Scaling millions of qubits: How Quantum Data Centers bring new possibilities for quantum computing

Quantum computing is making remarkable strides, rapidly progressing from processors with just tens of qubits to those with hundreds. While this milestone is noteworthy, developing a truly practical quantum computer ultimately demands scaling to millions of qubits. Along the way, we face significant obstacles, including the challenge of increasing qubit counts in monolithic architectures and offering quantum computing “as a service.”

One promising avenue is the emergence of Quantum Data Centers (QDCs). These facilities leverage distributed quantum computing by networking multiple smaller-scale quantum processors, much like traditional data centers. Housed in controlled environments—such as specialized data center warehouses—QDCs gain crucial benefits. The precision in temperature, timing, and phase stability, combined with the relatively short distances between processors, helps minimize loss and maintain high fidelity of quantum information. This setup not only ensures reliable interconnections among quantum computers but also provides a scalable and efficient foundation for quantum computing infrastructure.

Quantum Data Center architecture design principles 

To achieve true scalability, a QDC architecture should incorporate several key principles:

• Modularity and scalability: The system must facilitate flexible interconnect setups and seamless integration of new quantum processors.
• Multi-tenancy: It should support the concurrent execution of various quantum applications and algorithms, maximizing shared resources and efficiency.
• Any-to-any connectivity: All quantum processors should be able to interconnect freely, ensuring versatility and high performance.
• Multiple logical interconnect topologies: The design must allow for the creation of multiple, coexisting, and independent network topologies, each optimized for different quantum algorithms.
• Platform independence: The QDC should be capable of supporting heterogeneous quantum computing platforms, allowing diverse hardware to coexist within the same data center.
• Adaptive interconnect configuration: The logical interconnect topology should be modifiable on demand—even mid-algorithm—enabling sub-routine-level optimization for maximal performance.
• Shared critical resources: The architecture should include provisions for sharing specialized and costly resources, such as photon detectors and Bell State Measurement (BSM) devices, to improve operational efficiency.
• Quantum network-aware orchestration: A dedicated orchestrator, equipped with network-aware quantum circuit partitioning and scheduling mechanisms, should manage the entire data center, ensuring optimal resource utilization and smooth workflow.

Below, Figure 1 illustrates a high-level vision of this modular, scalable QDC network architecture, designed to remain independent of any specific quantum computing platform.

A pivotal element of the QDC architecture is the quantum-enabled switch, which serves as the foundational building block for a dynamically switchable quantum network fabric connecting multiple quantum processing units (QPUs). Its standout features include:

• Non-blocking switching of entangled photons, seamlessly directing signals from any input port to any output port.
• Support for diverse entanglement modes, including time-bin, frequency-bin, and polarization-based methods.
• Ultra-low loss and minimal time jitter, enabling reliable cascading of multiple switches without significant signal degradation.
• Expandable design, accommodating pluggable functional modules (e.g., BSM devices, detectors, delay lines, and memories) to be shared on demand among QPUs.

For inspiration, QDC network topologies can look to classical data center designs, specifically switch-centric or server-centric models.

Figure 1(a) illustrates a switch-centric Clos network. Here, groups of QPUs are housed in racks, each with a top-of-rack (ToR) switch. These racks are then interconnected via a tiered arrangement of optical switches, which are equipped with quantum hardware for entanglement generation. To address mismatches between qubit resonant frequencies (typically in the near-infrared range of 700–900 nm for many atomic or ionic platforms) and telecom wavelengths, the architecture separates:

• Intra-rack communication: Uses near-infrared (NIR) frequencies for short distances, leveraging emitter–emitter or emitter–scatterer protocols.
• Inter-rack communication: Operates at telecom wavelengths over longer distances, using scatterer–scatterer protocols.

Since the ToR switch and its connected components run at NIR frequencies, entanglement sources must be non-degenerate: one photon output remains in the NIR regime, while the other is at telecom wavelengths. Alternatively, quantum frequency converters (QFCs) can shift photons from NIR to telecom (and back) at the ToR switch ports, extending communication to other racks.

Figure 1(b) illustrates a server-centric topology. In contrast, server-centric topologies employ smaller switches with fewer ports but equip each QPU with multiple ports. These networks can often operate entirely at NIR frequencies, favoring emitter–emitter protocols (though emitter–scatterer may still be used). Routing and resource management in server-centric architectures entails addressing repeater networks—an extensive topic in its own right. 

