IN Brief:
- Second-quarter revenue increased 9% year on year to $1.604 billion.
- Platform work includes EliteSiC devices, silicon MOSFETs, controllers, and GaNEXUS products spanning 40V to 650V.
- Third-quarter revenue is expected to reach between $1.65 billion and $1.75 billion.
onsemi expects its AI data-centre business to more than double during 2026 as high-voltage power products gain positions across several stages of the rack conversion chain.
Second-quarter revenue reached $1.604 billion, 9% higher than a year earlier and 6% above the first quarter. The Power Solutions Group generated $829 million, rising 19% year on year and 13% sequentially, making it the strongest-performing of the company’s three operating segments.
onsemi cited growing adoption of EliteSiC silicon-carbide devices, silicon MOSFETs, and controllers in AI data-centre platforms. It has secured work with Great Wall, a Chinese cloud-infrastructure power supplier, and expanded its position within the NVIDIA MGX ecosystem as rack-level power requirements continue to rise.
The company has also introduced GaNEXUS, a gallium-nitride portfolio spanning 40V to 650V. The range is intended to address several conversion stages rather than a single device class, with applications across data centres, robotics, and industrial infrastructure.
The breadth matters because an AI rack does not rely on one power semiconductor. Facility power is converted through front-end stages, distributed at higher DC voltages, stepped down through intermediate buses, and finally regulated close to processors and accelerators. Each stage operates under different voltage, current, switching-frequency, isolation, and thermal conditions.
Silicon carbide can suit higher-voltage conversion where efficiency, blocking capability, and thermal performance are priorities. Gallium nitride can support high switching frequencies and smaller magnetics in selected stages, while conventional silicon remains competitive where cost, maturity, ruggedness, and control integration outweigh the benefits of a wide-bandgap material.
Device selection therefore follows the converter topology rather than the marketing label attached to the semiconductor. Engineers must still account for gate-drive behaviour, parasitic inductance, electromagnetic interference, short-circuit response, cooling, package resistance, and the interaction between switching edges and surrounding magnetics.
A faster transistor can expose a weak layout more efficiently. Wide-bandgap devices place greater demands on board geometry, package design, drive control, and measurement because voltage and current can change more rapidly than in established silicon implementations.
Packaging will determine how much of the material advantage reaches the finished system. Bond wires, leadframes, substrates, cooling interfaces, and module inductance can erase gains available at die level, particularly as current slew rates increase. Qualification must also cover repetitive switching stress and the operating profiles expected across a data-centre service life.
Production test is another constraint as devices enter higher-volume infrastructure programmes. Recent GaN test-system orders for AI power devices show that measurement accuracy, high-current handling, low-resistance characterisation, and multi-site throughput are becoming part of the manufacturing problem rather than an afterthought once the device has been designed.
The Great Wall and NVIDIA-related positions indicate that onsemi is participating in platform programmes rather than selling only isolated catalogue parts. The company has not disclosed the revenue contribution, device volumes, deployment schedules, or precise power stages covered by those wins.
Its forecast that AI data-centre revenue will more than double is therefore a management expectation rather than a customer-backed volume disclosure. Demand will depend on platform schedules, accelerator availability, data-centre construction, and the rate at which higher-voltage rack architectures move into production.
Automotive remains another substantial market. onsemi said its power products had extended their role in Rivian’s R2 platform, covering zonal architecture and onboard charging. That diversification matters because the capital and process capability required for power semiconductors cannot sensibly depend on one infrastructure cycle.
Second-quarter gross margin was 38.4%, compared with 37.6% a year earlier, while free cash flow reached $425.4 million. The company expects third-quarter revenue between $1.65 billion and $1.75 billion, with a non-GAAP gross margin of 40% to 42%.
Those figures give onsemi room to expand its portfolio, but the technical evidence will come from qualified designs, field reliability, and sustained orders across several conversion stages. AI racks provide the headline; the underlying work remains lower losses, controlled switching, repeatable packaging, and devices that survive the voltage and temperature they were specified to handle.
