AI billions, lousy margins. Can Supermicro profit from the data center boom?
Supermicro is one of the companies that have turned the AI investment boom into rocket-fueled revenue growth. Orders are measured in tens of billions of dollars, the backlog is at a record high, and demand for AI servers seems boundless. But the economics of the business look completely different from NVIDIA or other companies considered the main winners of the AI era. Supermicro operates at single-digit to low double-digit margins, consumes vast amounts of cash on working capital, and in the recent past has faced serious questions about accounting, internal controls, and export rules. The question for investors is therefore not whether Supermicro is growing. The question is whether it can turn that growth into real shareholder value.
Key points
Supermicro is a server integrator, not a chip manufacturer. According to management, revenue for FY2026 is expected to reach $38.9 to $40.4 billion, yet the gross margin still fluctuates between just 6 and 17 percent in individual quarters.
The backlog at the end of Q4 FY2026 jumped to a record of over $60 billion in new orders, while at the same time the margin for that quarter jumped to 15 to 17 percent from the originally guided 8.2 to 8.4 percent.
Free cash flow remains deeply negative, reaching approximately minus $6.85 billion over the last twelve months due to rapidly growing inventories and receivables, which are growing faster than revenue.
In March 2026, the company faces a new case involving export controls: the DOJ indicted three individuals connected with the company for the illegal export of $2.5 billion worth of AI servers to China; Supermicro itself was not indicted.
The valuation is significantly lower than Dell's; the shares trade at a forward P/E of around 8.9x versus 24.7x for Dell, reflecting both higher expected growth and a higher risk premium for margins, cash flow, and governance.
Supermicro $SMCI is an American manufacturer of servers and modular data center systems based in San Jose. Over the last three fiscal years, it has increased revenue from approximately $5.2 billion (fiscal year 2022) to $22 billion (fiscal year 2025), and according to current management, it is on track for revenue of around $39 to $40 billion in fiscal year 2026. At the same time, it is a company whose gross margin fluctuates between 6 and 17 percent quarter to quarter, well below the level the market typically associates with the technological winners of the AI revolution.
Business Model and Position in the AI Ecosystem
Before assessing whether Supermicro is a good investment, it is necessary to understand exactly what the company sells and where it stands in the artificial intelligence value chain.
What Supermicro Actually Sells
Supermicro does not develop chips and does not own any fundamental computing technology comparable to NVIDIA's GPU architecture. The company designs and manufactures server and rack-scale systems—complete computing units that integrate processors and graphics chips from NVIDIA $NVDA, AMD $AMD, or Intel $INTC together with its own motherboards, power supplies, chassis, cooling systems, and networking components. In other words, Supermicro takes finished silicon "brains" from other companies and builds around them the physical infrastructure that turns them into a functioning data center.
The portfolio includes, among others:
SuperBlade, MicroBlade, FlexTwin, GrandTwin, and BigTwin modules for dense computing deployments;
complete rack-scale AI clusters built on NVIDIA HGX and GB200/GB300 NVL72 architectures;
SuperStorage systems;
edge and 5G/IoT solutions.
Position Among Chip Makers, Server Vendors, and Data Center Operators
Simplified, the AI infrastructure value chain has three layers. At the top are chip designers like NVIDIA, who sell intellectual property transformed into silicon—the segment with the highest added value in the entire chain. In the middle are integrators and server manufacturers, including Supermicro, Dell $DELL, or HPE $HPE, who embed GPUs and CPUs into functioning systems. At the end of the chain are data center operators, i.e., hyperscalers such as Microsoft $MSFT, Amazon $AMZN, Alphabet $GOOG, or Meta $META, as well as specialized cloud AI capacity providers like CoreWeave $CRWV, who buy and operate servers.
The difference between individual server manufacturers is substantial:
NVIDIA designs chips and reference architectures (e.g., NVL72) and collects by far the highest margin in the entire chain from each system sold.
Dell Technologies and Hewlett Packard Enterprise are diversified IT giants with broad portfolios of computers, storage, networking, software, and enterprise services. AI servers are just one, albeit rapidly growing, segment alongside more stable and more profitable businesses such as enterprise support, financing, or networking solutions.
Supermicro, by contrast, is a virtually pure server hardware specialist. It has no diversification into software, cloud, or financial services, which means higher sensitivity to the server demand cycle but also an ability to respond faster to new generations of chips.