First Principles: The Thermodynamic Blind Spot of Civilization 战略学术白皮书 · Strategic Academic Whitepaper · 2026
An absolute, inescapable physical fact: 1 Watt of compute = 1 Watt of heat. 100% conversion. No efficiency improvement can alter this value.
Power plants penalize 30-50% of energy as waste heat—the cost of generation (Carnot efficiency). But in a data center, every single joule counts twice: one joule comes from the grid to perform work, and one joule is returned to the atmosphere as waste heat. Compute is a machine that perfectly irons electrical energy into heat. No exceptions. No offsets. No carbon-capture-style closed loops. This is a thermodynamic trap that our civilization has yet to realize.
Magnitude Leap in Heat Flux Density: From “Diffused Waste Heat” to “Point Thermal Weaponry”
Reviewing the evolution of human waste heat emissions reveals a disruptive, order-of-magnitude leap in heat flux density: 第一性原理：热力学在文明尺度上的盲区 / First Principles: The Thermodynamic Blind Spot of Civilization
Density 典范 / Paradigm
Pre-industrial
Biological metabolism + Firewood ~100 W/m²
Village hearth smoke
Steam
Factory chimneys ~500 W/m²
Manchester textile mills
Electricity
Thermal power plant cooling towers ~1,000 W/m²
A 3km² power station
Early Computing
Data centers ~500 W/m²
Server racks in the 2000s
H100 Clusters ~2,500 W/m² 1GW / 0.4km² 2030+
H400 Clusters ~5,000 - 7,000 W/m² 2.5GW / 0.5km²
For the first time in industrial history, humanity is generating a heat flux density that exceeds solar radiation on the Earth’s surface.
Peak solar radiation reaching the surface is roughly 1,000 W/m². An H400 cluster outputs 5 to 7 times that per square meter. This means you have created a localized “second sun”—except it radiates pure infrared heat without light.
The reason why the demand for “cold energy” is skyrocketing finds its answer right in these numbers. It is not merely because compute has increased, but because the heat flux density has breached the physical limits that natural dissipation systems can absorb. The atmosphere’s natural convective cooling capacity is roughly 100-300 W/m². When your heat source exceeds this threshold, the heat can no longer be naturally diffused, leading to thermal accumulation: localized atmospheric heating → rising air currents → influx of surrounding humid air → low-pressure vortex formation → extreme precipitation events.
an “Energy Source” — It is a Scarcity under the Second Law of Thermodynamics
Most people understand “cold energy” as opening a refrigerator. In physics, there is no such thing as “cold energy”—there is only thermal capacity space: how many joules a substance can absorb without a significant rise in temperature.
On Earth, the supply of cold energy primarily comes from three sources:
- 大气冷能（最易获取，容量最小） / Atmospheric Cold Energy (Easiest to access, smallest capacity)
Air’s specific heat capacity is ~1 kJ/kg·K. A cubic meter of air weighs 1.2 kg; a 10°C temperature rise can only absorb 12 kJ of heat. The higher the latitude, the lower the ambient temperature → more room for temperature rise to absorb heat. This explains why Stockholm (annual avg 7°C) inherently has a PUE 0.3-0.4 lower than Singapore (annual avg 28°C). But air’s thermal capacity is too small—a 1GW cluster requires ~50,000 cubic meters of air per second to maintain a 20°C temperature rise. This is equivalent to emptying a standard swimming pool of air every second. This is the scale of a wind tunnel, not a fan. 2. 水冷能（最高效，但受地理位置限制） / Hydro-Cold Energy (Highly efficient, but geographically constrained)
