The Bottleneck Is Not Cognitive

AI-Solutionism as Fetishistic Disavowal

In February 2025, Melanie Nakagawa, Microsoft’s Chief Sustainability Officer, published a blog post explaining that her company’s emissions had risen 23.4% since 2020, the year it pledged to be carbon negative by 2030. The pledge, she wrote, was a “moonshot.” Now, “the moon has gotten further away.” The cause was artificial intelligence. The solution, she wrote, was also artificial intelligence.

Five months earlier, Sam Altman published “The Intelligence Age,” in which he listed “fixing the climate” among the “astounding triumphs” superintelligence would render commonplace. Dario Amodei’s Machines of Loving Grace said the same thing in a more cautious register. Demis Hassabis has been saying it since 2023.

Two stories live inside the same boardrooms. In one, AI is breaking the climate goals. In the other, AI will fix the climate. Both are true. Neither is a contradiction the executives are eager to resolve, because the contradiction is the business model.

This is the structure to notice. AI will not solve climate change, because climate change is not a problem we lack the means to solve. The IPCC said so in 2022. The IEA said so again in 2025. The bottleneck is political. It has always been political. The AI-fixes-it narrative does not address the bottleneck. It performs a more interesting function: it makes the bottleneck disappear from view.

What we already have

The IEA’s Global Energy Review 2026 records 800 GW of new renewable capacity added in 2025, a 16% year-on-year increase and the 23rd consecutive annual record. Solar PV alone accounted for 605 GW. Battery storage costs fell roughly 40% in 2024 to about $150 per kWh. At least 1,650 GW of advanced renewable projects sit waiting at grid interconnection queues.

We do not have a technology gap. We have a permitting gap, a transmission gap, a subsidy gap, and a power gap. The OECD and IEA’s December 2025 joint report puts global fossil-fuel support at $916 billion in 2024. The IMF’s broader implicit-subsidy figure, including unpriced externalities, reached $7 trillion in 2022. Consumer-facing clean energy received about $70 billion in 2024. The ratio is roughly nine to one against the transition.

Capital follows the ratio. ExxonMobil reported $33.7 billion in 2024 net income. Saudi Aramco reported $106.2 billion and committed $85.4 billion in declared 2025 dividends. The fossil sector spent $219 million to influence the 2024 US election and $151 million on federal lobbying that year alone.

The COP30 outcome in Belém, November 2025, contained no binding fossil-fuel transition language. Colombia’s delegate told the closing plenary the conference had failed the science. The United States sent no delegation. Trump’s Executive Order 14154 froze IRA appropriations on his first day in office. Executive Order 14162 notified Paris withdrawal. EPA Administrator Lee Zeldin moved to revoke the 2009 endangerment finding, the legal foundation for federal greenhouse-gas regulation in the United States.

In Europe, the Omnibus Simplification Package, finalised in December 2025 and entered into force in March 2026, narrowed the Corporate Sustainability Reporting Directive to firms above 1,000 employees and €450 million in turnover. The threshold exempts roughly 92% of companies originally covered. The climate transition plan obligation in CSDDD was deleted outright. Pascal Canfin, a liberal MEP, said in plain words that the deletion came at American demand. Friedrich Merz pushed the bloc to drop the 2035 ban on internal combustion engines. The Green Deal is being gutted by the coalition that built it.

This is not a knowledge problem.

What AI is actually doing to the energy system

Hyperscaler emissions are rising, not falling. Google’s 2024 environmental report shows total emissions up 51% since 2019, with data-centre electricity demand growing 27% year-on-year. Microsoft is up 23.4% since 2020. Meta and Amazon are on the same curve.

The capital expenditure of five tech firms exceeded $400 billion in 2025 and is set to rise another 75% in 2026. The IEA notes this figure is now larger than global investment in oil and gas production. That capex is not buying solar farms. It is buying gas turbines, restarted nuclear units, and long-duration offtake agreements with whatever generation can deliver firm power on AI’s deployment timeline.

