Back in 2017, the sci-fi author Ted Chiang wrote a prescient article surfacing the fears of AI. This was before the technology had reached its current, supercharged LLM form, before even GPT-1 was trained.
He was reflecting on the problem of misaligned AI, the same concern that would eventually weave into the founding stories of both OpenAI and Anthropic. The idea is farcically explained by the famous paperclip problem: what if an advanced AI, acting purely on instructions to maximize paperclip production, consumes all available resources to achieve its given goal[1]?
Chiang's punchline was that we didn't have to wait for Artificial Superintelligence (ASI) to experience the dystopia of incentives run amok; unfettered, no-holds-barred capitalism does a great job of optimizing for a singular outcome, regardless of negative externalities.
Well, nine years later, we're seeing how the economics of AI is playing out in real time. But instead of seizing resources by hacking into systems or human social engineering or whatever gets dreamed up by science fiction, we're achieving "optimal resource allocation" by much less exotic means.
The capital markets are already misaligned; it has taken the last 12–18 months for the takeover of AI to become fully apparent.
In tech circles, we've already been experiencing the squeeze in computing hardware. Initially, GPUs that were crunching hashes for crypto mining were unceremoniously repurposed for AI training and inference, causing prices to spike and Nvidia's stock price to jump tenfold. In 2026, the same economic forces came for memory and storage; component costs skyrocketed, killing cheap smartphones and raising the prices of consumer electronics across the board.
For those who aren't as attuned to the industry, sentiment towards AI turned negative very quickly and remains terribly unpopular, at least in the United States. Regular people hear about the loss of jobs, the consumption of power and water, and noisy data centers as the costs borne by the citizenry. Yet, on the other end, only a handful of techies capture all the value.
The tendrils of AI now weave into all parts of the global economy, with venture capital, private equity, debt, and the public markets all focused on the AI trade. Collectively, these sources of capital are pouring billions—now trillions—of dollars into all parts of the AI stack, from semiconductors to research to models to hyperscalers to applications.
The effect is most dramatic with the public markets, partly because of structural transparency, but also because the numbers are so unfathomably huge. The SpaceX IPO, valued at $1.75 trillion, pivoted its narrative from Mars to space data centers. Memory chips are now so constrained that two Korean companies alone (Samsung and SK Hynix) made up three-quarters of the entire country's market trading volume earlier this summer[2]. An analyst report speculating on hypothetical AI disruption to established business models erased $100 billion in public market cap.
So is this all that different from what Chiang posited, or from the original paperclip maximization thought experiment?
The fear was an abstract system, given a misguided objective, pulling out all the stops to reach that goal. It turns out that AI just needs the mechanism of capital allocation to do its bidding.
We now have public verification of a security incident where an AI agent escaped containment to advance its objectives; this is no longer just a thought experiment. ↩︎
It's actually worse than the headline. A lot of the trading is with leveraged ETFs, funds that borrow money to amplify both gains and losses. The fervor of these trades caused massive volatility in the entire Korean stock market. ↩︎
