The future of Compute at the OT Edge

EXECUTIVE BRIEF: Where is OT edge computing going.... find out here...

Why Superconducting and Quantum Compute Won’t Centralize Industrial Control

>100 GHz Josephson-junction switching speed of superconducting circuits

250–500 µs typical update cycle for multi-axis motion control and CNC interpolation loops

2 ms minimum round-trip latency to a data center 400 km away — before any processing occurs

1. The Centralization Trap

Every major leap in computing power triggers a predictable, well-rehearsed chorus from the tech industry’s leading evangelists. When cloud computing matured, we were told local hardware was obsolete. When early quantum and high-performance AI architectures emerged, the prediction was the same: the “edge” was dead, and total centralization was inevitable. Now, as superconducting computing advances from a theoretical breakthrough into a looming operational reality, the exact same IT-centric thesis is being dusted off.

The argument seems simple on paper: if a centralized, near-zero-resistance superconducting super-brain can process data at unfathomable speeds with massive parallel efficiency, why waste money putting computing capabilities at the local level?

The flaw in this logic is born from a fundamental misunderstanding of the physical world.

IT-centric thinkers inherently view an industrial facility—whether it is a high-speed packaging plant, a semiconductor fab, or a petrochemical refinery—as just another group of “clients” tapping into a network, no different than a web browser, a mobile app, or a database query. They look at a Programmable Logic Controller (PLC) or an edge gateway and see an underpowered computer that desperately needs to be optimized out of the architecture.

What they miss is the dirt, the torque, and the sheer liability of physical operations.

In the world of Operational Technology (OT), code doesn’t just manipulate pixels or update financial ledgers; it drives physical-world outputs. It controls massive pumps, high-inertia motors, and explosive chemical processes. Because of this, the industrial sector operates under an entirely different set of rules than mainstream software development.

The edge is not a temporary compromise born of weak computing power — it is a permanent requirement of physical reality.

Superconducting computing will undoubtedly transform high-level industrial optimization, enabling massive multi-variable predictive models, deep-learning vision training, and plant-wide thermodynamic twins that were previously impossible. However, the architectural pendulum will never swing entirely back to total centralization. The unyielding laws of physics regarding time and latency, combined with the absolute mandate of industrial risk mitigation and safe-state autonomy, act as an unbreakable anchor. The execution layer must remain at the edge.

2. The Technical Reality of Superconducting Compute (Setting the Boundary)

To understand why superconducting compute cannot simply swallow the industrial edge, we must first look at what the technology achieves—and what it physically requires to survive.

At its core, superconducting computing replaces traditional silicon-based transistors with superconducting materials and ultra-fast switching elements, such as Josephson junctions. In a standard semiconductor, electrical resistance generates heat. As you pack more transistors into a silicon chip to get more power, that heat scales exponentially, hitting a thermal wall that limits clock speeds and energy efficiency. Superconducting circuits bypass this entirely. Operating with near-zero electrical resistance, they can switch at frequencies exceeding 100 GHz—orders of magnitude faster than traditional commercial processors—while consuming a fraction of the internal power.

For heavy industrial applications, this represents a massive paradigm shift. A centralized superconducting super-brain will be capable of processing immense, highly non-linear datasets simultaneously. It can ingest data from tens of thousands of IoT sensors across an entire corporate fleet to run real-time, high-fidelity thermodynamic simulations, train deep-learning models for predictive maintenance, and orchestrate complex supply-chain-to-production loops on a macro scale. However, this staggering computational power comes with a severe physical catch: the necessity of complex cryogenic isolation.

The Cryogenic Isolation Ward

Superconductivity is a delicate state of matter. To achieve zero resistance, the materials used in these advanced processors must be cooled to extreme cryogenic temperatures. Depending on the architecture, this means operating near absolute zero—around -269°C using liquid helium—or, at the higher end of the spectrum for High-Temperature Superconductors (HTS), around -196°C using liquid nitrogen. Maintaining this environment requires massive, highly complex, and delicate cryogenic cooling infrastructure, including multi-stage compressor systems, vacuum-insulated lines, and specialized dilution refrigerators.

