AI Is Getting Cheaper, But Deployment Costs Won’t Drop! NVIDIA’s Earnings Support Bullish Sentiment, Yet Market Must Carefully Calculate This Equation

AI prices are falling rapidly, yet deployment costs continue to rise. As model prices drop, chip and memory supplies tighten, and AI infrastructure projects increasingly rely on debt financing, the market has gradually shifted its focus this summer from "Will AI succeed?" to a more pressing question: Who will ultimately profit from the AI industry?

According to Jan-Patrick Barnert, Senior Strategist at Bloomberg, several recent developments over the past ten days have made this issue increasingly prominent.

First, a free AI model called Ox Alpha recently launched with performance comparable to cutting-edge models, yet its developer remains unknown. At the same time, OpenAI has cut prices for its flagship model for the third time in about a month, and AI token price indicators continue to decline—suggesting that "intelligence" as a commodity is rapidly losing value.

However, Barnert argues that falling AI prices alone do not constitute a major warning. AI price indicators reflect both usage volume and pricing, measuring more closely what the market is willing to pay for AI rather than just list prices.

More importantly, price declines have historically accompanied technological advancement. For new technologies to achieve mass adoption, usage costs must continuously decrease, and next-generation chips can significantly reduce the production cost per AI token. As long as AI production costs fall faster than selling prices, price declines can actually benefit industry expansion.

Currently, however, the market is beginning to show signs of this "arithmetic reversal." The semiconductor industry is entering a phase of structural supply shortages, with tight supply across foundry and memory segments, implying that computing resource prices may remain high. Bloomberg previously reported that some of NVIDIA’s largest customers have been informed that AI server prices for deliveries early next year could rise by over 15%.

NVIDIA’s latest financial outlook remains robust, with the company projecting approximately 70% revenue growth for its fiscal year 2028, indicating strong ongoing AI demand. On the other hand, rising memory prices are beginning to erode industry profit margins. The company has warned that increasing memory costs could pressure gross margins.

Memory suppliers are also raising costs. Samsung Electronics has increased prices for new advanced chip foundry orders by up to 15% and signed multi-year contracts with data center clients, reserving up to 70% of its memory capacity. SK Hynix warns that memory shortages could worsen by 2027 and is exploring joint ventures to fund new factory construction.

This creates a fascinating dynamic: AI product prices continue to fall, while the chips, memory, and other resources needed to produce AI are rising in price due to supply constraints.

Of course, the AI bull case still has a strong counterargument: volume.

Lower prices may lead to higher usage. In the U.S., nearly 60% of enterprises now pay to use AI, and AI spending across all expenditure tiers has more than tripled compared to the past. The three major hyperscale cloud providers generated approximately $106 billion in combined cloud revenue last quarter, a year-over-year increase of over 40%.

According to estimates by Ramp, the top AI-spending enterprises spend about $7,400 per employee per month on AI—though this figure may be inflated. However, the median spending across all enterprises is about $12 per employee per month, suggesting significant room for AI adoption to expand.

From this perspective, falling AI prices may be central to the industry’s business model. Cheaper AI drives greater usage, massive usage supports overall market scale, and ultimately, a growth cycle of "price decline → demand explosion → total revenue increase" could emerge.

The problem, however, is whether this revenue can translate into real returns on investment for enterprises. This is the hardest question the market faces today.

Milos Maricic, founder of AI consultancy Maximand, studied U.S. financial firms, analyzing 919 earnings call transcripts over three years from 60 large U.S. listed financial companies.

He found that about four-fifths of the calls mentioned AI, and over half discussed AI-related costs. Yet only one company clearly stated how many dollars in profit AI had generated—and the total return mentioned across two calls amounted to about $19 million.

This highlights the biggest contradiction in AI investment today: while executives, investors, and media interviews are filled with stories of AI-driven efficiency and growth, actual earnings calls still contain very few quantifiable examples of AI ROI.

Maricic bluntly states that after three years of large-scale AI deployment, enterprises purchasing this technology still cannot clearly articulate how many dollars in return AI has delivered.

As AI investment continues to rise while returns remain unquantifiable, credit markets are naturally becoming more cautious. Bloomberg reports that Broadcom is negotiating to raise over $60 billion in debt to fund AI chip-related investments. The entire industry is experiencing a financing boom, with companies borrowing heavily to bet on future computing demand.

However, bond markets are demanding higher risk premiums. The cost of credit default insurance on Broadcom’s debt is currently about 80 basis points higher than in January, and the increase is accelerating.

Rich Privorotsky, Head of Europe One-Delta Trading at Goldman Sachs, points out that if the market begins to believe companies may ultimately fail to fund all planned AI infrastructure, they will face a choice between increasing equity financing or cutting capital expenditure. Both outcomes make higher stock valuations difficult to sustain and are becoming a core contradiction in this summer’s AI rally: earnings estimates continue to rise, yet price-to-earnings ratios are beginning to contract.

Previously, investors focused on whether AI machines could run and whether demand would persist. Now, the market is asking: who will ultimately earn sufficient returns from massive AI capital spending?

NVIDIA’s latest earnings report has given bulls some breathing room, with strong revenue guidance proving that AI demand is still growing rapidly. But with rising input costs like memory, margin pressures are beginning to surface.

Can the AI price war be resolved by demand growth?

Bulls can still hope that volume will solve the problem—falling AI prices may drive widespread enterprise adoption, eventually creating enough demand to cover infrastructure costs. Just as e-commerce endured years of losses before achieving massive profits, AI’s returns may simply not yet be fully reflected in corporate earnings.

But another scenario is equally possible: if AI product prices continue to fall at their current pace, while input costs for chips and memory remain high due to supply shortages, and companies must rely on more expensive debt to fund capital spending, and AI-using customers cannot demonstrate sufficient profitability, the industry’s financial structure will need to satisfy too many conditions simultaneously.

Barnert notes that in July, the market was still debating whether AI could maintain pricing power. Now, new evidence shows that AI pricing power is under pressure from both ends—buyers demanding lower prices, while suppliers continue to raise input costs.

The market has not yet seen a full-scale panic or formed a clear bearish consensus, but stock valuations are beginning to reflect greater outcome divergence. Investors are paying higher risk premiums for this uncertainty. How this arithmetic ultimately resolves will determine how much further the next phase of the AI rally can go.