AI Pricing Collapses While Chip Costs Surge, Squeezing Margins

The economics of artificial intelligence are entering uncharted territory. AI model prices are collapsing at unprecedented speed — OpenAI has cut flagship model prices three times in roughly a month — while the cost of building AI infrastructure is climbing sharply. This inverse relationship is forcing investors to recalculate who profits from the technology boom.

Why it matters

This isn't abstract market volatility. The gap between falling AI prices and rising infrastructure costs directly threatens the business model underlying hundreds of billions in data center investments. If companies can't demonstrate returns while their input costs escalate, the entire AI capital cycle faces a reckoning that could reshape technology spending priorities across industries.

The cost squeeze intensifies

The semiconductor market is entering what industry analysts call structural undersupply. Nvidia has informed major customers that server prices containing its AI chips will increase more than 15% for systems shipping early next year, according to Bloomberg News reporting. Samsung Electronics has raised advanced chipmaking prices up to 15% for new orders and locked 70% of memory capacity into multiyear contracts. SK Group's chairman has warned the memory shortage will worsen through 2027.

Meanwhile, the Silicon Data LLM Token Expenditure Index — which tracks what buyers actually pay for AI inference — continues sliding. A free model called Ox Alpha recently appeared online performing near cutting-edge capabilities, with no disclosed creator. The product is deflating while raw materials face rationing and upward repricing.

The missing returns

Three years into massive AI buildouts, corporate buyers still cannot quantify financial returns. Research by Milos Maricic, founder of AI advisory firm Maximand, examined 919 earnings calls from the 60 largest U.S.-listed financial firms over three years. While four in five mentioned AI and more than half discussed costs, only one firm twice highlighted realized dollar returns — totaling approximately $19 million combined.

"Three years into the AI buildout, the firms buying the technology still cannot put a dollar figure on the payoff," Maricic said, as reported by Bloomberg.

Cloud revenue tells a different story. Three major hyperscalers recorded combined cloud revenue of roughly $106 billion last quarter, up over 40% year-over-year. Business AI adoption approaches 60% of U.S. companies, with spending tripling across distribution levels. Yet this volume growth hasn't translated into measurable productivity gains customers can report to investors.

Credit markets signal doubt

Debt markets are pricing increased risk. Broadcom is negotiating to raise more than $60 billion in debt for AI chip funding, part of broader infrastructure financing. Credit default swap costs for Broadcom debt have risen approximately 80 basis points since January, with the premium accelerating.

Rich Privorotsky, Goldman Sachs' head of European one-delta trading, noted that if companies cannot finance planned infrastructure, "the answer is either more equity or less capex. Neither deserves a higher multiple." Stock multiples have contracted even as earnings estimates rise — a market acknowledging uncertainty about who captures value.

Nvidia's recent earnings provided some reassurance with strong guidance, projecting approximately 70% revenue growth in fiscal 2028. However, the company warned about potential margin compression as memory costs increase.

The bull case rests on volume overwhelming cost pressures — that cheap intelligence drives adoption sufficient to justify infrastructure spending. The alternative is a product deflating faster than production costs can fall, funded by increasingly expensive debt, sold to customers unable to measure returns. That requires numerous variables aligning simultaneously.

These details were first reported by Bloomberg News.