The Planetary Cost of AI Acceleration A Thermodynamic Outlook on Four Possible Paths Forward

William Zhu Lei Zhu · 2026

Read The Planetary Cost of AI Acceleration A Thermodynamic Outlook on Four Possible Paths Forward by William Zhu Lei Zhu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-28949.txt.

Introduction

Zhu and Zhu argue that the growth of AI compute will hit thermodynamic limits before data or capital limits. They frame AI as the current physical form of a long process of intelligence evolving on Earth. If compute keeps growing at an exponential pace, heat and the finite heat capacity of Earth set the bound. The evolution of intelligence is, in their view, a problem of non-equilibrium thermodynamics. They outline four possible paths forward.

Catalog record

CREATORWilliam Zhu Lei Zhu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28783
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.28949

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The Planetary Cost of AI Acceleration A Thermodynamic Outlook on Four Possible Paths Forward William Zhu Lei Zhu Abstract The artificial intelligence industry is not an isolated economic phenomenon; it is the current physical substrate for a broader, multi-billion-year process: the evolution of an abstract ”intelligence” on Earth. As computation accelerates toward a planetary-scale phase transition, the dominant discourse remains largely confined to algorithmic architectures, alignment, and silicon supply chains. But physics invariably asserts itself. When analyzed from first principles, it becomes clear that if the current trajectory of exponential computation growth holds, the ultimate bottleneck of the coming decades will be neither data nor capital, but the laws of thermodynamics and the finite heat capacity of the Earth. The evolution of intelligence is fundamentally a problem of non-equilibrium thermodynamics , bound by strict hardware limitations, and ultimately, an absolute ecological boundary condition. Civilization itself is not a divine blessing, but an exceedingly rare and expensive ”thermodynamic algorithm” actively fighting the default settings of the cosmos. Navigating this trajectory requires a rigorous examination of the physical laws governing computation, complexity theory, and the narrow thermodynamic tightrope civilization must walk to survive it. 1. The Physical Cost of the Agentic Wave T he computing paradigm is rapidly transitioning from passive generative models to continuous swarms of persistent, autonomous economic actors. Driven by the explosive growth of LLM agents led by OpenClaw and the beginning of its systemic integration into many aspects of the economy, the market is pricing in an unavoidable physical reality. Foundational computing hardware providers are projected to experience a non-linear surge, reaching yearly multi-trillion-dollar production and revenue within three years, backed by substantial backlogs of next-generation processing units. However, beneath the financial metrics, a profound thermodynamic paradigm shift has already occurred in the state space of computation. The industry is no longer optimizing just for raw flops or parameter counts. The fundamentals of the next epoch have quietly become the energy-to-token and energy-to-intelligence rate. This trajectory is deeply non-linear. To support continuous, agentic intelligence, the globe will be blanketed with gigawatt-scale data centers. But beyond the high electrical demand lies an inescapable, physical corollary: every single joule consumed by these server racks—and by the dedicated power plants built to sustain them—must eventually be dissipated into the environment as vast volumes of waste heat. 2. Computation as Dissipative Thermodynamics Building intelligence is to wage a localized war on entropy. But physics exacts an absolute toll. Intelligence, whether biological or artificial, is not a disembodied abstraction; it is a physical process governed by statistical mechanics. To compute is to manipulate information, ordering a system into a state of lower internal entropy. By Landauer’s Principle [ 2 ] , there is a strict thermodynamic lower bound to this process: erasing a single bit of information requires a minimum amount of energy ( E\geq k_{B}T\ln 2 ), which is irreversibly released into the environment as waste heat. Viewed through the lens of non-equilibrium thermodynamics, intelligence is a dissipative structure [ 4 ] [ 3 ] —a concept pioneered by Ilya Prigogine to describe highly ordered, complex systems that can only sustain their internal organization by continually exporting maximum entropy (waste heat) into their external surroundings. Scaling computation inextricably means scaling energy, which then inevitably means scaling heat dissipation. When stripped to its fundamental physics, an expanding cognitive infrastructure essentially acts as a localized, growing entropy-reduction engine that demands a proportionally vast and growing thermodynamic exhaust. The rate of waste heat production is strictly proportional to the scale and speed of the computation. Computation, at its physical limit, is heat dissipation. 3. Earth’s Fragile Phase Space Here is where information theory collides with complex system ecology: Earth’s current thermodynamic balance sheet operates within a thin margin, and the non-linear growth of computational heat dissipation presents a foundational structural vulnerability. The Earth is virtually a closed thermodynamic system; it exchanges energy with space, but practically no mass. Its waste heat reservoir is strictly finite, and its natural heat outlet—the boundary interface radiating infrared energy into the 2.7-Kelvin vacuum of space—is constrained by the Stefan-Boltzmann law. For the past few thousand years, human tribes and empires have operated under the arrogant assumption that the available ”low-entropy frontiers” were infinite, allowing them to freely dump their internal systemic waste heat outward. Historically, the planetary heat source has been the Sun. The biosphere has evolved a fragile equilibrium with it: reflecting a massive portion back into space (albedo) and absorbing just a fraction to drive weather patterns and biological life. Utilizing solar or wind power for compute is thermodynamically safe because it only intercepts and redirects existing solar heat flows within this closed loop. However, the chemical (fossil fuels) and atomic (nuclear fission) energy being unearthed to power the sprawling computational infrastructure boom represent entirely net-new additions to this balance sheet. New heat is being continuously injected into the system from within. How large is Earth’s buffering reservoir? Results from complexity theory, and in particular catastrophe theory, suggests that it could be vastly smaller than what a linear model would estimate. In complex, highly coupled ecological networks, continuously injecting energy does not cause a slow, linear degradation. Instead, the system absorbs the stress until it reaches a critical bifurcation point—the edge of its current basin of attraction. The Earth’s existing heat outlet is already performing near maximum capacity just in the effort to vent the accumulated warmth of historical greenhouse gasses. When the sheer rate of anthropogenic waste heat exceeds the dissipation capacity of the reservoir, the planetary ecological system will not just be incrementally warm—it can fail via an abrupt ecological phase shift. 4. The Thermodynamics of Growth Faced with the prospect of overloading the biosphere through the exploding scale of computation, a reactionary impulse is deceleration. Is the exponential pursuit of cognitive capacity a misstep? Should this growth be halted? Objectively analyzed from first principles, the answer is an unequivocal no. The development, scaling, and complexification of systems are not arbitrary technological choices; they are innate, physical imperatives of the universe. From the first single-celled organism to planetary-scale silicon neural networks, open systems naturally evolve to maximize the flow and dissipation of energy across gradients. According to the Constructal Law and the principles of evolutionary thermodynamics, nature favors systems that optimize energy currents. Life itself is a complexification engine that continually evolves to harness steeper entropy gradients. The drive to process more information, to compute the environment more accurately, and to expand structural capacity is the fundamental defining characteristic of living systems. Intelligence is simply the latest, highest-bandwidth substrate for this cosmic process. Attempting to artificially halt this evolutionary trajectory is fundamentally contradictory to the nature of life. In a dynamic, highly entropic universe, static equilibrium is synonymous with death, and slowing or freezing the development of a complex system is a direct path to systemic failure when viewed from a geological and cosmic timescales. Systems that cease to adapt are inevitably erased by shifting environments. Therefore, paradoxically, accelerating the development of intelligence is the most potent defense against the thermodynamic limit it exacerbates , because advanced cognitive systems possess the unique, emergent capacity to act as a planetary Maxwell’s Demon—sorting, optimizing, and radically reducing

Read the complete text