2026-11-12 16:00

Introduction: NVIDIA’s 800V architecture fuels a new battleground in power semiconductors. On November 10, Infineon released its Q4 and full-year 2026 financial results: Q4 revenue reached €4.706 billion, with full-year revenue totaling €16.353 billion, up 12% year-on-year. More telling than the numbers is the structural shift—Power & Sensing Systems (PSS) segment generated €1.796 billion in Q4, a 48% YoY increase, with a segment margin of 30.8%. Infineon also disclosed for the first time that dedicated AI data center power revenue amounted to approximately €1.7 billion in FY2026, with total data center-related revenue reaching around €2.2 billion, and the company anticipates sustained strong demand for data centers in 2027. Just two weeks prior to this report, on October 28, Infineon announced an investment of approximately €1.1 billion in its Philer facility in Austria to build the world’s first mass-production factory for 300mm (12-inch) gallium nitride (GaN) power semiconductors, expected to come online by end-2027. “The limit of computing power is power supply.” Summarizing the context of the AI race, Infineon Greater China President Pan Dawei noted that the frontier has shifted from chip process technology to power delivery—GaN is evolving from a “trending material” in charger markets into a foundational choice for data center power architectures. This is precisely the industrial signal pointed to by both Infineon’s financial performance and its capacity expansion.

AI power has become the primary growth driver. The past quarter marked the most impressive period in Infineon’s PSS division history. Company investor materials indicate that the restructured Power Systems business unit generated €5.434 billion in FY2026, with AI and data center applications accounting for 40% of revenue. Infineon CEO Jochen Hanebeck cited “data center power” as the top growth catalyst in his earnings statement. Behind this surge lies a steep curve of compute demand. According to data from China’s National Data Administration and Ministry of Industry and Information Technology, daily token requests in China surpassed 100 billion in early 2024, jumped to 100 trillion by end-2025, and exceeded 500 trillion by June 2026—representing thousands of times growth over two years. Based on this trajectory, Infineon estimates global AI data center capacity will reach 156 GW by 2030. Faced with escalating GPU and rack-level power demands, traditional power architectures are hitting severe bottlenecks, fueling explosive growth in AI data centers. However, Pan Dawei emphasized that the real constraint lies in the market’s ability to deliver corresponding compute capacity: “Three months ago, customers requested one number; now it’s another.” He revealed that power-related orders have piled up so much that they are now constrained by the physical pace of data center construction. Under growing “power scarcity” anxiety, upstream power devices are undergoing unprecedented valuation re-rating.

€1.1 Billion Plant Investment: Betting Big on GaN—Opportunity Lies in Power Losses. Inside AI data centers, electrical energy undergoes multi-stage conversion: “solid-state transformer → power supply unit (PSU) → intermediate bus converter (IBC) → GPU core power.” Each stage incurs efficiency losses that get amplified into compute costs. GaN offers faster switching speeds and lower losses, enabling significant miniaturization of power supplies under high-frequency operation. Citing Yole Group data, Infineon notes that the GaN power semiconductor market will grow to approximately $3.5 billion by 2031, with a CAGR of around 35% between 2025 and 2031. The 300mm wafer size is the technological cornerstone of this bet. Two years ago, Infineon successfully achieved the world’s first 300mm GaN wafer production—this larger diameter increases chip yield per wafer by 2.3x compared to 200mm wafers. Moreover, since GaN manufacturing processes closely resemble those of silicon, Infineon can directly repurpose its existing 300mm silicon production line equipment. By leveraging its current cleanroom infrastructure in Philer, the company performs a “cage replacement” strategy, using roughly €1.1 billion to unlock large-scale manufacturing. In investor materials, Infineon refers to this new facility as the “first commercial-scale 300mm GaN wafer fab scheduled for launch in 2027,” effectively setting a clear market timeline commitment.

Discovering New Demand Beyond the Rack. Indeed, beyond data centers, embodied robotics is emerging as an unexpected yet definite growth scenario for GaN. Pan Dawei revealed that Infineon’s GaN power market share has risen from fifth globally in 2024 to second globally, with first 300mm samples delivered to customers in Q4 2025. Real-world test data from the past year shows: replacing silicon devices with GaN in robot joints reduces motor drive size by about 30–40%, improves joint module efficiency by 5% to 20%, and lowers device temperature rise by approximately 10°C—directly translating into extended battery life and reduced thermal management costs in humanoid robots with limited battery capacity. “If GaN can truly solve the pain points we face in robot joints, even at a slightly higher price, enterprises will still choose it.” In Pan’s view, GaN enables fewer passive components like capacitors on PCBs under high-frequency designs, thereby lowering system-level costs. Nevertheless, GaN is not universally applicable. Infineon adopts a three-pronged strategy: silicon, silicon carbide (SiC), and GaN. Late October saw Infineon partner with NVIDIA to advance the 800VDC power architecture for AI data centers. Additionally, through prior agreements on 200mm SiC wafers and the acquisition of C2i Semiconductors, Infineon has established a full-stack solution—from grid-side solid-state transformers (SST) to 800V rack-level high voltage, down to vertical power delivery to core chips. Pan stated that for the same power level, “one customer may use SiC, another GaN”—depending on their design choices. “It’s difficult to define a fixed GaN share in future data centers.” In reality, the true variable lies at the architectural level: traditional transformers are evolving toward SST. Pan revealed that seven or eight leading SST vendors are already publicly collaborating with Infineon, but widespread adoption of SST “will take some time,” and the current transition phase relies on customized modular container units. With the 300mm GaN production plant scheduled to go live by end-2027, Infineon positions itself ahead of most competitors who remain stuck on 200mm lines, entering the scale-cost sweet spot early—a move aligned with internal projections that mass production will bring GaN solutions’ cost close to silicon-based alternatives. Clearly, from exceeding €2.2 billion in data center revenue to committing €1.1 billion to build a GaN fab, Infineon sees a “power revolution” surging behind compute supply—and recognizes that the competition in power semiconductors has now intensified at the levels of fundamental materials and large-scale manufacturing capacity. It has chosen to bet on wafer size to sprint ahead. (Originally published on Titanium Media APP, by Qian Fang)

Source: Qian Weijing