GE Vernova ended 2025 with an 80 GW gas turbine backlog stretching to 2029, with hyperscaler offtake deals reaching out to 2035. Siemens Energy reported its largest order intake on record. Mitsubishi Heavy Industries doubled gas-turbine manufacturing capacity. Constellation Energy is restarting Three Mile Island under a 20-year power purchase agreement with Microsoft. The pipeline of small modular reactor offtake agreements grew from 25 GW at the end of 2024 to 45 GW by April 2026.

These are infrastructure decisions being made now. They lock in fossil and nuclear dependency through the 2030s and 2040s. They are being made because AI demand is the most credible buyer of firm power in the global energy market.

The IEA’s measured assessment of AI’s hypothetical climate benefits, in its 2025 Energy and AI report, concedes that the projected emissions reductions from AI applications “would be far smaller than what is needed to address climate change,” and adds the cleaner sentence: “there is currently no momentum that could ensure the widespread adoption of these AI applications.”

Translated: the alleged benefit is conditional on a political revolution that is not happening. The cost is being spent now. The Altman, Amodei and Hassabis claims are not predictions. They are sales pitches with a planetary cost.

The technological sublime

Jameson, in Postmodernism, or the Cultural Logic of Late Capitalism (1991), gave us a name for the aesthetic mode in which AI now operates: the hysterical sublime. The eighteenth-century sublime confronted us with nature too vast to comprehend. The storm. The abyss. The alpine peak. The postmodern sublime confronts us with something else, the totality of capital itself, expressed in technological systems whose scale and operation exceed individual representation. Jameson’s example was the computer network. The data centre is its perfected form.

The point of the sublime, Jameson argued, is misrecognition. We mistake the difficulty of cognitive mapping, the difficulty of seeing where we are inside the system, for the impossibility of cognition altogether. Awe substitutes for analysis. The system becomes a face of God. The political question of who built it, who runs it, and on whose terms it operates, falls away.

This is what AI does to climate discourse. Microsoft’s $80 billion in fiscal-year 2025 capex is sublime. The five-firm $400 billion total is sublime. The model parameter counts are sublime. The IEA’s 945 TWh by 2030 is sublime. Each figure produces awe. None of them produces a sentence beginning with “Sam Altman should.” Reverence is what defers the political question. The technological sublime is what allows trillion-dollar infrastructure decisions to feel like weather.

Jameson’s other move is more specific, and more useful. Late capitalism, he argued, dissolves the distinction between cultural production and economic production. AI is the literal fulfilment of this thesis: a technology whose product is culture, whose output is text and image and code, and whose business model is the metering of cognition itself. To be inside this technology is to be inside late capitalism. The Microsoft that bought Three Mile Island and the Microsoft that pledged carbon negativity are the same firm. The contradiction is not ironic. It is organisational, material, financial. It is the structure.

The fetish

Žižek’s contribution is to name the psychic mechanism that lets the contradiction stay sustainable.

The relevant Lacanian concept is fetishistic disavowal, captured in the formula Octave Mannoni gave us and Žižek made canonical: Je sais bien, mais quand même. I know very well, but nevertheless. The fetish is not what you believe instead of the truth. It is what allows you to know the truth and act against it without psychic damage. The classical Marxist model of ideology was “they do not know what they are doing, but they are doing it.” Žižek’s update, in The Sublime Object of Ideology, is sharper: they know what they are doing, and they are doing it anyway. The cynicism is not a failure of ideology. It is the form ideology now takes.

AI as climate saviour is a textbook fetish. Microsoft’s executives know their AI buildout is breaking their climate goals. They publish the numbers. Altman knows climate action is a political and not a cognitive problem. He has been told by every climate scientist he has met. They know. Nevertheless. The future AI carries the belief on their behalf. The fetish does the believing so they do not have to.