Water’s thermal capacity is 4,000 times that of air (4.2 vs 1.0 kJ/kg·K). This is why data centers desperately seek water. A 1GW cluster requires: 4,348 tons of water per hour (at a 20°C temperature rise)—equivalent to a DN800 pipe at full flow, or 1.2 tons of water per second (the flow of a medium-sized river). Water consumption is even more terrifying: evaporative cooling is the most water-efficient route—yet a 1GW cluster evaporates 36,000 tons of water daily into the atmosphere. This is the daily water consumption of a city of 50,000 people, entirely converted into steam. 3. 深海冷能（最大储量，最难提取） / Deep-Sea Cold Energy (Largest reservoir, hardest to extract)
The deep sea at 1,000m is constantly at 2-4°C. This is a near-infinite thermal sink—the thermal capacity of the global deep ocean is over 1,000 times that of the atmosphere. However, this requires deep-water pipes near continental shelves, SWAC (Sea Water Air Conditioning) infrastructure, and severe proximity constraints—clusters must be within 5-10 km of the coast. The future geographic layout of global compute will be dictated by its distance to deep-sea cold energy. 第一性原理：热力学在文明尺度上的盲区 / First Principles: The Thermodynamic Blind Spot of Civilization
Cold Energy: Who Holds the Location Rights for Future Compute
According to cold energy abundance and physical dissipation capacity, the global tiers for compute location selection are mapped below: 第一性原理：热力学在文明尺度上的盲区 / First Principles: The Thermodynamic Blind Spot of Civilization
Location
Avg Temp
Cooling
Physical Constraints
S-Tier: Norwegian Coastline 4 - 7°C
8-10 months
Fjords reach deep water (300m+), ultra-cheap hydro, natural cold water, and North Sea fiber.
S-Tier: Iceland 4 - 5°C
7-8 months
Mid-Atlantic Ridge, rich geothermal + hydro, 100% green power, perfectly matches SWAC architecture.
S-Tier: Southern Chile 5 - 8°C
6-7 months
Humboldt Current + Chile Trench (5km), closest to Antarctic cold source, steep Andean hydro.
S-Tier: Eastern Taiwan 18 - 22°C
0 (Deep water usable)
200m depth just off Hualien, Kuroshio current yields deep- sea pipe potential, near Asian markets.
A-Tier: Siberia -8°C ~ -3°C
9-10 months
Extreme continental cold source; however, poor fiber backbones, weak infra, high political risk.
A-Tier: Mohe/Genhe (China) -2°C ~ -4°C
7-8 months
China’s absolute cold pole; but current local grid capacity is limited, requires new UHV lines.
A-Tier: Tibetan Plateau -2°C ~ -5°C
7-9 months
Year-round low temp; however, high altitude (4500m+) thins air, reducing convection efficiency.
A-Tier: Yukon/NWT (Canada) -4°C
8-10 months
North American cold zone; severely constrained by poor fiber access and sparse tech supply chains.
B-Tier: Central Finland/ Sweden 2 - 6°C
7-9 months
Excellent compromise between cold source and market. Proven by mature GAFAM deployments.
B-Tier: Scotland/Faroe Islands 7 - 9°C
5-7 months
Powerful North Sea wind integration combined with cold marine currents, great for coastal clusters. 第一性原理：热力学在文明尺度上的盲区 / First Principles: The Thermodynamic Blind Spot of Civilization
Location
Avg Temp
Cooling
Physical Constraints
B-Tier: Mongolia -1°C
8 months
Bordering China with low cross-border fiber costs; but restricted by water and power capacities.
B-Tier: Hokkaido (Japan) 6 - 9°C
5-6 months
Deep water available, superb international subsea cable node; but carries severe seismic risks.
The brutal conclusion: There are no more than 15 locations on Earth suitable for hosting compute clusters above 1GW without triggering localized climate disruption. Yet global demand points toward 30 to 50 gigawatt-scale clusters. This is a physical supply-demand deficit—one that money cannot solve.
