The King Arthur Class: A Thermodynamic and Biomimetic Framework for Next-Generation Hyper-Scale Datacenters

Patent Status: U.S. Provisional Patent Application No. 64/046,446 (Patent Pending)

Executive Summary

Modern datacenters are some of the most carefully engineered facilities in the world. Power distribution is modeled in detail, network fabrics operate at microsecond speeds, and cooling systems are tuned with extreme precision. Even so, most facilities still rely on one basic assumption that has barely changed: the rectangular building shape borrowed from traditional warehouse design.

That assumption is becoming harder to defend as artificial intelligence and high-performance computing push rack densities far beyond earlier limits. In many legacy facilities, the building itself works against efficient operation. Rectangular layouts encourage thermal mixing, drive up water use, and leave power capacity stranded in ways that become more costly as scale increases.

The King Arthur Class proposes a different approach. It describes a new category of datacenter architecture built around radial symmetry, circular organization, and vertically stacked cylindrical forms. Drawing on the passive environmental control seen in Macrotermes termite mounds and the adaptive behavior of the Thaumoctopus mimicus, or mimic octopus, the framework shifts the focus from optimizing systems inside a fixed box to using geometry itself as part of the solution.

When the form of the building works with the movement of heat, air, and utilities, more of the burden can be carried by durable architectural design instead of constantly active mechanical systems. In practical terms, that means a path toward lower PUE, near-zero WUE, more efficient radial utility routing, and a much smaller land footprint through vertical expansion.

1. The Geometric Paradox of the Rectangular Box

The standard, flat, rectangular datacenter design is not a product of deliberate thermal optimization; it is an artifact of convenience. While internal equipment (such as variable-speed fans, aisle containment, and economizers) has been completely re-engineered across multiple generations, the floor plate has stayed flat and the walls have stayed straight.

This persistence of a familiar floor plate creates several problems at hyper-scale. In a rectangular room, cooled air loses effectiveness as it travels across long parallel rows, so the racks farthest from the source often see warmer intake temperatures than intended. Hot exhaust can recirculate, chilled air can bypass the servers altogether, and temperature layers can form vertically inside the room. To compensate, operators typically overcool at the source and drive fans harder, which contributes to the higher PUE ranges still common in older facilities. The same geometry also stretches power, cooling, and network paths across long horizontal distances, increasing copper use, resistance losses, voltage drop, and latency constraints. At the campus level, continued horizontal expansion consumes large amounts of land and brings higher permitting costs, real estate pressure, and community pushback.

2. Biomimetic Anchor I: The Macrotermes Termite Principle

One useful model comes from the African savanna, where Macrotermes termites build mounds more than 30 feet tall that keep internal temperatures within a narrow range even as outdoor conditions shift dramatically between day and night. They do this without conventional air conditioning, fans, or ongoing water consumption. The mound works because its shape and internal passages create a passive thermal system that guides airflow instead of forcing it.

The structure is arranged around a central chimney, with chambers and passageways distributed in concentric patterns. As heat builds inside the mound, warmer air rises through the central shaft and exits near the top. That upward movement draws cooler air inward through lower channels connected to the ground. The mound does not overpower heat with machinery. It uses form to guide heat and airflow in a stable, efficient cycle.

That principle sits at the center of the King Arthur Class. The core idea is that geometry is not just a container for infrastructure. It is an active efficiency multiplier.

3. The King Arthur Class: Architectural Blueprint

The King Arthur Class applies that radial logic to hyper-scale computing. The name is intentional. Much like King Arthur’s round table symbolized equal standing, this design places power, cooling, networking, and operations around a shared center rather than along a long linear edge. The result is a floor plate organized as a set of concentric thermal zones that shorten and equalize supply and return paths for every rack.

The Concentric Horizontal Layout

Instead of arranging equipment in long straight rows, the design divides each floor into four rings. At the center is a hot aisle core that collects server exhaust in a dense vertical plenum, typically running between 95°F and 113°F. Around that core sits a circular ring of 12 to 16 radial rack pods designed for high-density compute loads, often in the 30 kW to 50+ kW range per rack. Outside the racks is a cold air distribution ring that feeds conditioned air inward from all directions at roughly 64°F to 77°F. At the outer perimeter, a dedicated equipment ring houses CRAH units, CDUs, and primary heat exchangers.

This arrangement means no rack sits meaningfully farther from the infrastructure core than another. Supply and return paths stay short and balanced. The layout also makes it much harder for hot exhaust to re-enter server intakes or for chilled air to bypass the compute load. In effect, several persistent airflow problems in rectangular facilities are reduced not by more controls, but by the physical organization of the floor itself.

