The silicon cargo cult: why AI accelerationism must be rejected

In a revealing conversation on The Ezra Klein Show, Nvidia CEO Jensen Huang articulated a bizarre defense of a rapid fossil fuel-powered AI data center buildout. Confronted with the reality that hyperscalers are running into severe electrical capacity shortages, Huang argued that the United States has failed to build sufficient energy infrastructure because it got “really gummed up in climate change and sustainable energy”. His proposed remedy: we must aggressively expand fossil fuel power generation in the near term to feed AI data centers, because AI is supposedly our “best opportunity to turn a corner on climate change”. To justify this Faustian bargain, Huang utilizes a medical analogy:

“It’s kind of like, in order to save you, they’ve got to hurt you first — that’s the nature of surgery. They’ve got to cut you open to save you. They’ve got to inflict an enormous amount of pain and suffering on you so that they can save you. And so I think AI’s kind of like that”.

Huang’s pitch for AI accelerationism frames the reliance on fossil fuels as an unavoidable necessity in service of future technological salvation. However, under closer examination, the delusional nature of this thesis becomes evident. Accelerating compute by delaying decarbonization has little to do with rational pragmatism; instead, it reveals a fundamental misunderstanding of humanity’s current predicament and the physical reality behind it.

The category error: confusing technological innovation with physical solution deployment

A foundational premise of the accelerationist argument made by Huang is that the rapid decarbonization needed to achieve the goals of the 2015 Paris Agreement is held back by a lack of intelligence, and that the climate crisis is fundamentally an innovation bottleneck. It is worth noting that Huang at least acknowledges the climate crisis as an urgent matter of concern (unlike proponents of “effective accelerationism” such as Marc Andreessen, who reject ecological constraints entirely). Yet across the tech-optimist spectrum, there is a shared belief that artificial superintelligence will miraculously unlock commercial fusion, ultra-efficient carbon capture, or room-temperature superconductors in time to avert catastrophe. A related hope that using AI for innovation and operational improvements could avoid more greenhouse gas (GHG) emissions than would be caused by data centers was articulated by Nicholas Stern et al. (2025), though they did not advocate for expanding fossil fuel generation. However, the primary driver accelerating the climate crisis is not merely the operational GHG emissions footprint from data centers. Crucially, the unregulated utilization of AI by the fossil fuel industry increases the productivity of exploration and extraction activities. Will Alpine et al. (2026) demonstrated that the additional GHG emissions enabled by AI-driven gains across fossil fuel value chains substantially outweigh the GHG emission avoidance enabled through AI applications in renewable energy.

Treating a physical deployment challenge as an innovation challenge represents a category error. Novel scientific insights generated by AI do not automatically translate into rapid physical infrastructure deployment. In addition, diverting scarce grid capacity, clean baseload power, and electrical equipment away from civil infrastructure toward speculative data centers is further delaying the displacement of fossil fuels by clean energy solutions. Although machine learning is genuinely useful for specific research tasks in biotechnology and materials discovery, such breakthroughs do not resolve the primary constraints that have stalled decarbonization to date. The vast majority of technologies required for rapid decarbonization are already commercially mature and cost-competitive, as documented in Ember’s (2026) report The Age of Power. Unlike at the turn of the 21st century, the climate crisis is no longer an innovation problem but a problem of an insufficient pace of industrial deployment, compounded by insufficient economic incentives for a Paris-aligned climate mitigation pathway and widespread misinformation among voters and policy makers. The insufficient pace of decarbonization is not only a result of market failure, but also the result of institutional and regulatory failure: Owing to regulatory capture and intense lobbying by entrenched fossil fuel interests, policymakers have lacked the political will to establish effective regulatory tailwinds for clean electrification, to internalize carbon externalities through robust polluter-pays mechanisms and climate liability frameworks, or implement market incentives like The Global Carbon Reward (Chen 2026).

The thermodynamic fallacy of “surgery”

Huang’s surgery metaphor relies on an intuitive biological model: a surgeon makes a controlled incision, and excises the pathology, after which the patient’s homeostatic biology heals the wound. But applying this model to the Earth’s climate system commits a fatal thermodynamic error. A surgeon is highly skilled, possesses anatomical knowledge and follows established medical protocols designed to preserve life. In contrast, the promised salvation through hypothetical superintelligence is entirely speculative, whereas the ecological damage inflicted by fossil-fuel expansion is tangible and irreversible. Moreover, the planetary biosphere represents a complex thermodynamic system governed by global energy balances and geochemical feedback mechanisms. The GHG emissions resulting from the combustion of fossil fuels to power data centers do not behave like a controlled incision. Instead, they contribute to a cumulative radiative forcing that traps excess thermal energy in the oceans and atmosphere across millennial timescales. Finally, the Earth’s climate system possesses numerous non-linear tipping points. If near-term emissions push the biosphere past systemic thresholds, resulting in e.g. the collapse of the Atlantic Meridional Overturning Circulation (AMOC), the irreversible destabilization of polar ice sheets, or self-sustaining boreal and permafrost feedback loops, the system locks into an alternate, permanently degraded state of equilibrium. If the surgical analogy applies anywhere, it is to the rapid phase-out of fossil fuels. Yet even there, the commercial availability of climate solution technologies makes that transition far less painful than transition skeptics often claim, in particular for energy-importing nations.