This creates an insurmountable form-factor mismatch with the reality of the industrial plant floor. An industrial edge device—whether it is a PLC, a variable frequency drive (VFD), or an edge gateway—must be built for survival. It is routinely bolted into tight, unventilated Motor Control Center (MCC) buckets, mounted on vibrating machinery, or shoved inside NEMA 4X enclosures exposed to ambient heat, dust, moisture, and intense electromagnetic interference (EMI).

The idea of scaling down a liquid-helium cryogenic cooling apparatus to fit into a standard DIN-rail assembly or a field junction box is a mechanical and economic non-starter. Furthermore, the parasitic energy required to run the cooling system vastly outweighs the efficiency gains of the chip unless it is aggregated at a massive, centralized scale. The economics and the power requirements simply do not scale to meet the physical environment constraints of OT.

Therefore, superconducting computing is bound by its own physics to remain isolated inside highly controlled, specialized data centers. It cannot travel down the wire to the plant floor. It can never be field-deployable; it can only be reached via a network. And it is that reliance on the network that brings us directly to the first breaking point of total centralization: the unyielding laws of time and physics.

3. Silicon and Beyond: The Edge Gets Denser, Not Centralized

While superconducting compute chases raw throughput inside cryogenic data centers, a parallel and arguably more consequential shift is happening in the materials and packaging of edge-native silicon itself: monolithic 3D (M3D) integration.

Traditional chip scaling improved performance by shrinking transistors laterally. That approach is hitting atomic-scale limits. M3D takes a different path—stacking multiple active device layers vertically, with interconnects separated by only nanometers rather than the millimeters of a traditional multi-chip PCB layout. Vertically stacking layers of silicon circuits lets chipmakers dramatically increase computing density and speed while reducing energy use, offering a route to extend Moore’s law without further transistor shrinkage.

The historical barrier has been thermal: building high-performance silicon typically requires temperatures near 1,000°C, but once a first circuit layer is complete, any additional layer must stay below roughly 400°C to avoid damaging what’s already built. Researchers have recently demonstrated a process that gets around this by transferring ultrathin single-crystalline silicon layers onto finished circuits rather than growing them in place—enabling true layer-by-layer stacking without cooking the layers below.

Parallel work is pushing this further with alternative materials—atomically-thin 2D semiconductors like molybdenum disulfide, and ferroelectric and oxide-based transistors—stacked directly with memory in the same 3D structure. This architecture eliminates costly data transfers to distant memory arrays, enabling near-sensor inference and low-latency decision-making with substantially reduced energy consumption.

Superconducting compute gets faster by getting colder and more centralized. 3D-stacked silicon gets faster by getting denser and staying local.

For the industrial edge, this matters for a specific reason: it is the inverse of superconducting compute’s constraint. M3D and 2D-material stacking gets faster by getting denser and staying local—shorter wire lengths cut parasitic capacitance and boost bandwidth between logic and memory, at ambient temperature, in a footprint no bigger than today’s PLC or edge gateway. That means the next decade’s edge devices will not just hold the line against centralization—they will get materially more capable at running deterministic control loops and real-time inference, without requiring a single watt of cryogenic overhead or a single microsecond of network round-trip.

This is the real story of the hybrid hierarchy: while the superconducting core gets smarter about macro optimization, 3D-stacked silicon and 2D materials make the edge smarter about micro execution—independently, and in parallel.

4. The Physics Boundaries (Temporal Thresholds & Determinism)

Even if we completely ignore the cryogenic hardware constraints of superconducting computing and assume the centralized super-brain has an internal processing latency of exactly zero seconds, total centralization still collides with a hard, unyielding wall: the speed of light.

In the IT world, latency is measured in human perception. If a webpage takes 100 milliseconds to load, or a cloud-hosted video buffers for a fraction of a second, the user experience is slightly degraded, but the system functions. In the OT world, latency is measured against mechanical and electrical velocity. Here, a delay of a single millisecond is not an inconvenience; it is a stability threshold.

To understand why, we have to look at the math of data propagation. Information traveling through fiber-optic networks does not move instantaneously; it travels at the speed of light in glass, which is roughly two-thirds its speed in a vacuum. This introduces a hard physical propagation delay of approximately 5 microseconds per kilometer of round-trip distance.