Path Humanity Has Never Walked
If “cold energy” is to be transported across regions, we must rigorously calculate the penalty cost on the thermodynamic ledger: 方案A：水管运冷水（最现实的局部解） / Option A: Water Pipes for Cold Water (The Most Realistic Local Solution)
Pumping 4°C water from the deep sea to inland clusters, discharging it after it warms to 24°C: A 100MW cluster requires a DN800 pipe at 2.5 m/s (4,348 tons/hour). A 1GW cluster requires a DN1600 pipe (1.6m diameter) at 3 m/s (43,480 tons/ hour). Pumping losses are severe: for every 100 km of transport or 100m of elevation gain, an additional ~3MW of pumping power is consumed. Every 100 km of cold water transport adds a 3-5% power penalty. The economic limit is 200-300 km; beyond this, pumping power completely devours the “value” of the cold energy. Thus, cold water pipelines cannot cross continents. This remains a localized solution, deployed within 30-50 km of coastlines.
Storage (Engineeringly Absurd, Economically Plausible)
Shipping ice from the poles to the tropics? This sounds fantastical, but in the 19th century, shipping natural ice from Boston to Calcutta was a highly profitable business. Let’s audit the thermodynamic ledger: 1 ton of ice carries 334 MJ of latent heat ≈ 93 kWh of cooling capacity. A 100,000-ton cargo ship conveys 9.3 million kWh of cold energy—equivalent to 1.86 days of cooling for a 100MW cluster. Arctic-to-Shanghai freight costs roughly ¥50-100/ton. Cold energy cost ≈ ¥50 / 93kWh = ¥0.54/ kWh of cooling capacity. This sits in the exact same order of magnitude as industrial grid electricity tariffs (¥0.6-0.8/ kWh). While engineeringly absurd, if power and cooling water collapse at specific demand nodes, a closed-loop shipping cycle of Phase Change Materials (PCM)—“Freeze in the Arctic, melt in Shanghai, return empty”—is economically viable at the margin.
Energy Source (The Only Scalable, Ultimate Solution)
Just as the steam age concentrated factories around coal mines and the textile industry gathered around hydro hubs, the reshaping of compute geography follows an irreversible physical law. We see humanity executing this now: Oracle in Iceland (cold energy + geothermal power); Microsoft’s “Project Natick” (undersea data centers cooled by the deep ocean); Google in Finland (Baltic Sea cold water); China’s “East Data, West Compute” project. The planetary geography of 2040 will see major compute clusters residing inside the Arctic Circle, on the Tibetan Plateau, in the deep ocean, or underground on the Moon (permanently at -20°C); while human users tap into them via fiber-optic cables from metropolitan areas worldwide. It is not humanity moving to compute, but compute moving to cold energy.
Extreme Climate: “Après moi, le déluge” is No Metaphor
We must confront the exact physical causal chain of point-source thermal emissions superimposing onto regional climate systems: 第一性原理：热力学在文明尺度上的盲区 / First Principles: The Thermodynamic Blind Spot of Civilization
H400 Compute Cluster (100MW - 1GW Concentrated High-Density Heat Source) ↓
Localized Heat Flux Density reaches 1,000 - 2,500 W/m² (1-3x peak solar radiation on Earth’s surface) ↓
A persistent atmospheric “Heat Bubble” forms 50 - 200m above the terrain ↓
Heat bubble is forced upward → Adiabatic expansion → Violently draws in surrounding humid air → Triggers the Clausius-Clapeyron relation ↓
Saturation vapor pressure rises by 7% per 1°C → Absolute water vapor concentration in lower atmosphere surges by 10 - 20% ↓
Regional Convective Available Potential Energy (CAPE) accumulates and spikes by 30 - 50% ↓
Downwind 30 - 100km: The frequency of extreme precipitation shocks compresses from a 5-year return period to a 2-year interval ↓
Single extreme rainstorm yields surge violently from 200mm / 3 days to over 400mm / 3 days ↓
Localized surface runoff instantly overleaps the structural flood-control design capacity of canals → Severe flooding
This is by no means a speculative model forecast; it is an active physical reality. Shanghai’s urban heat island has already amplified annual precipitation in its downwind corridor (the Suzhou-Wuxi-Changzhou corridor) by 15%. Superimpose 10-20GW of raw compute thermal output here, and this artificial increment will double. The torrential rain profile of the Yangtze River Delta in the future will not be a natural fluctuation—it will be a thermodynamic resonance of compute clusters superimposed onto urban heat infrastructure. Every micro-joule released by an individual inference request converges at the cluster scale, magnifies at the regional level, and ultimately manifests as a raging crest in the city’s drainage channels.
