Despite the continuing uncertainties in technology and timing, business investment in quantum computing is on the rise. BCG’s latest research shows that enterprise spending reached new heights of about $550 million in 2025, surpassing total investment by academia and government for the first time. More than 90% of leading global finance and insurance firms, 56% of top global health care and biopharma companies, and 52% of industrials companies are investing in quantum computing. More than half of top companies across sectors (excluding consumer goods and real estate) have some form of investment in the technology.
Deciding to invest in a powerful, but unproven, technology is only the first step. Companies new to the field soon discover that there is no such thing as a single quantum computer. In fact, five distinct technological approaches to hardware, each with different pluses and minuses, are advancing at their own stages of maturity. (See Exhibit 1.)
Each of these systems comes with its own set of engineering decisions. In quantum circles, they are known as modalities. To keep an already complex technology as simple as possible, we’ll refer here to hardware systems. Here’s our assessment of today’s landscape and the business considerations and tradeoffs that companies face.
Which Factors Matter?
Qubits (the quantum equivalent of classical 1-0 bits) can be built in different ways. The choice of hardware matters because each system, and the qubits it creates, behave differently when you try to put them to work. For example, some qubits are stable for long periods of time (a big challenge in quantum computing) but slow to operate. Others are fast actors but lose their quantum properties almost immediately. Some can be manufactured at scale using existing semiconductor techniques. Others require bespoke assembly in conditions that are difficult to reproduce outside a laboratory (such as near-absolute zero temperatures). Others operate closer to room temperature, though never quite at it.
These are not minor variations on the same engineering theme. They are fundamentally different approaches to the underlying technological challenge: isolating and controlling a quantum system well enough to perform useful computations before the system loses its quantum properties, a phenomenon known as decoherence.
Four factors define how well a system performs and how it compares with the alternatives. Each approach trades strength in one area for weakness in another.
Scaling—how many qubits a system can support today and how many it can realistically support. Qubit count matters because most commercially useful quantum algorithms require far more qubits than current systems provide.
Most systems reach their current qubit counts by linking multiple self-contained units together. We refer to each of these units, such as a chip in a superconducting system or a trap in an ion-trap system, as a module, or a self-contained quantum-system unit that vendors link with other modules to scale up the total qubit count.
The vast majority of vendors report module-level specifications only. However, linking modules can have an impact on fidelity, connectivity, and overall system performance, which requires additional metrics to properly assess these architectures. In some R&D systems, we have observed module-to-module fidelities as low as 90%.
Addressability and Noise—how accurately individual qubits can be controlled and how reliably two-qubit operations, which are the building blocks of quantum algorithms, can be performed. This is typically measured by gate fidelity: a two-qubit gate at 99.9% fidelity means one error per thousand operations.
Coherence Time—how long a qubit retains its quantum state before decoherence sets in. Longer coherence times allow for more complex computations before errors accumulate beyond the ability to recover.
Clock Speed—how fast the system can execute operations. A highly accurate but extremely slow system may be impractical, or no more useful than current supercomputers, for computationally intensive tasks.
Five Contenders
Since no system today leads across all four factors, the challenge for business is to understand which tradeoffs best match the desired application or use case. This understanding separates a well-informed quantum strategy from a naive one. The task is complicated by the fact that systems vary in maturity, infrastructure footprint, and commercial trajectory. (See Exhibit 2.)
Scaling a quantum computer may ultimately require connecting smaller quantum processing units, or modules, rather than enlarging a single processor. This can simplify manufacturing, testing, and control, while extending a system’s effective qubit capacity and connectivity. But every module boundary requires an interconnect—electrical, photonic, or otherwise—that can introduce latency, error, and control overhead. The practical value of modularity will therefore depend on the quality and speed of these links, not simply the number of modules connected.
In terms of maturity, the five systems can be divided into three groups: the proven leader (superconducting circuits), capable challengers (trapped ions and neutral atoms), and high-potential unknowns (photon-based and spin qubits). Other technologies, such as topological qubits, are in earlier stages of development but are not yet mature enough to compare against the four factors.
