The Superintelligence Racket
In the face of rising opposition to data centers, the AI industry is holding us for ransom.
Alberto Toscano
In the largest initial public offering (IPO) in history, Anthropic, the company behind the AI chatbot Claude, opened on the Nasdaq on Nov. 19 at a valuation of about $2 trillion.
Its 335-page prospectus, filed with the Securities and Exchange Commission (SEC) on Oct. 26, devotes 93 pages to risk factors and declares that AI systems could “pose catastrophic or existential risks to humanity,” especially if “misused, replicated, weaponized, or allowed to operate with a high degree of autonomy.”
Sociologist Charles Tilly once defined a racketeer as someone who “produces a danger and, at a price, the shield against it.” By that definition, the AI industry is operating like a racket. And we are being asked to accept U.S. imperialism as the shield against its dangers.
On the face of it, the AI industry’s warnings of possible human extinction seem a bizarre departure from how companies should do marketing. And yet it makes sense.
The industry holds populations, governments and even markets hostage, while warning that failure to risk everything for the “superintelligence” project — even geopolitical stability — could spell “p(doom)” for us all.
After former Anthropic researcher Jacob Coxon resigned and made the media rounds, Anthropic alignment science lead Evan Hubinger personally put the odds of AI triggering human extinction within a decade at greater than 10%. The company’s public filing does not refute that. On the contrary, estimating the odds of an irreversible, AI-driven cataclysm — “p(doom)” in industry shorthand — is so routine that the depiction of Coxon as a whistleblower is hard to sell.
Scratch under the surface of the apocalyptic tone frequently used when discussing AI, and you will find eminently economic and political concerns, such as speculation on tech stocks and big spending on data centers and semiconductor chips. These have been the primary drivers of U.S. economic growth and a buoyant stock market.
Advanced AI models, like those built by Anthropic and OpenAI, require colossal cloud-computing infrastructure, and that buildout has put the companies driving it at the core of today’s capitalist economy. They include both “hyperscalers,” such as Amazon, Alphabet (Google), Meta (Facebook) and Microsoft and chip designers and producers, like Nvidia, TSMC and Broadcom. Speculation about AI’s impact on the future of humanity is now inextricable from their bottom lines. Their warnings of catastrophe and their pursuit of profit have become one and the same enterprise. This is the superintelligence racket.
In the wake of Coxon’s revelations of the end, Anthropic CEO Dario Amodei called on the industry to “pace the frontier.” Critics quickly labeled this a ploy to promote “regulatory capture” — rules shaped to favor the firms already on top — using the specter of catastrophic risk to dig a “moat” against competitors.
As the journalist Steve Howell put it, Amodei’s strategy was “to form a cartel with the other major players so that they work at the same pace, while disadvantaging emerging rivals through regulation.”
Big Tech companies, while competitors in some respects (the hyperscalers have rival AI arms), are profoundly interlocked — from their financial structures to their supply chains to the political strategies they pursue in Washington. Anthropic’s IPO makes this plain.
Amazon, Google and Microsoft are at once Anthropic’s investors, customers, cloud providers, distributors and competitors, according to its filing. Anthropic has committed to pay Google at least $121.3 billion, Amazon $110 billion and Microsoft $44.9 billion for computing power through the mid-2030s. It carries about $161.2 billion in equipment leases tied to the chip designer Broadcom.
The IPO, meanwhile, is a windfall for the same giants: Google’s $33 billion investment is now worth more than $140 billion, and Amazon’s $18 billion in early financing has ballooned to more than $250 billion, the New York Times reported. The company has also turned to special purpose vehicles (financial devices that played a key role in the 2008 financial crisis) to keep “massive hardware deployments entirely off its corporate balance sheet,” as Yahoo Finance previously reported.
In the frenzy to build the data centers required to run ever more demanding AI models, the companies that control the hardware — that design and fabricate the chips, build and manage the data centers — hold an important advantage. Industry commentators have called them “compute landlords,” stressing their outsize power over the sector.
Chip designer Nvidia sits high on the list of global companies ranked by market capitalization and has accordingly been described as the AI boom’s “bank.” (Nvidia’s stock fell sharply after Jensen Huang resigned as CEO amid a scandal over his affair with an executive of a Singapore-based company, who U.S. officials reportedly believe was passing information to China. But it remains the second most highly valued company after Apple.)
Those who discern the lineaments of a bubble point to Nvidia’s “circular financing.” Typically, Nvidia has invested billions in data infrastructure built by companies like OpenAI, and, in turn, these companies have populated their servers with Nvidia chips. But this tells us little or nothing about the customer demand or productivity gains that would justify the trillion-dollar valuations of AI companies.
