The AI Profit Paradox: Why the Boom Risks a Massive Bubble

Apollo Chief Economist Torsten Slok warns that the AI boom is structurally unsound, as profits are currently funded by investor capital rather than customer revenue. While chipmakers see 41% margins, AI model developers face -59% operating margins, creating a precarious dependency on continuous venture and corporate funding to sustain the supply chain.

The AI trade has shifted from a productivity play to a capital-expenditure cycle. For the past two years, the market has rewarded the “picks and shovels”—the hardware providers—while ignoring the massive burn rates of the companies actually building the intelligence.

The Bottom Line

• Margin Inversion: The highest-margin sector (Silicon/Equipment) relies entirely on the lowest-margin sector (Models/Apps) to raise capital.
• Debt Escalation: Hyperscalers issued $121 billion in debt in 2025 alone to fund AI capex, four times their five-year average.
• Systemic Risk: A pullback in spending by the “Magnificent Seven” would likely trigger a debt crisis for secondary infrastructure borrowers.

The Margin Gap: Who is Actually Paying for AI?

In a standard business model, the company selling the end product to the consumer captures the highest margin. In the AI value chain, the opposite is happening. According to data from Pitchbook and Bloomberg analyzed by Torsten Slok, the “upstream” providers are the only ones seeing real returns.

But the balance sheet tells a different story for the developers. While Nvidia (NASDAQ: NVDA) and AMD (NASDAQ: AMD) enjoy massive premiums, the firms creating the actual AI applications—such as Anthropic and OpenAI—are operating at a loss. Slok calculates the operating margin for models and applications at -59%.

Here is the math: The profits at the top of the chain are not being generated by consumers paying for AI subscriptions or efficiency gains. Instead, they are being funded by shareholders and VCs pouring money into the application layer, which then spends that money on chips and cloud compute.

AI Value Chain Segment	Estimated Profit Margin	Primary Entities
Silicon & Equipment	41%	Nvidia (NASDAQ: NVDA), AMD (NASDAQ: AMD), Micron (NASDAQ: MU)
Energy & Grid	Moderate/Growing	Constellation Energy (NASDAQ: CEG)
Cloud & Compute	Variable	Microsoft (NASDAQ: MSFT), Amazon (NASDAQ: AMZN)
Models & Applications	-59%	OpenAI, Anthropic

The Hyperscaler Debt Trap

The risk isn’t just limited to startups. The “hyperscalers”—the five largest cloud providers—are increasingly leveraging their balance sheets to maintain the AI arms race. According to a Bank for International Settlements (BIS) report, AI investment is outpacing both earnings and free cash flow.

This has forced a pivot toward debt. A Bank of America analysis indicates that in 2025, these five firms issued $121 billion in debt. This is a staggering increase compared to the previous five-year average. When companies borrow to fund capital expenditures (capex) without a corresponding rise in productivity or revenue, they create a “capex bubble.”

The danger here is a feedback loop. If the ROI for the end customer doesn’t materialize, the hyperscalers will stop spending. Because the supply chain is so tightly integrated, a sudden halt in GPU procurement would leave infrastructure providers with massive, unserviceable debts.

The Oracle Gamble and the $300 Billion Question

Nowhere is this risk more concentrated than at Oracle (NYSE: ORCL). By the end of fiscal 2026, Oracle reported negative cash flow of $23.7 billion. The firm is currently carrying nearly $130 billion in outstanding debt and $260 billion in lease commitments for AI infrastructure that hasn’t even been built yet.

Much of this gamble is tied to a $300 billion deal signed with OpenAI last September. As tech analyst Ed Zitron noted, the stability of Oracle’s current financial position hinges on OpenAI’s ability to actually spend that compute budget.

But there is a broader macroeconomic headwind. Goldman Sachs projects AI investments to exceed $1 trillion in 2026, yet there has been no significant shift in general economic productivity outside of the “Magnificent Seven.” This suggests that AI is currently a cost center for the majority of the S&P 500, not a profit driver.

Market Implications: The Risk of a Protracted Bust

If the market realizes that AI applications cannot monetize their users fast enough to sustain the hardware spend, we won’t see a gradual decline. We will see a “sudden pullback in financing,” according to the BIS. This would turn the current boom into a protracted investment bust.

Institutional sentiment is beginning to reflect this caution. As noted by Bloomberg, the focus is shifting from “capacity” (how many chips can we buy?) to “utilization” (how much money are these chips making?).

The stability of the entire sector now rests on a single variable: the speed of ROI for the end customer. If the enterprise adoption of AI remains a novelty rather than a necessity, the “supercycle” will be revealed as a capital transfer from investors to chipmakers, with no sustainable engine to power it.

Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.