Oracle Corp. has ridden a wave of investor enthusiasm this year, emerging as the top-performing megacap stock in 2025. Executives at the database giant last month touted projections of $381 billion in revenue from renting out specialized cloud servers to artificial intelligence developers like OpenAI, positioning the company as a key player in the AI infrastructure boom.

But beneath the rosy forecasts, internal documents reveal a starkly different picture of profitability. Oracle’s fast-growing cloud business, fueled by rentals of Nvidia Corp.’s high-demand chips, has operated on razor-thin gross profit margins over the past year, sometimes dipping into losses, according to a report by The Information.

Struggles with Small-Scale Rentals and High Costs

In the three months ending in August, Oracle generated about $900 million in revenue from servers powered by Nvidia chips, yet it eked out only $125 million in gross profit—a margin of roughly 14%, far below the company’s overall cloud margins of around 70%. The challenges stem from the economics of renting out these pricey graphics processing units, or GPUs, which are essential for training and running AI models but come with hefty upfront costs.

For smaller-scale rentals, the situation is even bleaker. Oracle has incurred significant losses on deals involving fewer than 1,000 GPUs, with some contracts resulting in negative gross profits due to high energy, maintenance, and depreciation expenses that outpace rental fees, as detailed in the same The Information analysis of internal data.

Losses on Nvidia’s Cutting-Edge Blackwell Chips

Particularly telling is Oracle’s experience with Nvidia’s latest Blackwell chips, which are among the most sought-after in the AI space. In its most recent quarter, the company lost nearly $100 million from renting out access to these advanced processors, per insights from Sherwood News, highlighting the financial strain of scaling up amid intense competition.

This revelation sent Oracle’s shares tumbling as much as 7% on Tuesday, erasing some of the gains that had propelled the stock higher on AI hype. Analysts note that while demand for AI computing power remains voracious, providers like Oracle must navigate a market where Nvidia holds significant pricing power, often leading to squeezed margins for cloud intermediaries.

Broader Implications for AI Infrastructure Providers

The thin margins underscore a broader tension in the AI ecosystem: Companies are racing to build out massive data centers stocked with Nvidia hardware, but the path to profitability is fraught with risks. Oracle’s push includes plans for enormous investments in infrastructure, yet the internal figures suggest that without larger, longer-term contracts, the economics may not add up quickly.

Competitors like Amazon Web Services and Microsoft Azure face similar dynamics, though their scale might afford better bargaining leverage with chip suppliers. As Seeking Alpha reported, Oracle’s documents indicate that gross margins in this segment have hovered in the low teens, prompting questions about the sustainability of its aggressive growth targets.

Investor Reactions and Future Outlook

Wall Street insiders are now recalibrating expectations. While Oracle’s executives remain bullish, emphasizing multiyear deals with big clients that could improve margins over time, the leaked data has sparked debates about whether the AI gold rush will yield enduring profits or just fleeting revenues.

For industry observers, this serves as a cautionary tale about the capital-intensive nature of AI infrastructure. Oracle may need to secure more high-volume commitments to offset costs, but in a market where chip shortages persist, the financial tightrope walk continues. As the sector evolves, such disclosures could influence investment strategies, reminding stakeholders that hype alone doesn’t guarantee healthy bottom lines.