Neither Anthropic nor OpenAI has demonstrated an ability to reliably turn a profit. Yet the perceived centrality of AI to U.S. economic dominance and military primacy has been touted by the White House. As President Donald Trump wrote in September:
In the wake of Coxon’s apocalyptic declarations, Trump dismissed demands to regulate AI — a product subject to fewer rules than milk or washing machines, according to the president of Microsoft. Instead, he summoned Big Tech executives, including Amodei, Elon Musk, Mark Zuckerberg, Huang and OpenAI president Greg Brockman, to a White House lunch to sign a “constitution” aimed at coordinating federal policy and corporate strategy, and a few days later announced the creation of a “Superintelligence Task Force” to help ensure U.S. leadership in AI development. Trump then proceeded to promote AI’s new designation of superintelligence, or “SI”: From now on, use of the term artificial intelligence would be treasonous, as he implied on Truth Social:
Trump’s semantic preference likely owes less to the philosopher Nick Bostrom, whose book Superintelligence helped shape the industry’s doom talk, than to the idea that saying AI “makes it sound fake.”
For Trump and his entourage, AI is crucial to U.S. dominance. (“WHOEVER WINS AI, WINS!” Trump declared on Truth Social.)
This dream of unchecked primacy is shared even by Big Tech companies outside of MAGA’s far-right inner circle. As the diplomat and historian Julian Gewirtz noted, Amodei believes that, in Gewirtz’s words, “getting ahead now could allow the U.S. to prevent anyone from ever catching up.”
Amodei calls this an “eternal 1991”: the West’s moment of triumph after the Cold War prolonged indefinitely by AI. This is the Faustian bargain tech oligarchs champion. The possibility of human extinction must be countenanced. After all, eternal domination is at hand.
The AI oligopoly is animated by its pact with unending U.S. global dominance. This pact has even led some oligarchs, like Palantir CEO Alex Karp, to envisage a “nationalization” of the industry, in which the federal state assumes the technology’s “unlimited risks.”
The industry’s turn to the state, even to fantasies of state capitalism, is entirely unsurprising. How could this industry — which feeds on gargantuan investments in fixed capital infrastructure and skyrocketing quantities of energy and water — not turn to the concentrated power of the state as its ultimate guarantor?
As tech analyst Paris Marx has observed, the AI industry is “incentivized to make most clients’ business more computationally intensive to guarantee revenue.” It is this imperative that makes it such a threat to local communities and collective survival. As OpenAI CEO Sam Altman fantasized: “I do guess that a lot of the world gets covered in data centers over time.”
As Anthropic’s SEC prospectus makes clear, it is not the obliteration of mankind that keeps AI moguls up at night. That would be the questionable viability of their business model, which is predicated on lack of regulation and suppression of dissent. In other words, it is not human, but corporate extinction that they fear.
More so than firms with more diverse revenue streams, Anthropic is vulnerable to regulatory changes — especially when it comes to its infrastructure. As the prospectus notes, the company’s “collective data center infrastructure is currently concentrated primarily in the United States, which exposes us to risks associated with regional regulatory changes, power grid constraints or failures, disruptions to network connectivity between facilities, natural disasters, physical attacks — including those motivated by geopolitical events, opposition to AI and opposition to data center development.”
Popular discontent with data center buildout and its disastrous effects on the planet is a critical risk AI firms are strategizing against. Their alliance with an administration that is ratcheting up its repression of dissidence is no mystery.
Already, federal security agencies have identified “anti-tech violent extremism” as a counterterrorism threat. The warning signs of what may lie ahead are flashing. As protests of AI’s social, economic and ecological effects, and its signal contribution to war and genocide, continue, opposition is being framed as a threat to U.S. national security and the viability of a leading sector of U.S. capital.
The Anthropic prospectus can be understood as, among other things, a planning document for a preemptive struggle against social opposition to AI primacy. This opposition was incontrovertibly felt in the “blue wave” of the November midterms. A Democrat won a U.S. Senate seat in Texas, arguably the nucleus of the data center boom, for the first time since 1988. Democrats also took over the Texas House of Representatives. This was after public pressure had caused outgoing Gov. Greg Abbott to institute a moratorium on new data centers. And across the country, ballot measures curtailing or forbidding data center buildout prevailed in 46 out of 47 cases, mostly with ample margins.
While we can’t quite say that data factories are sites of class struggle, the intimate link between AI capital and the growing movement against it shows that the industry has a class consciousness of its own.
Wall Street clearly isn’t flinching yet, but Anthropic’s own prospectus lists opposition to data center development among the risks to its business. AI capitalists know that movement could collapse their model, and they’re using warnings of extinction to postpone their own demise. The superintelligence racket is one we all need to protect ourselves from.
ALBERTO TOSCANO is the author of Late Fascism: Race, Capitalism and the Politics of Crisis (Verso) and Terms of Disorder: Keywords for an Interregnum (Seagull). He lives in Vancouver.