However, compared to long-haul quantum communication, QDC-scale networks benefit from an efficient global control plane. Moreover, entanglement swapping can be carried out deterministically, as QPUs have the capacity to apply deterministic gates between their communication qubits. This capability further streamlines network operations, enabling more reliable and scalable QDC infrastructures.

Entanglement generation and protocols

In a Quantum Data Center (QDC), each QPU typically contains two types of qubits: Data qubits, used primarily for executing quantum computations and communication qubits, responsible for creating and storing entangled qubits (often called “ebits”) that enable networking between different QPUs. The primary objective of a QDC network is to generate entanglement between the communication qubits of various QPUs. These entangled pairs are then used to perform remote gates between data qubits residing in different QPUs. Depending on the underlying QPU technology, there are three main methods for establishing entanglement:

• Emitter–emitter: Both communication qubits in the QPUs emit photons, which then undergo a post-selected BSM at a central location.
• Emitter–scatterer: The first QPU’s communication qubit (emitter) emits a photon, which is captured by the second QPU’s communication qubit (scatterer) and finally measured by a photon detector.
• Scatterer–scatterer: Two entanglement sources each produce entangled photon pairs. One photon from each pair is directed to a central BSM, while the other photon is sent to the QPUs at the “ends” of the network.

A key similarity across all three methods is that long-distance quantum communication relies on single-photon states (flying qubits). The emission or scattering process is designed so that each flying qubit becomes entangled with its corresponding stationary (communication) qubit. Because QDCs rely on the dynamic generation and immediate consumption of entangled qubits (“ebits”), on-demand scheduling protocols play a crucial role. These protocols allow communication qubits—typically of lower quality and shorter coherence times than data qubits—to be used effectively by generating entanglement only when needed. As soon as an ebit is created, it is quickly consumed to perform remote operations, eliminating the need for long-term storage of quantum information on the communication qubit.

Quantum Data Center orchestrator

In a QDC, an orchestrator is essential for managing two primary tasks: circuit partitioning and network scheduling. The circuit compiler takes a quantum circuit (expressed in terms of logical qubits) and maps it onto physical qubits within QPUs. It assigns logical qubits, executes inter-QPU gates via teleportation, and strives to minimize the number of remote gate operations.

Meanwhile, the network scheduler, overseen by a central controller, coordinates the sequential control of quantum hardware and optical switches. This involves reconfiguring optical switches and allocating quantum network devices to establish end-to-end optical paths between communication qubits. Because multiple attempts may be needed to generate entangled qubits (ebits) until a heralding signal confirms success, synchronization across terminal QPUs and other devices is crucial to ensure smooth execution.

Beyond these core functions, the orchestrator must also support multi-job (multi-circuit) scheduling. New jobs can arrive at any time—even while others are mid-schedule or mid-execution. In multi-tenancy scenarios, where a complete list of jobs is available, the orchestrator aims to create an optimal, unified schedule that minimizes both latency and infidelity, ultimately enhancing the efficiency and reliability of the Quantum Data Center.

Quantum Data Center architectures for distributed quantum computing

We have outlined a foundational approach for developing large-scale quantum computing infrastructure within a data center framework—bringing us a step closer to realizing practical quantum advantage. QDC network architectures move beyond traditional point-to-point connections, creating new opportunities for distributed quantum computing.

This provides a scalable solution based on dynamic, circuit-switched quantum networks capable of distributing entanglement among multiple QPUs. By leveraging shared quantum resources and embracing modular topologies—such as switch-centric and server-centric designs—the network can offer on-demand, all-to-all connectivity while minimizing reliance on expensive quantum hardware.

To seamlessly manage the execution of distributed quantum jobs, a network-aware quantum orchestrator, along with entanglement generation protocols, can be implemented. This orchestrator bridges the gap between physical-layer architectures and quantum applications, ultimately advancing the vision of scalable, efficient, and accessible Quantum Data Centers.

Explore our latest findings on Quantum Data Center architectures and learn how the Cisco Research team is exploring how to build reliable quantum networks to meet future computing needs in our paper, Quantum Data Center Infrastructures: A Scalable Architectural Design Perspective.