If a centralized superconducting data center is located just 400 kilometers (about 250 miles) away from a manufacturing plant, the physical transit time for a signal to travel from a sensor to the data center and back to an actuator is a minimum of 2 milliseconds. This 2-millisecond physical floor immediately disqualifies centralized compute from managing high-performance industrial control loops:

•        High-Speed Motion Control: Modern multi-axis vector drives, CNC machines, and robotic interpolation loops routinely operate on update cycles of 250 to 500 microseconds.

•        Precision Execution: In high-speed packaging, printing, or web-handling applications, material moves at velocities where a 2-millisecond blind spot translates to inches of unguided product, resulting in catastrophic registration errors, web breaks, or mechanical collisions.

•        Vibration and Pulse Monitoring: Advanced turbo-machinery diagnostics and high-speed electrical protection loops require microsecond-level sampling and immediate deterministic execution to prevent catastrophic mechanical failure.

This is why centralization will never work for real-time operations, despite the power of the underlying compute technology.

Furthermore, raw speed is only half the battle. In automation, determinism—the absolute predictability of when a decision will arrive—is far more critical than average throughput.

Wide Area Networks (WANs) are inherently non-deterministic. When data leaves a plant floor and travels across external network infrastructure, it is subjected to packet routing variances, micro-burst congestion at commercial switches, and unpredictable serialization overhead. This introduces jitter variance in latency.

In a closed-loop control system, jitter kills stability. The system begins to hunt, oscillate, and will ultimately tear itself apart mechanically.

Superconducting computing can drastically reduce the thinking time, but it cannot shrink the physical distance between the data center and the factory floor. Because the laws of physics prevent a remote network from ever matching the deterministic, sub-millisecond thresholds of a local industrial bus, the localized execution brain remains an absolute necessity.

5. The Risk Bound (Autonomy, Survivability, and Safe State Isolation)

While the laws of physics establish that centralized control is technically unviable for high-speed loops, industrial risk management dictates that it is an absolute business and safety non-starter. In Operational Technology, the ultimate metric isn’t computational efficiency; it is survivability.

To understand the core of this argument, one must apply the fundamental “What If the Pipe Breaks?” test.

In the IT world, network disconnection or a cloud data center outage is an expensive inconvenience. Video streams buffer, emails are delayed, or an e-commerce checkout fails. The system state is saved in a database, and operations resume seamlessly once the connection is restored.

In the OT world, a loss of connectivity while a physical process is running is a high-stakes hazard. If a petrochemical refinery, an offshore drilling rig, or a water treatment plant loses its connection to a centralized “super-brain” for even a few consecutive seconds, the consequences can be catastrophic. Without a local brain to modulate control, high-pressure pumps can over-pressurize vessels, chemical reactions can experience thermal runaway, and heavy material handling systems can violently collide.

Because a network connection to the outside world can always be severed—whether by a backhoe cutting a fiber line, a localized power grid failure, or a targeted cyberattack—the plant floor demands Air-Gapped Autonomy. Every localized cell, machine, and production line must possess enough native intelligence to execute two critical strategies completely independent of an external network:

•        Graceful Degradation: The ability to drop from a hyper-optimized “cloud-guided” state into a safe, localized “steady-state” production mode, keeping the plant running safely at a baseline level until connectivity returns.

•        Emergency Shutdown (ESD): If steady-state operation is impossible, the local controller must have the independent compute power to execute a highly complex, sequential shutdown sequence to bring the physical assets to a verified safe state.

This requirement for localized autonomy is not just an engineering preference; it is codified by global functional safety standards.

Regulatory frameworks such as IEC 61508 (functional safety of electrical, electronic, programmable electronic safety-related systems) and IEC 61511 (safety instrumented systems for the process industry) strictly dictate the design of Safety Instrumented Systems (SIS). These safety loops—consisting of dedicated sensors, safety controllers, and final control elements like emergency trip valves—must be completely segregated from standard control networks, highly reliable, and physically localized.

Liability and regulatory compliance demand that the ultimate authority over physical output remains hardwired to a local, immutable execution brain.

Entrusting a life-safety trip function, or even the basic interlocking logic that prevents a machine from crushing an operator, to a remote, centralized infrastructure—no matter how fast or advanced that infrastructure may be—violates the foundational tenets of industrial safety isolation.

