazon-web-services), Lawrence Berkeley National Laboratory (LBNL), NASA, and [NVIDIA (Quantum)](https://quantumintel.tech/companies/nvidia). The paper — available on arXiv at arxiv.org/abs/2607.15426 — introduces a communication-to-computation ratio (R_cc) that lets engineers and infrastructure planners determine, quantitatively, whether a given [hybrid quantum-classical](https://quantumintel.tech/glossary/hybrid-quantum-classical) workload is bottlenecked by latency or by compute.

The key finding is striking in its simplicity: sample-based quantum diagonalization (SQD), measured on 77 qubits of an [IBM Quantum](https://quantumintel.tech/companies/ibm) Heron device, yields an R_cc of approximately 10⁻⁴ — meaning communication overhead is negligible even over wide-area networks. In contrast, quantum error correction (QEC) for superconducting devices operates on microsecond timescales where every nanosecond of decoder latency matters. The two regimes are not just different in degree; they require fundamentally different infrastructure strategies.

---

## The R_cc Model and Why It Matters

The paper decomposes the total runtime of a hybrid workflow into three components: classical compute time (T_C), quantum compute time (T_Q), and communication overhead (T_comm). From these, the researchers define:

**R_cc = T_comm / (T_C + T_Q)**

The analogy to classical HPC's roofline model is deliberate and apt. Just as the roofline model separates compute-bound from memory-bandwidth-bound programs, R_cc separates compute-bound hybrid quantum workloads from communication-bound ones. A workflow with low R_cc can tolerate cloud-level network latency without meaningful performance degradation. A workflow with high R_cc — where communication dominates the cycle — demands tight, low-latency coupling between quantum and classical hardware.

This kind of quantitative framework is something the field has needed. Infrastructure discussions in quantum-HPC integration have frequently conflated two distinct engineering problems, and the paper separates them explicitly.

---

## Two Levels, Two Different Engineering Problems

The researchers draw a sharp line between what they call the **real-time level** and the **application level**.

**Real-time level:** This is the domain of QEC decoders and control stacks. Classical processors must receive error syndromes from the quantum processor, decode them, and feed corrections back within the logical gate cycle — on the order of microseconds for superconducting devices. There is no negotiating with physics here: if the correction latency exceeds the logical gate cycle, [logical qubit](https://quantumintel.tech/glossary/logical-qubit) fidelity collapses. Co-location of QPUs with high-performance classical compute is not optional at this level.

**Application level:** This is where today's [NISQ](https://quantumintel.tech/glossary/nisq) algorithms live — variational solvers, quantum-enhanced sampling, generative quantum eigensolvers. These workflows exchange data between circuit executions, not within them. The classical host is not racing against a device clock. For these algorithms, the question of whether low-latency integration helps reduces to a calculation: run the R_cc numbers.

---

## Specific Algorithm Benchmarks from the Paper

The paper applies the R_cc model to published NISQ demonstrations, replacing intuitions with arithmetic. Three results are particularly instructive:

**Sample-Based Quantum Diagonalization (SQD):** A leading near-term approach for quantum chemistry. Combines shallow quantum circuits with computationally heavy classical diagonalization. Measured on 77 qubits of IBM's Heron device, R_cc ≈ 10⁻⁴. Communication overhead is negligible. Remote cloud access is adequate. This is a significant data point for teams considering whether to invest in co-located infrastructure for chemistry workloads.

**Generative Quantum Eigensolver (GQE):** Trains an AI model using quantum energy feedback. The paper reports R_cc values around 10⁻¹⁰ even under extensive compute scaling assumptions. An extraordinarily compute-bound workflow.

**Quantum-Enhanced Markov Chain Monte Carlo (QE-MCMC):** A contrasting case. With R_cc ≈ 10⁻¹ under remote-access assumptions (measured on a 10-qubit IBM device), this workflow is communication-sensitive. But critically, the paper notes the classical step per iteration is estimated at approximately 10 nanoseconds — meaning tight latency is needed, but heavyweight HPC resources are not. A dedicated, low-latency link to modest classical compute may be the right infrastructure call here, not a rack inside a supercomputer facility.

