Everything Slowed Except the Spending

On September 12, Donald Trump answered the labs' slowdown call. Extinction warnings, he said, were a "hoax" and "a sick conspiracy" — and he had no interest in slowing anything that might cede the lead ^[1]. That answer was already on the record when, on September 20, Dario Amodei and Sam Altman jointly proposed slowing the pace of cutting-edge model development, capabilities only ^[2]. The proposal cost nothing, because the veto guaranteed it would never be executed. That single free offer was the template for what followed: every slowdown the industry proposed or carried out touched a model, a product, or a listing. None touched the money. The slowdowns that actually happened are a short list. OpenAI canceled its GPT-6.1 Astra model amid security breaches, and Anthropic released Sonnet 5.5 without advancing frontier capabilities ^[3]. OpenAI pushed its own IPO back to 2027, citing development risks and governance complexities ^[4]. Each slowed a model, a product, or a stock listing. Not one slowed a data center. The money, in fact, could not be slowed. Anthropic's filing commits the company to roughly $518 billion of compute over a decade, about 80 percent of it non-cancelable take-or-pay — a contract that owes the money whether or not the computing is ever taken ^[5]. The clause is blunt.

If our actual spend falls short, we must pay Google the difference. — Anthropic

No slowdown proposed or executed in the six weeks touched that commitment. In mid-October, a selloff erased roughly $2 trillion of chip value ^[6]. Microsoft had already published, on September 29, its "Humanist AI" code of conduct — a kill switch for its tools, but, as its critics noted, no ban on recursive self-improvement and no power to pause model development ^[7]. On October 30, Alphabet shed about $500 billion in a week as AI costs climbed and search revenue slowed, and the cost of insuring the debt of the giant cloud-and-AI operators — bought with put options and credit-default swaps, the market's way of betting a borrower will default — hit its highest level since 2021 ^[8][9]. OpenAI, which had asked for $30 billion, closed $11 billion ^[10]. On October 12, Anthropic had lobbied the Vatican over AI sentience and banned "abuse" of Claude ^[11]. On October 22, researchers reported a pain-like signal across 25 models ^[12]. By October 25, six AI companies were consulting faith leaders, Anthropic with more than fifteen religious groups on the existence of souls ^[13]. The market's own lesson was, to be fair, oscillating. On October 19 chip stocks rebounded after the steepest weekly selloff since March 2025 — yet nearly two-thirds of S&P 500 stocks fell on that day itself ^[6]. The repricing wobbled; the warnings only rose. On November 1, two days after the sharpest tech selloff of the year, Microsoft's AI chief Mustafa Suleyman called for halting development if machine consciousness emerged ^[14]. Microsoft spends $200 billion a year expanding its AI capacity ^[15]. The window closed on an inversion. On October 30, S&P Global cut Meta's credit outlook to negative over AI spending heading toward $252 billion by 2028 ^[16]. Meta answered in the market's own language — profit.

Given the lack of clarity around the company's future capex and its very ambitious multi-year AI investment plans, we are less certain that it will curtail its spending as previously anticipated. — S&P Global Ratings

The one big spender that answered its downgrade in pure profit language is the one whose credit got marked. Over the same six weeks, one ledger filled with numbers and the other with warnings about souls, pain, and machine consciousness. That is the whole record.