The Cylindrical Skyscraper Form

The same logic extends vertically. The concept envisions stacking these circular floors across roughly 12 to 20 stories, reaching heights of about 197 to 328 feet with floor diameters between 98 and 131 feet. When the hot aisle cores align from floor to floor, they create a continuous building-scale chimney.

That chimney makes use of the stack effect, the natural tendency of warm, lower-density air to rise. Instead of relying entirely on mechanical fan power to move heat upward, the building gains a convective assist that becomes stronger as the temperature difference between exhaust air and outdoor conditions increases. As server utilization rises, the architecture naturally supports the heat rejection process rather than fighting it.

4. The Hydrological Emergency and Waterless-by-Design Strategies

The industry is also facing a growing water problem. In Texas alone, more than 400 datacenters used an estimated 25 billion gallons of water in 2025. If grid demand rises to the projected 77,965 megawatts by 2030 and water intensity remains similar, annual consumption could climb to about 161 billion gallons, or roughly 2.7% of the state’s full water budget. That level of industrial demand is especially difficult to justify in regions already dealing with recurring drought.

Traditional facilities often achieve efficiency through evaporative cooling towers, which depend on water loss to the atmosphere as part of the heat rejection process. The King Arthur Class responds by pairing its more efficient geometry with closed-loop and dry-cooling strategies designed to cut Water Usage Effectiveness from a legacy average near 2.0 L/kWh to a level that is functionally close to zero.

Closed-Loop Liquid and Immersion Cooling Systems

In this model, water stops being a disposable operating input and becomes a long-lived infrastructure medium. Direct-to-chip cold plate systems circulate coolant through microchannel copper plates mounted to CPUs and GPUs, typically operating with warm fluid in the 65°F to 85°F range. These systems can support about 30 to 50 kW per rack and can lower PUE to roughly 1.04 while removing the evaporative burden and relying on dry coolers for more than 6,000 hours each year. Rear door heat exchangers offer a drop-in retrofit path for similar rack densities by intercepting hot exhaust air before it enters the room, and when paired with external dry coolers they can cut direct water use by about 40% to 60%. For higher density environments, single-phase immersion places the server hardware in a sealed dielectric fluid bath that circulates through an external heat exchanger, supporting 100+ kW per rack, removing the need for server fans, and pushing PUE toward 1.02 with near-zero direct water use. Two-phase immersion takes that further by using a low-boiling dielectric fluid that vaporizes on contact with hot components, condenses, and returns in a passive cycle, allowing maximum-density computing with PUE below 1.02 and no direct cooling water consumption.

Air-Cooled Chillers and Adiabatic Pre-Cooling

In spaces where liquid cooling is not yet universal, the design can still rely on sensible air-cooled chillers. These systems are less efficient in very hot climates because they are limited by dry-bulb conditions rather than wet-bulb conditions, but the radial layout helps offset some of that penalty by reducing internal airflow resistance and keeping distribution paths more uniform.

For extreme heat, adiabatic pre-cooling can be used as a selective hybrid measure rather than a constant dependency. Water is applied to porous media only when outdoor temperatures pass a high threshold, such as 75°F dry-bulb. That step can cool intake air by about 10°F to 20°F before it reaches the refrigerant coils. Compared with open cooling towers, this approach can reduce water use by 70% to 85%, limiting consumption to peak summer periods while allowing fully dry operation for more than 5,000 hours each year.

5. Biomimetic Anchor II: The Mimic Octopus Operational Strategy

If the termite mound provides the structural model for managing heat and airflow, the Thaumoctopus mimicus, or mimic octopus, offers a useful model for operating the facility under changing compute conditions.

The mimic octopus does not depend on a single fixed defense. It shifts its appearance and posture based on context, presenting itself differently depending on the threat it faces. That kind of real-time adaptation provides a practical analogy for datacenter control systems that need to respond to changing workloads, temperatures, and power conditions without staying locked in one static operating mode.

Many older facilities still run as though peak conditions are constant. Cooling systems stay aggressive overnight even when loads are low, and electrical reserves remain fixed around worst-case assumptions. The King Arthur Class replaces that with a context-aware operating model built around four coordinated phases.

The 4-Phase Lifecycle Architecture

Phase 1: Sensor Baselines & Observation

The facility is heavily instrumented across all mechanical, electrical, and IT layers. Continuous telemetry tracks local thermal bounds, air pressure differentials within the Zone 1 chimney, power consumption profiles per rack, and pump/fan performance curves. This serves as the sensory foundation, mapping out the facility's real-time physical state.

Phase 2: Posture Response Library

First, the building establishes a detailed baseline through instrumentation across mechanical, electrical, and IT systems, tracking local temperatures, pressure differentials in the central chimney, rack-level power use, and pump and fan performance. Second, the control layer maintains a library of response profiles for recurring conditions. Those profiles can include an AI training posture that accelerates localized fluid flow and widens the cold-air perimeter path, an arid night mode that shifts into fresh-air economization and raises server inlet thresholds toward ASHRAE Class A3 limits, and a peak utility preservation mode that power-caps selected loads and reallocates stranded electrical capacity to the most active nodes.

