Dr. Kris Naudts + Zeynep Koruturk (Founding & Managing Partners) + Donald Harmitt (Associate) @ Firgun Ventures.

For years the headline metric in quantum computing has been a simple one: how many qubits a company can fit onto a single chip. Qubits are the basic units of quantum information, and rising counts signalled a field moving beyond laboratory prototypes, yet that race is destined to meet a wall that physics and manufacturing impose together. Beyond a certain size, fabricating a flawless monolithic chip becomes punishingly difficult, and wiring every qubit to every other qubit grows harder with each addition. The route to a genuinely useful machine therefore runs through a different strategy, one the field now calls modular quantum computing: rather than building a single enormous processor, the smarter path is to link many smaller modules so they behave as one larger machine, scaling outwards through connection rather than upwards through density.

Scaling One Modality by Connecting Many Modules

There are two distinct routes worth separating. The homogeneous approach scales a single qubit modality by networking many identical modules, while the heterogeneous approach combines several different qubit types, superconducting, trapped-ion, neutral-atom, photonic and others, so that each contributes what its physics does best. Both point toward the same destination: a future in which quantum computing lives less in a single exotic device and more in something resembling a high-performance computing centre.

Industrial technologies tend to move from heroic single machines to networked systems, and quantum appears to be approaching the same transition. One of the earliest commercial demonstrations of homogeneous modularity came from Xanadu, the Canadian company whose Aurora system, unveiled in January 2025, was described as the first modular and networked photonic quantum computer. In practical terms the team connected 35 photonic chips and roughly 13 kilometres of optical fibre across four server racks to build a 12-qubit machine. The qubit count was modest, but it was never the point; the significance was that the architecture could in principle scale to thousands of racks and millions of qubits, the foundation of a future quantum data centre. 

Photonics lends itself to this because light travels naturally down a fibre, so the networking layer and the computing medium are one and the same, and much of the system runs at room temperature rather than demanding elaborate refrigeration. Photonic Inc., a Vancouver-based developer of silicon spin-photonic hardware and one of Firgun Ventures’ portfolio companies, has taken a scalability-first posture, treating how to connect modules as the problem to solve before all others. A parallel bet is being placed not on the qubits but on the wiring between them, with Nu Quantum, a Cambridge spin-out, building the networking layer as a standalone product, a sign that the interconnect is increasingly treated as infrastructure in its own right.

The incumbents have reached modularity from the opposite direction, scaling established hardware by stitching chips together. IBM offers the most detailed public roadmap, with its Heron processors designed to link through chip-to-chip couplers and its forthcoming Flamingo processor expected to carry quantum information between separate chips over longer ranges, all within Quantum System Two, a platform built to host interconnected modules rather than one large chip. 

The same instinct is visible in China, where the Wuhan-based in CAS Cold Atom Technology unveiled Hanyuan-2, in May 2026, described as the world's first dual-core neutral-atom quantum computer, pairing two processor cores in a single cabinet. Independent benchmarks are not yet available, but the direction of travel is clear: even state-backed national champions are building modular designs rather than one ever-larger array.

Different Qubits, Better Together

The heterogeneous view is the more ambitious of the two, and it begins from a simple observation, namely that each qubit modality has distinct strengths rooted in its physics, and those strengths are unlikely to converge as the technology matures. Superconducting qubits, favoured by IBM and Google, are fast and benefit from mature semiconductor fabrication; trapped ions offer exceptionally high fidelity; neutral-atom arrays scale to large numbers and suit the simulation of physical systems; and photonics excels at networking while operating at room temperature. No single technology delivers everything. 

That insight now carries institutional weight, with DARPA launching its Heterogeneous Architectures, for Quantum programme, known as HARQ, in 2026 to combine multiple qubit technologies into one system, across a software work stream that assigns each task to the right qubit type and a hardware work stream that builds the interconnects linking distinct platforms. The same shift is underway in classical chipmaking, where Patrick Vandenameele, chief executive of the Belgian nanoelectronics institute Imec, argues that next-generation AI systems are becoming heterogeneous assemblies of technologies, that must be co-optimised across the whole stack rather than improved in isolation.

A battery-chemistry problem makes the division of labour concrete: neutral-atom arrays could simulate candidate electrolyte molecules, trapped-ion systems could handle the high-fidelity calculations of reaction and degradation pathways where tiny differences decide stability, superconducting processors could run the algorithms passing work between quantum simulation and classical machine learning, and photonics could serve as the networking layer binding the modules into one. Each component does what it is best at, and the fibre between them makes the whole behave as a single machine. This is not a prediction that all these pieces will be integrated tomorrow. It is a way of seeing where the field may be heading.

Where the Investment Case Sits

For investors, the strategic implication of all this is the most important part, because modularity subtly relocates where value in quantum hardware is likely to accrue. If the defining problem is no longer how to build a better qubit but how to connect qubits, whether identical or diverse, then the interconnect ceases to be supporting infrastructure and becomes the main event. That reframing favours a particular kind of company. Firms treating connection as their first-order problem, among them Photonic Inc., and the spin-photonics work pursued by companies such as Quandela, are building the layer on which both homogeneous and heterogeneous machines will depend, and the photonic interconnects that HARQ advances for heterogeneous systems will benefit homogeneous ones just as much. The clearest expression of the thesis is Nu Quantum in Cambridge, which has made the networking layer itself the product rather than a feature of someone else's processor.

The Shape of What Comes Next

Progress here is genuine but early, since cross-module operations remain slower and noisier than those within a single chip, and none of this implies that every piece will be integrated tomorrow. The trajectory is nonetheless consistent across photonic challengers, semiconductor incumbents and state-backed champions alike: the industry is moving from isolated processors toward connected systems, and from architecture debates toward system integration. The future quantum data centre housing is on the trajectory to house several technologies side by side, each carrying the part of the problem it handles best. The companies that learn to connect quantum processors, rather than merely enlarge them, are building the architecture on which a useful quantum industry will eventually run.