AI Servers Sell Out! Foxconn Industrial Internet Earns 131 Million RMB Daily in First Half, but Net Profit Margin at 4.26% — Experts: Pricing Power in AI Chain Lies Entirely Upstream

Foxconn Industrial Internet's latest half-year report reveals staggering numbers: revenue of 557.861 billion RMB (all figures in Chinese yuan unless specified) in the first half of this year, up 54.64% year-on-year. Net profit attributable to parent company reached 23.7 billion RMB, equivalent to a daily net profit of 131 million RMB. AI server revenue increased 2.3 times year-on-year, and GPU AI rack shipments rose 3.2 times. However, gross margin was 7.15%, and net margin was only 4.26%.

Jingjing Jian, Partner and Head of the AI Team at MIR睿Industrial, revealed the core issue: "Companies like FII earn assembly service fees—hard-earned money from system integration. The real high margins lie upstream—in GPUs, HBM, high-speed networking, liquid cooling, and power distribution systems."

Over 55% of an AI server's cost is attributed to the GPU. FII handles assembly, testing, logistics, and after-sales, achieving a gross margin of around 7%. After expenses, net profit margin drops to just over 4%. In contrast, NVIDIA’s B200 chip has a manufacturing cost of about $6,400 but sells for $30,000–$40,000, yielding a gross profit of over $20,000 per chip. The net profit FII earns from assembling an entire server may not even reach a fraction of the gross profit from a single GPU.

Jian stated bluntly: "An H100 sells for tens of thousands of dollars, while FII earns just a few percentage points per unit assembled. This is the reality of the supply chain."

Contract manufacturing is the thinnest-margin segment, with pricing power held by chip and key component suppliers. While computing capacity continues to expand, returns are outpacing commercialization.

Jian emphasized: "Intelligent computing is not a bubble." Beyond cloud giants, traditional enterprises are beginning to build private computing infrastructure. North America’s four major cloud providers spent $131.6 billion in capital expenditures in Q1 (up 70.3% YoY) and approximately $163.9 billion in Q2 (up 86% YoY). Full-year guidance has been raised to $745 billion. China’s investment in computing infrastructure under the '15th Five-Year Plan' is expected to reach 4 trillion RMB.

However, most projects currently have low return on investment (ROI) because AI commercialization is not yet mature. Jian noted: "Demand exists, but infrastructure deployment is ahead of monetization."

The next high-margin frontier lies in power architecture. If 800V DC + SST (solid-state transformers) becomes the mainstream power solution for AI factories, Eaton, the U.S. smart power and power management company, has already partnered with NVIDIA to develop a dedicated power solution for NVL72. While still in technical validation and not yet widely commercialized, a breakthrough would unlock a new high-margin component market.

Jian said: "Whoever controls scarce resources earns the profits. When shovels are scarce, shovel sellers profit; when shovels are abundant, users profit. Right now, upstream players are eating meat, midstream players are sipping soup—and this will continue for some time. Short-term returns may not be optimistic, but not investing means falling behind. This is a decade-long infrastructure race."