OCP 2026 峰会：Navitas、英飞凌、TI 展示 AI 数据中心 800V 供电方案

As AI workloads redefine the power requirements of modern data centers, power design solutions won a prominent spot at the Open Compute Project (OCP) Global Summit 2026 in San Jose, California. The data center power solutions showcased at the San Jose McEnery Convention Center ranged from 800 V-to-12 V power delivery boards to server power supply units (PSUs) to high-voltage hot-swap controllers.

Navitas Semiconductor displayed a 24-kW, 800 V-to-12 V DC-DC power delivery board it developed with Nvidia. The gallium nitride (GaN)-based 800 V DC power solution for AI data centers supports Nvidia’s MGX server hardware. The board was featured on Nvidia’s AI Factory MGX Wall partner display area at the OCP show floo`r.

The power delivery board, built around the company’s GaNFast power architecture, targets 97% efficiency at full load and a power density exceeding 2,000 W/in³. “The 24-kW solution demonstrates how GaNFast technology can eliminate power-conversion stages, increase efficiency and power density, and bring power closer to the GPU,” said Chris Allexandre, president and CEO of Navitas.

Another board-level power solution for AI data centers displayed at OCP was Infineon’s 27-kW, three-phase power supply unit (PSU). The reference board for AI servers meets the requirements of the OCP Open Rack V3 (ORV3) specification and AI server rack architectures.

By YMIN  10.19.2026

By Hynetek Semiconductor Co., Ltd.  10.15.2026

By Ayar Labs  10.14.2026

The board features a five-level ANPC power factor correction (PFC) topology and a three-level LLC converter, achieving a power density of 116 W/in³ versus the ORV3 minimum requirement of 94 W/in³. The PFC and LLC stages are digitally controlled to respond to demanding AI load transients.

The programmable power control accelerator (PPca), integrated into the PSOC Control C3 microcontroller, supports precise current and voltage regulation with fast dynamic response. In addition to PSOC microcontrollers, the reference board includes 650-V CoolSiC and CoolMOS MOSFETs, EiceDRIVER gate drivers, and XENSIV current sensors.

Another key feature of this reference board for AI servers is an integrated energy buffer that provides 20 milliseconds of hold-up time during short power grid disturbances. This helps smooth the line voltage during rapid GPU load changes, which in turn eliminates the need for a separate capacitor bank unit (CBU).

Power path protection and monitoring

As AI clusters scale from 48 V to higher rack power levels, they increasingly demand reliable hot-swap functionality to maintain uninterrupted system operation. “Future AI data centers require power delivery systems that combine high efficiency, maximum reliability, and comprehensive protection capabilities,” said Magdalene Boebel, senior vice president and business line head of power system ICs at Infineon.

At OCP, Infineon also unveiled a high-voltage hot-swap controller for digital power-path protection and monitoring in AI server power architectures operating at ±400 V and 800 V DC. The XDP8HD40S-1 enables power modules, converters, GPUs, and server subsystems to be connected and disconnected safely while preserving system integrity.

Infineon’s new hot-swap controller is designed for a range of high-voltage applications in AI data centers. Besides 400-V to 800-V single- or dual-side hot-swap implementations, that includes battery backup systems, high-voltage rack and server power delivery, peak shaving, and two-wire 800-V or ±400-V IBC systems.

Texas Instruments also showcased its first 800-V isolated hot-swap controller that delivers in-rush current and power path protection while enabling safe power insertion and fast fault isolation. The LMI800PS, available in a single high-density package, replaces a complex discrete solution that requires five chips.

It is a scalable solution for diverse rack architectures, supporting both ±400-V and 800-V distributions. In AI data center environments, this isolated hot-swap controller prevents a single load failure from disrupting the shared power bus, maximizing data center stability.

Nanosecond-level telemetry

Like Navitas, TI also showcased its 800 V DC power solutions on Nvidia’s MGX AI Factory Ecosystem wall. This included a high-performance 3 gigasamples per second (GSPS) analog-to-digital converter (ADC) that enables 24/7, low-power, event-based remote monitoring at nanosecond resolution.

In data centers, existing monitoring solutions often fail to detect nanosecond-level power-rail glitches, leading to costly, unplanned downtime. The ADC34DSO75 combines ultra-high-speed sampling with advanced on-chip digital signal processing in a small form factor to detect and capture transient fault events in real time.

“As AI workloads drive massive scale, the need for real-time visibility and architectural stability has never been greater,” said Karthik Vasanth, vice president and general manager for TI’s data, converters, and clocks business unit. Power chip vendors like TI are confident that high-performance ADCs and hot-swap controllers featuring high-speed telemetry, advanced power protection, and precision timing capabilities can help data center operators and hyperscalers optimize performance and minimize downtime.

As GPU performance increases and rack densities rise, data center operators face significant challenges regarding power instability and inadequate real-time monitoring. The power solutions showcased at the OCP Global Summit 2026, held Oct. 12–15, aim to close this technology gap by accelerating the deployment of the high-efficiency power systems AI infrastructure requires.

Read also:

Why Data Centers Must Become System‑Aware

Get a grip on your data center power efficiency

Amperesand Aims to Smooth AI Data Center Power Spikes

AI AND BIG DATA, POWER MANAGEMENT