Superconducting Circuits. The most-often-pictured quantum computer, a gold-colored chandelier hanging inside a cryogenic cylinder, is a superconducting system, the dominant hardware today. Backed by the likes of IBM and Google for more than a decade, it is in a maturity category of its own and has a significant share of the quantum hardware industry. Tech leaders have made operational systems accessible to external users for years.
Qubits are fabricated on chips using loops of superconducting material, typically aluminum or niobium, and controlled using microwave pulses. They operate at temperatures of approximately 15 millikelvin (colder than interstellar space), requiring large dilution refrigerators and dense control wiring. The engineering challenges are significant, but superconductivity is among the most studied phenomena in physics, with decades of academic and industrial research behind it.
The result is hardware with the highest current digital qubit counts, fast gate speeds, and a fabrication logic that borrows from semiconductor manufacturing. (Unless otherwise noted, when we cite current qubit counts, they reflect what’s been achieved within a single chip, trap, or module; not connected totals.)
The tradeoffs are well understood. They include fidelity that lags competing modalities, qubit connectivity that is limited by the geometry of the chip, and scaling that requires solving hard problems in wiring density and thermal management. The engineering challenges at scale are real, but they are known challenges, which is itself a form of advantage.
Trapped Ions. Trapped ion systems use individual charged atoms, typically ytterbium or barium, suspended in electromagnetic fields and manipulated with lasers. Rather than being fabricated on a chip, qubits exist in the atoms themselves, which means they are naturally identical to one another, an advantage that contributes to hardware’s exceptional fidelity profile.
Trapped ions achieve two-qubit gate fidelities at or above 99.9%, the highest of any system, and coherence times measured in minutes rather than microseconds. The tradeoff is speed: trapped ion operations run at up to the kilohertz range—and often in the single-digit hertz range—four to five orders of magnitude slower than superconducting systems. Optical control systems also introduce infrastructure complexity that complicates scaling. The technology is well-established but reaching the qubit counts needed for fault-tolerant computation remains a significant engineering challenge.
Neutral Atoms. Neutral atom systems trap and arrange uncharged atoms in arrays using focused laser beams called optical tweezers. Like trapped ions, they operate at the atomic level, offering long coherence times and strong fidelity but slower clock speeds than superconducting systems. Unlike trapped ions, neutral atoms have demonstrated a compelling path to scale.
This system also enables a distinctive capability: quantum memory. Atoms can be moved between active computation zones and storage positions, allowing information to be held while other parts of the system process. This is useful for complex algorithms that require intermediate results to be preserved and, at least in the current generation of hardware, makes error correction easier than on other platforms, since idle qubits can be shielded from ongoing computation rather than exposed to it.
Neutral atoms also support an analog mode, which is distinct from the gate-based digital model used elsewhere in the field. The analog approach performs instant computations by using the laws of physics to bypass discrete logic gates. Early results are promising; researchers have reported real signs of quantum advantage in domains such as materials science. The tradeoff is precision. Analog computations are easier to achieve but harder to control, not unlike the analog recording technologies (such as vinyl records and magnetic tape) that preceded more precise digital formats (such as CDs and MP3s). It’s an early-stage capability, but one worth watching.
Clock speed remains the primary constraint—operations run orders of magnitude slower than superconducting, but for algorithms where fidelity and coherence matter more than raw throughput, this tradeoff is navigable. Hardware advances are also coming quickly.
Photon-Based. Photon-based systems encode quantum information in particles of light rather than in matter. Photons travel through optical circuits, and their quantum states can be manipulated at room temperature, or close to it, but readout detectors and some entanglement operations still require cryogenic conditions. The advantage is partly architectural: photons are natural carriers for quantum networking, which makes photon-based systems attractive for applications in which distributing quantum computation across nodes matters.
The scaling math for photonics differs from that of other systems. Because photons tend to disappear, a phenomenon called photon loss, hardware systems compensate by using large numbers, with error correction codes designed to tolerate losses. This creates a scaling challenge: million-qubit photonic systems may be achievable, but current systems are far from that scale. Fidelity also remains a significant issue.
Spin Qubits. Spin qubits are the new kid on the quantum block. They encode quantum information in the spin state of electrons or atomic nuclei inside solid-state materials (typically silicon or carbon). In addition to being the newest entrant in the hardware race, they are in some respects the most strategically interesting.
