AI server business is booming, but all the money is being made by Nvidia.
```
Amid the surging demand for artificial intelligence, major global AI server manufacturers are facing a common challenge: despite significant revenue growth, profit margins continue to shrink.
As companies like HPE, Dell, and Supermicro release their financial reports, a harsh reality has come to light: despite soaring AI server orders, the high cost of Nvidia chips and fierce market competition are making hardware manufacturers’ profit margins extremely thin.
HPE’s third quarter financial results released on Wednesday showed the company’s revenue grew 18% year-on-year to $9.14 billion, with earnings per share of $0.44, both exceeding analysts’ expectations. But its server division’s operating margin dropped from 10.8% in the same period last year to 6.4%.
Facing investors’ concerns, HPE CEO Antonio Neri sought to reassure the market during an analyst call, promising margins would recover to about 10% by the end of the current fiscal quarter. This temporarily stabilized the stock, with HPE shares rising 1.5% in after-hours trading.
This trend is spreading throughout the industry, and the fundamental reason is that the core of AI servers—high-performance GPU chips—are almost monopolized by Nvidia. These expensive chips make up the bulk of server costs, leaving most of the value with the upstream supply chain.
The Dilemma of “Revenue Without Profit”
For server original equipment manufacturers (OEMs), the current AI market presents an awkward situation of “increasing revenue without increasing profit.”
In addition to HPE, whose margins have almost been “cut in half,” another server giant, Supermicro, is facing the same challenge. Although its fiscal 2025 Q4 revenue soared 46.59% year-on-year, its gross margin dropped to 9.7%.
Dell faces a similar situation; its fiscal 2026 Q2 gross margin fell from 22% in the same period last year to 18.7%, which the company attributed to pricing pressure from AI servers.
This phenomenon of "high revenue, low profit" is becoming a common dilemma for AI hardware manufacturers. In stark contrast to the meager profits of server makers, chip giant Nvidia enjoys staggering profitability.
Nvidia, with a data center GPU market share as high as 98%, has absolute pricing power. Its fiscal 2026 Q2 report showed a non-GAAP gross margin as high as 72.7%, several times higher than server manufacturers. Furthermore, data shows its latest Blackwell GPU platform can achieve a profit margin of 77.6% on AI inference workloads.
This stark disparity in profit distribution clearly illustrates the current state of the AI value chain:
Nvidia: gross margin over 70%.
Dell: gross margin about 18.7%.
HPE: server division’s operating margin fluctuates between 6.4% and 10%.
Supermicro: gross margin falls to 9.7%.
The Pressures Behind Declining Profits
The continuous pressure on server manufacturers’ profits mainly arises from three structural factors.
First is the high cost of components. The core of AI servers is Nvidia’s GPU, which is expensive and in short supply, leaving OEMs with almost no bargaining power. A report even points out that in the cloud service provider sector, for every $7.90 of AI hardware revenue, OEM hardware makers might lose $1, highlighting the asymmetry in the cost structure.
Second is fierce market competition. To gain market share, server manufacturers have waged intensive price wars. Offering deep discounts to win big customer orders has further eroded already thin profits. For example, Dell’s Infrastructure Solutions Group’s operating margin has fallen to 8.8%.
Lastly is complex supply chain management. In order to meet the urgent delivery needs of AI components, manufacturers are forced to bear additional logistics costs, while challenges in inventory management further increase operating costs and profit pressure.
For investors, the growth story of AI server makers is becoming increasingly complex. Although these companies’ revenue forecasts are very optimistic, sustainability faces severe tests. Analysts warn that if companies like Supermicro maintain long-term profit margins of only 10-11%, it will be insufficient to support long-term technological innovation or provide substantial returns to shareholders.
In this grand game of AI, hardware assemblers are more like “porters,” while Nvidia is the true “big winner.”
Risk disclaimer and noticeThe market has risks; investment requires caution. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific situation. Investment decisions made on this basis are at your own risk. ```