In Lacan’s register, AI occupies the position of the subject supposed to know: the figure to whom we transfer our agency, the doctor, the analyst, the priest, the leader who is assumed to possess the knowledge that orders our world. We do not need to know how the climate will be fixed. The AI will know. The transfer is not a clinical delusion. It is a structure of belief that keeps daily life moving. It is also why pulling the AI fantasy out from under educated professionals feels like a personal attack. You are not refuting an argument. You are removing a load-bearing wall.

Two further moves. The big Other: the symbolic authority we pretend to believe in even when we know it is fictional. The AGI that will solve climate is the big Other of contemporary tech ideology. It does not exist. It is not coming. It functions, regardless. Investors price equities around its imminent arrival. Governments shape industrial policy around its needs. The non-existence of the object is no obstacle to its political effects.

Then the obscene superego. Standard ethics commands restraint: thou shalt not. The Lacanian superego commands the opposite: enjoy. Push harder. Accumulate faster. Produce more. Marc Andreessen’s manifesto sentence, “Any deceleration of AI will cost lives. Deaths that were preventable by the AI that was prevented from existing is a form of murder,” is the obscene superego in undisguised form. The ethical injunction to slow down is recoded as homicide. Refusal becomes complicity. The only permitted ethical posture is unrestrained acceleration. This is not a marginal eccentricity. It is the operating ideology of a sector with $400 billion in annual capex.

Žižek’s term for the people who deliver this ideology is liberal communist: the Davos figure who concentrates wealth and casts himself as the only conceivable agent of its redistribution; the hyperscaler CEO who orders the gas-turbine grid and gives a TED talk on planetary stewardship. The disavowal is not personal hypocrisy. It is structural. They cannot mean it and continue to do what they do. They continue. They mean it.

The fantasy is also why the public goes along. Fantasy, in Lacan, structures desire: it tells us what to want. The AI-fixes-climate fantasy structures the only desire still permitted to a depoliticised citizen. Within the fantasy, you wait, you invest in the index fund, you trust the process. Outside it, you confront named oil executives, captured regulators, and a political class that has chosen the wrong side. The fantasy is more comfortable. That is its function. You continue to recycle, drive the EV, sign the petition, and feel that something is being done, while the structure that produces the crisis goes unmolested. Žižek calls this decaffeinated politics. Climate politics under AI-solutionism is decaffeinated to the point of homeopathy.

The pattern

The shape repeats outside climate. In 2024, FAO recorded 673 million people hungry while global cereal production hit record levels. Up for Growth puts the US housing deficit at 3.78 million units; the Minneapolis 2040 plan cut counterfactual rents by 17 to 34%. The WHO Pandemic Agreement was adopted in May 2025 and stalled within months on the pathogen-sharing annex. Gabriel Zucman’s 2% global minimum tax on roughly 3,000 billionaires would raise $250 billion a year. Distribution, zoning, intellectual property, sovereign coordination. Four bottlenecks. None cognitive.

The landing

Refusing the AI-fixes-it narrative is not Luddism. The technology has uses. The objection is to the narrative function, not the artefact. The narrative converts political problems into engineering problems and sells the engineering at a margin. It is a category error wearing the costume of a solution.

The positive-nihilist position is unsentimental about meaning and uncompromising about action. There is no cosmic guarantor. There is no AI that will redeem us. There is no future in which the structures that produced this crisis spontaneously decide to dissolve themselves. There is only the question of what we do with the time and the tools we already have. The IPCC said the tools exist. The IEA’s deployment data confirms it. The bottleneck is fossil capital, captured legislatures, and a culture that has confused reverence for machinery with strategy.

The stakes are simple. If the AI-solutionist story holds, the next decade is consumed building an energy system optimised for inference compute, and the political work of decarbonisation is deferred indefinitely while we wait for a saviour that has no incentive to arrive. If the story breaks, we are left with the unglamorous task of organising against the actors who profit from delay. That task is not new. It is just ours.

We know what to do. We are choosing not to do it. That is a political fact. The point of saying so out loud is that political facts can be changed.