---

## The QEC Imperative and the AWS-QuEra Angle

The paper is explicit that the real-time QEC regime is where co-location becomes architecturally mandatory, and this maps directly onto where the industry is heading. [Amazon Web Services (Quantum)](https://quantumintel.tech/companies/amazon-web-services) has a stated strategic collaboration with [QuEra Computing](https://quantumintel.tech/companies/quera-computing) to bring [fault-tolerant quantum computing](https://quantumintel.tech/glossary/fault-tolerant-quantum-computing) to Amazon Braket. QuEra's neutral atom platform, which has demonstrated logical qubit operations, will eventually require exactly the kind of low-latency decoder infrastructure this paper characterizes. Publishing this framework now is, in part, an act of infrastructure roadmapping — telling the HPC community what to build and when.

The paper also flags two specific crossover conditions that could shift these conclusions even for application-level algorithms: dramatic shot-budget reductions via amplitude estimation, and the emergence of practical quantum memory. If either materializes, R_cc values for currently compute-bound workflows could shift toward the communication-sensitive regime. The model is designed to accommodate these updates as hardware capabilities evolve.

---

## What This Means for HPC Center Investment Decisions

The practical implication for national labs, hyperscalers, and enterprise buyers is that the co-location decision is not binary and should not be treated as such. The paper provides a quantitative method to evaluate specific workloads against specific infrastructure configurations.

For HPC center directors: today's quantum chemistry and optimization workloads likely do not justify the capital expenditure of physical QPU co-location — remote network access is sufficient. For teams architecting systems around fault-tolerant timelines, the calculus is reversed: every microsecond of decoder latency counts, and planning for QPU-HPC co-location should start now, before device scales make retrofitting prohibitively expensive.

The paper's authors — Dimitar Trenev, Sebastian Stern, Tyler Takeshita, Cedric Lin, Peter Komar, Pooja Rao, and Jerome (name truncated in source) — have produced what is, analytically, a tool for separating marketing from engineering in quantum-HPC integration discussions. That is a more durable contribution than any single benchmark result.

---

## Key Takeaways

- **R_cc = T_comm / (T_C + T_Q)** is the central metric: values well below 1 indicate compute-bound workflows where network latency is irrelevant; values approaching or exceeding 1 indicate communication-sensitive workflows.
- **SQD on IBM Heron (77 qubits) has R_cc ≈ 10⁻⁴** — remote cloud access is sufficient for today's leading quantum chemistry approach.
- **GQE shows R_cc ≈ 10⁻¹⁰** — among the least latency-sensitive hybrid algorithms analyzed.
- **QE-MCMC has R_cc ≈ 10⁻¹** under remote access, requiring low latency but not heavyweight classical compute.
- **QEC is the non-negotiable case**: superconducting decoder loops operate on microsecond timescales where co-location is architecturally mandatory.
- **AWS's QuEra collaboration** provides the commercial context for why this framework is being published now, ahead of fault-tolerant deployments on Braket.
- Two future conditions — amplitude estimation reducing shot budgets, and practical quantum memory — could shift application-level R_cc values and change infrastructure requirements.

---

## Frequently Asked Questions

**What is the communication-to-computation ratio (R_cc) in quantum-HPC workflows?**
R_cc is defined as T_comm divided by the sum of classical and quantum compute times (T_C + T_Q). A low R_cc means the workflow is compute-bound and tolerates network latency without performance loss. A high R_cc means communication overhead dominates and demands low-latency coupling between quantum and classical hardware.

**Do NISQ algorithms need to be co-located with supercomputers?**
Based on the AWS-LBNL-NASA-NVIDIA analysis, most current NISQ algorithms do not. Sample-based quantum diagonalization on IBM's Heron device showed an R_cc of approximately 10⁻⁴, meaning wide-area network access introduces negligible overhead. The exception is communication-sensitive workflows like quantum-enhanced MCMC, which need low latency but not necessarily HPC-scale classical resources.

**Why does quantum error correction require co-location with classical compute?**
QEC decoders must process error syndromes and return corrections within the logical gate cycle of the quantum processor — on the order of microseconds for superconducting devices. Any communication latency that exceeds this window degrades logical qubit fidelity. This is a hard physical constraint, not an engineering preference.

**Which organizations published this quantum-classical integration framework?**
The paper was published on July 21, 2026, by researchers from Amazon Web Services, Lawrence Berkeley National Laboratory, NASA, and NVIDIA. The full paper is available at arxiv.org/abs/2607.15426.

**When will co-location of QPUs with HPC become necessary for commercial quantum workloads?**
The paper identifies the onset of practical fault-tolerant quantum computing — specifically, when applications begin drawing on logical qubits requiring active QEC — as the threshold. The AWS-QuEra collaboration targeting fault-tolerant quantum computing on Amazon Braket is a concrete near-term indicator of when this transition becomes commercially relevant.