The reason is that most spin qubits are built in silicon, the foundation of the semiconductor industry. If spin qubit performance can be maintained as systems scale, they may eventually be manufactured using adapted versions of existing semiconductor fabrication infrastructure, a pathway to high-volume production that no other approach can credibly claim.
The technology is also compact. Spin qubits are small enough that a meaningful number could fit on a chip the size of a fingernail. This advantage has significant implications for data center integration compared with other systems that require rack-sized optical setups or large dilution refrigerators.
Current qubit counts are in the tens for spin qubit systems. Demonstrating that fidelity and coherence can be preserved at scale is the central open question. The engineering is hard, and the timeline is uncertain, but the physics are sound and the long-term potential is credible.
The Unsolved Challenge
The principal challenge facing all five hardware approaches—and quantum computing generally—is fault tolerance. Today’s systems are noisy. Errors accumulate during computation, and current rates are too high for most commercially valuable algorithms to run without producing unreliable results. Existing systems have enough capability to run certain specialized algorithms that can tolerate noise but not enough to reach fault tolerance.
The path to fault-tolerant computation runs through quantum error correction, a technique that uses many physical qubits to protect one logical qubit, the unit that performs the computation. (See Exhibit 3.) The required overhead is significant. Depending on the error correction code and the underlying error rate, a single logical qubit may require hundreds, or even thousands, of physical qubits for protection.
The developers of each hardware system are pursuing error correction in parallel with scaling. The race is not simply about which system can produce the most qubits, but which system can produce qubits that are good enough, as well as sufficiently numerous, that error correction becomes possible without requiring an impractical number of physical qubits to do it.
Modularity adds another unresolved dimension. As systems grow by linking multiple modules, performance across modules may differ from performance within a single module, with potential implications for fidelity, connectivity, and ultimately error-correction efficiency. Yet these effects are not consistently captured in today’s hardware specifications. The industry will need to develop standardized metrics that quantify the performance impact of modularity as systems scale.
On this dimension, the systems have genuinely different profiles. Trapped ions and neutral atoms have fidelity levels that make error correction more efficient—they need fewer physical qubits per logical qubit. Superconducting systems compensate for lower error correction with faster speed and scale. Spin qubits, if their fidelity profile holds up, could combine semiconductor manufacturability with competitive error correction efficiency. Photonics takes a different approach entirely, building error correction into its architecture from the start.
An important error correction milestone—demonstrating a logical qubit that outperforms the best physical qubit that it is built from—has only recently been crossed and only in controlled conditions. Moving on to practical fault-tolerant computation at commercially useful scale remains the defining quantum hardware challenge.
The Implications for Companies Today
Most organizations do not need to place a bet on a specific hardware system now, and doing so offers little advantage. But they do need to develop a working-level system-aware view of the quantum landscape, for three reasons.
Infrastructure decisions are not hardware-neutral. Different systems have different physical footprints, cooling requirements, control system complexity, and integration challenges with classical computing infrastructure. These factors affect capex and opex decisions such as whether to host a quantum system on-premise or run workloads on cloud-accessible hardware.
Vendor claims require a hardware lens. The quantum hardware market is competitive and, at times, promotional. Qubit count comparisons between systems are often misleading because raw qubit numbers do not account for fidelity, connectivity, or coherence time. A 1,000-qubit superconducting system and an 80-qubit trapped ion system cannot be compared apples-to-apples—and neither may be more capable for a given application.
The timeline depends on which system reaches fault tolerance first. Forecasts for commercially useful fault-tolerant quantum computers range from five to 20 years depending on which system is being tracked and the assumptions about engineering progress. Understanding which systems are closest to narrowing this range is essential for calibrating quantum strategy.
The quantum industry’s center of gravity is shifting. The debate has shifted from technology feasibility to hardware viability, specifically which architecture will define the commercial era and on what timeline. Following that debate requires a working understanding of the systems in play, not at the level of the underlying physics, but in terms of tradeoffs, maturity, and strategic implications for business.
The authors would like to thank David Shaw, chief analyst in quantum technologies at Global Quantum Intelligence (GQI), for his feedback on this article.
