Apollo’s Slok: AI’s income are ‘being funded by traders somewhat than earned from clients’

In a blog post revealed on Friday, Apollo Chief Economist Torsten Slok highlighted that the elements of the AI worth chain with the very best revenue margins—corporations making AI fashions and purposes—even have the bottom ranges of profitability, a departure from the usual enterprise mannequin of, properly enterprise, by which revenue margins are increased for corporations promoting an finish product to shoppers.

Slok broke down AI corporations into 4 classes: fashions and purposes, cloud and compute, vitality and grid, and silicon and tools. Utilizing knowledge from Pitchbook and Bloomberg for corporations together with OpenAI, Anthropic, Microsoft, Amazon, Constellation Power, Nvidia, AMD, and Micron, Slok calculated that silicon and tools—similar to chipmakers—has the very best revenue margin, 41%, within the AI worth chain. In the meantime, fashions and purposes—like Anthropic—have a -59% working margin.

Slok warns that this sharp disparity is as a result of cash from the AI growth isn’t coming from pure demand for AI purposes, however somewhat shareholders hoping to money in on what they hope is the following technological revolution.

“AI growth’s income are presently being funded by traders somewhat than earned from clients,” Slok mentioned. “The upstream margins are actual, however they’re paid for out of capital raised by the layer shedding cash, not out of money generated by finish demand.”

Goldman Sachs now tasks AI investments to swell beyond $1 trillion in 2026, however thus far, the Technology has little to indicate for itself, with no important modifications in financial productiveness or revenue margin development outdoors of the Magnificent Seven. Ought to AI financing decelerate, the lopsided revenue margin construction threatens to topple the soundness of your entire quickly increasing trade, Slok warns.

“The underside line is that probably the most worthwhile a part of the AI worth chain depends upon the least worthwhile half persevering with to develop income or elevate capital,” he concluded. “Capital can bridge the hole for some time, however not indefinitely. And therein lies the chance: will the ROI present up for AI’s finish clients quick sufficient to maintain the spending that’s producing these upstream margins?”

Wider unfold fears of an unsustainable AI enlargement

Slok isn’t the primary economist to sound the alarm on AI’s outsized reliance on investments. In its annual report revealed in June, the Financial institution of Worldwide Settlements famous the onslaught of AI investing, primarily from the 5 main hyperscalers, is outpacing earnings and free money circulation, which has led to those corporations issuing debt to boost further financing. A Financial institution of America evaluation from final November discovered that in 2025, these 5 hyperscalers issued $121 billion in debt, 4 instances the typical debt ranges issued by these corporations yearly over the earlier 5 years. 

“Disappointment in returns might set off a sudden pullback in financing and switch the capex growth right into a protracted funding bust, with potential knock-on results on monetary circumstances,” Financial institution of Worldwide Settlements mentioned within the report. “Ought to hyperscalers sluggish or halt the aggressive tempo of capex deployment, many debtors throughout the provision chain might wrestle to interchange misplaced income and repair their debt.”

Tech author Ed Zitron took this concern a step additional, arguing AI spending is extra precarious than IT even seems on the floor. He used the instance of Oracle, which has a damaging money circulation of $23.7 billion, as of the top of fiscal 2026, and almost $130 billion in excellent debt and $260 billion in lease commitments for AI infrastructure tasks which have but to start. Its large gamble on AI buildout is in service of OpenAI, with whom IT signed a $300 billion deal final September.

“Oracle’s existence — and Larry Ellison’s private wealth — hinges on whether or not OpenAI could make good on its promise to spend $300bn in compute,” Zitron wrote in a Substack post in June. 

He known as this dynamic “most-obvious and under-discussed a part of the AI bubble”: Whereas 13-figure hyperscaler capital expenditures are fueling a semiconductor growth, there’s little proof thus far that the present AI growth will translate to widespread purposes of Technology that may justify all this spending.

Extra harmful than simply Oracle failing to ship on its $300 billion promise to OpenAI is different tech corporations giving up on their very own exorbitant AI spending, Zitron defined.

“If Microsoft, Google, Amazon and Meta determine that IT’s time to cease spending $30 billion or extra 1 / 4 on GPUs, RAM, storage, and knowledge heart building,” he mentioned, “that’ll tear a gap within the aspect of what individuals assume is a everlasting supercycle.”