Phase 3: Automated Transition & AI Control Loops

Third, transitions between those modes are automated through AI-driven control loops and digital twin simulations that continuously model the thermal behavior of the building using server telemetry and weather inputs. If an AI training job ramps up, the facility can shift posture before the full heat wave arrives. This coordination between compute scheduling and thermal management can improve energy performance by roughly 20% to 40% over more static operating baselines.

Phase 4: Continuous Structural Adaptation

Fourth, the system is designed to keep adapting as hardware changes. Legacy air-cooled areas can be upgraded with rear-door exchangers or replaced with immersion systems, and the radial infrastructure core makes those upgrades easier to integrate without forcing a broader redesign of the building itself.

6. The Cooling Debt Framework

To connect real-time AI workload scheduling directly to physical infrastructure costs, the operational software tracks an immutable metric: Cooling Debt.

In this framework, every compute node and rack carries an implied cooling debt, which can be described as a function of power density, fluid resistance, distance from the radial core, and water intensity. Legacy air-cooled servers operating in less efficient positions accumulate more debt with each compute cycle, while racks using direct-to-chip or immersion cooling close to the core carry much less.

Racks filled with uncontained, air-cooled legacy servers operating far from the primary cooling inputs accumulate a high cooling debt score per compute cycle. Conversely, racks utilizing direct-to-chip or immersion methods positioned close to the central infrastructure core maintain a negligible cooling debt profile.

The facility’s workload orchestration platform utilizes this framework to route intense computing tasks dynamically:

The workload orchestration layer uses that score to decide where demanding jobs should run. If a high-density AI workload arrives during a hot afternoon, the system routes it toward the internal rings with the lowest cooling debt. If a rack crosses a debt threshold, protective actions can follow automatically, such as migrating the job to a cooler radial pod, applying a local power cap, or adjusting cold-plate fluid flow. This keeps infrastructure decisions tied directly to physical efficiency instead of allowing hidden thermal penalties to accumulate over time.

7. Macro-Scale Infrastructure Integration

The sustainability model also extends beyond the data hall. Photovoltaic systems mounted on the cylindrical facade, roof, and surrounding support structures can generate electricity with essentially no operating water draw, far below the lifecycle water intensity associated with traditional thermoelectric coal generation. At the same time, high-grade waste heat from dense server loops does not have to be treated as a disposable byproduct. It can be exported to district heating systems or nearby industrial processes, and in some cases it can even drive on-site absorption chilling for office support loads. In that setup, waste heat becomes an energy asset rather than a constant penalty.

8. Proven Technical Benchmarks (Industry Case Studies)

The technical ideas behind the King Arthur Class are not speculative in the sense of having no precedent. Several large-scale facilities already demonstrate pieces of the model in operation. Google’s Hamina site in Finland uses cold seawater from the Gulf of Finland as a heat sink and has reached a PUE of about 1.10 while keeping WUE near zero for cooling. Equinix’s SY4 and SY5 sites in Sydney combine sensible air-cooled chillers with targeted adiabatic pre-cooling and have reportedly reduced water use by around 80% compared with conventional open-loop towers. Iron Mountain’s Phoenix facility uses high-efficiency outside-air filtration and elevated inlet temperature allowances to operate around 6,000 hours each year with direct-air economization, reaching a WUE near 0.20 L/kWh, an 89% reduction from a regional baseline of 1.80 L/kWh. Nautilus Data Technologies has shown that closed-loop river-based heat rejection can support a PUE around 1.15 with no evaporative cooling water use, while Digital Realty in Singapore has paired direct-to-chip cooling for dense AI workloads with reclaimed municipal water so that facility cooling places no demand on potable supplies.

Conclusion: Geometry as Destiny

The industry’s dependence on large volumes of evaporated water and heavy mechanical overhead is not an unavoidable law of datacenter design. It is, to a significant extent, a consequence of relying on rectangular geometry long after the operational context has changed. Incremental improvements inside that format are starting to produce smaller returns.

The King Arthur Class changes that premise by treating the building itself as part of the thermal and operational strategy. When form is aligned with physics, the structure can do more of the work of managing heat, routing utilities, and supporting vertical growth.

Rethinking the foundational layout of hyper-scale computing is no longer just a theoretical exercise. It is becoming an engineering requirement. By combining radial geometry, biomimetic design principles, and advanced closed-loop cooling strategies, the King Arthur Class presents a credible path toward dense, resilient, and water-conscious digital infrastructure.