The largest AI profits are accumulating furthest from the customer, while the companies closest to the end user are absorbing the losses.
That verdict is the latest insight from Apollo Chief Economist Torsten Slok. Per his estimates, margins for silicon and equipment stand at 41%, while models and applications serving the end users (including frontier labs such as OpenAI and Anthropic) operate at a -59% margin.
“AI boom’s profits are currently being funded by investors rather than earned from customers,” Slok wrote. “The upstream margins are real, but they are paid for out of capital raised by the layer losing money, not out of cash generated by end demand,” he warned.
According to Goldman Sachs Research, AI investment will total around $1 trillion globally and just under $600 billion in the U.S. in 2026. Such spending puts hyperscalers, frontier labs, and hardware & infrastructure providers in a three-way standoff that neither can step away from unscathed.
Risk or Concede
The first pressure points are the frontier labs. A combination of long-term computing constraints is clashing with revenue models unproven at the scale of those obligations. OpenAI’s reported $300 billion multiyear agreement with Oracle Corporation (NYSE:ORCL) is perhaps the best example.
Adoption is expanding, but cash generated per unit of compute has yet to match the cost of training and operating frontier systems.
Microsoft Corporation (NASDAQ:MSFT), Alphabet Inc. (NASDAQ:GOOG), Amazon.com, Inc. (NASDAQ:AMZN), Meta Platforms, Inc. (NASDAQ:META) and Oracle are driving annual capital expenditure toward roughly 3% of U.S. gross domestic product from 2027 through 2029, according to Apollo — more than twice the late-1990s telecom peak as a share of the economy.
They issued $121 billion of debt in 2025, four times their average annual issuance over the preceding five years, according to a Bank of America analysis.
Slowing investment risks conceding capacity to competitors, while continuing to spend risks diluting returns on capital if AI revenue fails to scale.
Hardware suppliers sit at the other end. Their pricing power is substantial, but revenue is concentrated among the same buyers whose spending is increasingly debt- and equity-financed. Customers are also seeking alternatives.
Meta Platforms, Inc. and OpenAI have each agreed to deploy as much as 6 gigawatts of Advanced Micro Devices, Inc. (NASDAQ:AMD) Helios capacity under multiyear arrangements. At the same time, Elon Musk has gone all-in on NVIDIA Corporation (NASDAQ:NVDA).
Bottlenecks Make the Cycle Less Flexible
The infrastructure layer has genuine scarcity. Taiwan’s advanced semiconductor ecosystem alongside South Korea’s memory supply is central to large-scale AI development.
Still, infrastructure lacks flexibility. Semiconductors, memory capacity, power, and cooling require multiyear commitments and significant upfront investments.
But demand can change faster than factories can be repurposed. Export controls and geopolitical tensions only compound those risks.
What Breaks the Standoff
The clearest near-term risk is concentration. Oracle had approximately $130 billion of outstanding debt, $260 billion in long-term lease commitments, and negative cash flow of $23.7 billion at the end of fiscal 2026.
Its exposure to OpenAI’s ability to fulfill a $300 billion compute agreement is unusually direct.
A broader retrenchment would be more consequential. The Bank for International Settlements warned that disappointing returns could trigger a sharp financing pullback and turn the capex boom into a prolonged investment bust.
Thus, if hyperscalers slow deployments, suppliers across the entire upstream would face lost demand while carrying investment and debt burdens. To avoid that scenario, enterprises must demonstrate measurable productivity gains and margin improvement that justify such spending.
Until then, investors should watch enterprise software monetization and return on investment; hyperscaler free-cash-flow conversion relative to depreciation and whether multigigawatt commitments become active.
It’s a tall order, even for seasoned investors, but investing through the largest capex cycle in history requires vigilance of a similar order.
Image via Shutterstock