The Say’s Law fallacy: confusing factor cannibalization with infinite demand

Another core tenet of accelerationism is the belief in virtually limitless economic abundance unlocked by superintelligence. Rapid adoption scenarios, such as Anthropic’s models projecting 15% sustained annual GDP growth, assume an unprecedented productivity explosion. This thesis rests on an economic article of faith: that compute demand is functionally infinite, and that building fossil plants is simply a pragmatic response to an inescapable Jevons paradox. In reality, this infinite demand curve is a macroeconomic fiction. It invokes Say’s Law in an aggregate-demand-agnostic vacuum, presuming that the scaling of synthetic cognitive labor automatically generates the purchasing power to absorb its output. To be sure, growing compute demand is temporarily supported by sovereign defense and intelligence spending (which operates on state balance sheets independent of commercial ROI) and factor cannibalization (where enterprises replace white-collar payroll with software subscriptions to defend margins during economic stagnation). However, military budgets face strict fiscal limits and divert capital desperately needed for climate mitigation and adaptation, while factor cannibalization is constrained by a finite pool of substitutable tasks. In the real economy, markets face numerous realization problems:

• The underconsumption trap. The ultimate buyers of B2B enterprise software and commercial services are consumer-facing industries. Consumer spending in advanced economies is sustained by wage earners whose marginal propensity to consume (MPC) is roughly 0.8 to 0.95. By contrast, corporate treasuries and capital owners exhibit an MPC of only 0.2 to 0.4. If AI systems displace cognitive workers at scale without an unprecedented redistribution of capital profits, aggregate consumer demand declines, forcing output into decline as well.
• The limits of Jevons. The Jevons paradox requires high price elasticity of demand (>1) driven by expanding marginal economic utility. While early task automation provides clear value, marginal reasoning tokens quickly encounter diminishing returns. When the enterprise utility of generating additional synthetic text or speculative code flattens, demand becomes inelastic (<1), and total compute spending contracts rather than expands.
• The circular financing loop. A substantial portion of current compute “growth” is an accounting carousel detached from real-world revenue. Hyperscaler balance sheets deploy tens of billions directly or indirectly to back AI startups, which immediately return those funds to the hyperscalers to lease cloud clusters.

The millenarian trap: burning the seed corn

History offers sobering precedents for societies that deliberately sacrificed their immediate, physical means of survival in the conviction that doing so would summon a transcendent power to resolve their crisis. In 1856, facing the combined pressures of British colonial encroachment and a devastating outbreak of bovine lungsickness, the Xhosa people of southern Africa followed the prophecies of a young woman named Nongqawuse. The spirits promised that if the nation proved its faith by slaughtering its cattle and burning its grain fields, ancestral warriors would rise from the earth, granaries would overflow with supernatural abundance, the sick would be healed, and colonial invaders would be swept into the sea. A majority of the Xhosa went ahead and slaughtered their cattle and burned their grain, but the ancestors never appeared. Over 40,000 people starved to death, and the survivors were forced into indentured labor, destroying the society’s autonomy for generations. Drawing this historical parallel is not intended to claim the equivalence between a 19th-century colonial catastrophe and the modern collective mania surrounding AI. Rather, it highlights an identical failure mode of human decision-making under crisis. Just as the Xhosa believed in the necessity of ritually sacrificing their tangible, physical means of survival, accelerationists demand that we surrender our ecological real-world buffers (finite carbon budgets, grid stability, and clean air), trusting that AI-salvation will arrive in time to cancel humanity’s ecological debt. Any doctrine demanding that society suspend well-established scientific insights and exhaust its real-world resources on the promise of technological deliverance is essentially a cargo cult.

True optimism belongs to the physical world

There is an authentic, inspiring version of techno-optimism. It repairs the fractured relationship between humanity and nature by subordinating technology to planetary boundaries and social foundations, as articulated in Doughnut Economics. Human civilization is recognized as an integral part of a fragile living biosphere that must be responsibly stewarded. This philosophical stance finds its cultural expression in the solarpunk movement. It is the kind of optimism that builds high-speed rail networks, deploys hundreds of gigawatts of solar and wind, electrifies industrial thermal processes, and constructs resilient, abundant infrastructure within a safe and just operating space.

AI has a genuine, valuable role to play in that future: as an analytical tool to optimize electrical grid dispatch, streamline logistics, organize human knowledge, and assist in physical modeling. But AI proponents who demand that society prioritize speculative data center buildouts over physical decarbonization have ceased to be champions of progress. They are effectively asking us to burn our grain fields in the hope that an algorithmic messiah will provide abundance for all (which, between the lines, means windfall profits for a few). The ecological stakes and real-world harms are too high to tolerate AI accelerationism as an acceptable worldview. It is time for genuine techno-optimists to insist that, as the Pro-Human AI Declaration demands, AI must serve humanity and foster genuine progress, rather than enfeeble society and undermine our future.

illuminem Voices is a democratic space presenting the opinions of leading Sustainability Thought Leaders, their views do not necessarily represent those of illuminem.

The world needs sustainability knowledge. At illuminem, no interest group or shareholder can influence our work. Thank you for supporting our mission to make high-quality and independent sustainability information free for all. Every contribution helps. Thank you for donating today.