6. Back to the Future – Better than the Past (Hybrid Hierarchy)

Because centralized superconducting compute cannot overcome the speed of light, because industrial operations cannot compromise on localized survivability, and because 3D-stacked silicon and emerging 2D materials are making local execution more capable rather than less, the future of Operational Technology will not be a story of centralization replacing the edge. Instead, it will be a story of a hybrid hierarchy—a clean division of labor where each technology does exactly what it is architecturally built to do.

Superconducting computing will not shrink or minimize edge controllers like PLCs, PACs, and industrial PCs. Rather, it will act as the ultimate high-level orchestrator, redefining the boundaries of decision quality while leaving execution to the edge. This future industrial architecture will split cleanly along two lines of decision-making:

The Centralized Superconducting Layer (Strategic & Probabilistic)

Located in specialized, cryogenically cooled data centers, this layer will ingest massive, aggregated datasets from thousands of assets across global operations. It will specialize in heavy, non-linear, and mathematically complex workloads that require raw processing power over sub-millisecond timing:

•        Fleet-Wide Macro Optimization: Analyzing asset degradation, thermal profiles, and energy market pricing in real time to calculate the most profitable and energy-efficient operating parameters for an entire enterprise.

•        Deep Learning & Generative Vision Training: Training massive machine vision and predictive maintenance models that require immense parallel processing, then compiling those models into lightweight profiles.

•        High-Fidelity Digital Twins: Running live, multi-physics thermodynamic simulations of entire chemical processes or manufacturing lines to predict long-term mechanical failures or product defects before they occur.

The Localized Edge Layer (Tactical & Deterministic)

Located on the harsh plant floor, embedded in MCC buckets, or mounted directly to machine skids, this layer will remain highly computed, ruggedized, and completely autonomous. Increasingly built on 3D-stacked silicon and 2D-material logic-memory architectures, it will receive high-level strategic directives, optimized setpoints, and compiled AI models from the superconducting cloud, and then execute them locally:

•        Real-Time Edge AI Inference: Running pre-trained vision or vibration models locally at the microsecond level to catch defects or anomalies instantly without waiting for a network round-trip.

•        Deterministic Control Loops: Handling sub-millisecond, closed-loop vector drive adjustments, valve positioning, and motion synchronization that keep the physical machine stable.

•        Immutable Safety & Autonomy: Ensuring that if the network connection to the centralized super-brain drops, the local PLC seamlessly maintains steady-state production or safely executes an emergency shutdown.

Superconducting compute will provide a massive competitive advantage in large-scale data centers where the cost and power requirements for cooling make sense at scale. In fact, it can drastically reduce the overall footprint required for equivalent capability over current semiconductor technologies.

Closing Perspective

The pendulum of computing architecture will always swing. For decades, the broader tech industry has watched this cyclical rhythm play out: from centralized mainframes to distributed desktop PCs, and from the hyper-centralized cloud back down to the containerized mobile edge. Whenever a disruptive leap in raw processing power emerges, it is entirely natural for technologists to assume that the pendulum will inevitably swing all the way back to the center.

But when that pendulum enters the arena of Operational Technology, it hits a hard stop.

Superconducting compute represents a truly breathtaking frontier for industrial intelligence. It promises to unlock optimization models and predictive capabilities that will make today’s cloud-driven operations look prehistoric. Yet, as majestic as a centralized, cryogenically cooled super-brain may be, it cannot rewrite the fundamental laws of nature. It cannot force a photon to travel through fiber optics any faster, it cannot eliminate the chaotic jitter of a wide area network, and it cannot absorb the severe legal and human liability of a physical asset left blind by a severed network cable.

The edge is the critical last mile where digital logic converts into kinetic force, heat, and pressure.

Instead, superconducting compute will thrive as an essential part of a broader ecosystem. By hosting intensive analytical and AI engines within centralized data centers, we can use this monumental power to orchestrate and optimize the operational edge without ever compromising its autonomy. Meanwhile, advances in 3D-stacked silicon and emerging 2D materials ensure that the edge itself keeps getting more capable, more efficient, and more autonomous on its own terms.

The industrial world is uniquely grounded by physical reality. It is a permanent architectural necessity mandated by physics and locked in place by functional safety. As the industry looks toward the horizon of next-generation computing, the smart play is clear: we should not look to replace the edge with the cloud but rather use the monumental power of centralized superconducting systems—and the parallel gains in dense, local silicon—to make our localized brains smarter, more resilient, and more autonomous than ever before.