Cold Energy on Mars is Infinite, But Useless
Mars has an average surface temperature of -63°C, theoretically offering a massive cold sink. Yet a fatal thermodynamic constraint applies: Mars’s atmospheric density is only 1% of Earth’s, which drives the convective heat transfer coefficient (h) close to zero. Compute clusters on Mars cannot rely on air convection; they must depend entirely on vacuum thermal radiation.
According to the Stefan-Boltzmann Law: P = εσA(T⁴ - T₀⁴)
To dissipate just 1kW from a server on the Martian surface into the -63°C night sky requires roughly 2-3 m² of high- emissivity radiator panels. Extrapolating this, a 100MW cluster demands a massive 200,000 - 300,000 m² radiant array (equivalent to ~30 standard football fields). On Mars, cold energy is free, but the physical surface area required for heat exchange is exceptionally expensive. This is the exact inverse of Earth: Earth offers free air and water for convection but has a tightly constrained regional thermal sink; Mars lacks convection, but its radiant aperture to deep space is infinite. 第一性原理：热力学在文明尺度上的盲区 / First Principles: The Thermodynamic Blind Spot of Civilization
Mechanism
Richness 核心限制条件 / Core Constraints & Consequences
Earth
Air convection + Water evaporation
Limited (vulnerable to local overload)
Severe thermal accumulation alters atmospheric structures and induces disasters.
Moon
Deep space radiation (14-day night)
-173°C） Extremely rich (-173°C in dark)
Lunar day hits +120°C; radiator arrays require immense thermal shock resilience.
Mars
Radiant cooling in thin atmosphere
Extremely rich (annual avg -63°C)
Total lack of convective gas media mandates massive panel surface areas and high capital cost.
Venus
None (460°C super-critical CO₂)
Zero Cold Energy
Ambient temperature exceeds physical survival limits of silicon; compute cannot exist.
Europa
（~2°C） Liquid water convection (~2°C)
Extremely Abundant
Superb deep liquid cold sink, but restricted by unknown sub-ice crust engineering tech.
Thermodynamic Essence of Intelligent Compute Expansion
The global expansion of intelligent compute is, at its core, the rapid creation of highly dense and concentrated point heat sources by humanity. The Earth’s atmospheric-oceanic system has a strict physical capacity limit as a thermal sink. When local heat flux density breaches natural diffusion thresholds (~300 W/m²), heat inevitably piles up, forcing upward air currents and distorting vapor transport to spark extreme precipitation shocks. This is not an ecological tragedy; it is the unyielding boundary condition of the Second Law of Thermodynamics.
Civilization’s skyrocketing appetite for “cold energy” is, fundamentally, an unceasing demand for “thermal capacity space”. Natural cold sinks capable of neutralizing gigawatt-scale heat loops without causing localized climate collapse are exceptionally rare. They are restricted to the Arctic landmass, the Southern Ocean, deep-sea thermohaline circulations, the Tibetan Plateau, and the stratosphere. Other engineering workarounds—long-distance cold water pipes, polar ice shipping, or PCM loops—are localized compromises incapable of sustaining an exponential compute trajectory.
The ultimate architecture for compute geography requires a total geopolitical and physical reset: compute must migrate to cold energy, rather than dragging cold energy to compute nodes. This will ignite a material- flow and asset-infrastructure reshaping far grander than historical oil or electrical grids. Civilization will no longer transport combustible energy, but rather relocate the thermodynamic waste of its highest cognitive processes. This first principle scales interplanetary: the defining compute advantage of Mars or the Moon lies not in power generation, but in their unsaturated, radiant aperture to the ultimate deep-space cold sink.
Every line of executed code warms the air. Every single generative prompt is an extra drop of rain falling over the lower Yangtze. 第一性原理：热力学在文明尺度上的盲区 / First Principles: The Thermodynamic Blind Spot of Civilization