As AI and machine learning workloads accelerate, data center power consumption is beginning to outstrip existing infrastructure capacity. To meet this rising demand, new high-voltage DC standards support the higher-power, denser server racks now found at gigawatt-scale facilities. These high-voltage standards create engineering challenges when monitoring high-voltage power rails.

Designers need reliable, accurate, and fast-acting voltage supervision to prevent overvoltage damage to downstream components, and to help ensure a timely system response to undervoltage conditions. This article presents a supervision approach that addresses these requirements and enables the reliable deployment of next-generation high-voltage DC architectures.

 The push toward high-voltage DC architectures

The power profile of modern data centers is undergoing a dramatic shift as AI becomes the dominant application. Machine learning with large graphics processing unit arrays consumes power at levels once associated with industrial equipment rather than IT hardware. It is increasingly common for a single rack to draw 60 kW to 100 kW. Next‑generation AI systems are expected to push beyond 150 kW per rack.

Because traditional 48-V distribution designs cannot efficiently support these levels, designers are turning to a new class of high‑voltage DC standards centered around ±400 V or 800 V distribution. This shift, as shown in Figure 1, is not simply an incremental upgrade; it represents a fundamental change in the delivery of power across gigawatt‑scale facilities.

Figure 1 Conventional versus high-voltage data center power distribution. (Source: Texas Instruments)

Efficiency continues to drive the transition to higher voltages. Higher voltages reduce current and the I²R losses that dominate high power distribution, while also substantially cutting current and reducing conduction losses in cables, busbars, and connectors. Higher efficiency at large AI campuses means lower cooling requirements, improved energy performance, and increased computing density.

Higher voltages also unlock greater power‑delivery capability. Delivering 150 kW to 300 kW per rack at 48V requires heavy conductors, parallel cabling, and complex routing. High voltages deliver the same power with manageable current levels, enabling simpler infrastructures and longer distribution distances without excessive copper mass.

Cost provides yet another compelling factor. Smaller conductors, lighter busbars, and reduced copper usage lower material and installation expenses. At modern hyperscale data center campuses, these reductions are substantial.

Challenges in monitoring high-voltage power rails

As data‑center power architectures migrate toward higher‑voltage DC distribution, the need for monitoring and protection circuitry increases significantly. Higher-voltage DC distribution increases demands on monitoring and protection circuitry. Operating at ±400 V or 800 V means that every disturbance or transient condition carries more stored energy, with components operating closer to their absolute limits. These conditions reduce the margin for error and make precise power‑rail supervision essential.

Designers must contend with higher fault energy levels, faster electrical dynamics, increased electromagnetic noise, and tighter system‑level coordination requirements. In this environment, monitoring circuits must distinguish between harmless fluctuations and true fault conditions, with far greater accuracy and speed than lower‑voltage systems.

With these broader challenges in mind, let’s look more closely at two specific issues surrounding under- and overvoltage events:

• Response time. The voltage monitor must respond to faults fast enough to prevent damage to downstream components, but should not trigger erroneously from a noisy environment or short transient voltage fluctuations. For example, imagine a large current spike causing the supply voltage to drop while the power supply responds. If the voltage drops for only a very short time, it may not be considered a fault condition, thus requiring no action. As soon as the voltage is low enough to be considered a fault, however, the voltage monitor should take action as soon as possible to prevent damage.
• Size requirements are another common challenge for voltage monitoring. High-voltage data-center power supplies have extremely limited space, requiring the smallest possible monitoring solution. But it also has to be reliable. Ensuring that the voltage monitoring solution can be trusted to respond to faults is imperative to a reliable power supply and distribution system.

Requirements for monitoring high-voltage power rails

Figure 2 shows a minimal high-voltage monitoring circuit implementation using:

• A high-voltage resistor ladder to step down the power rail for sensing comparators.
• Two comparators to signal under- and overvoltage faults.
• A voltage reference for comparators.
• Filtering components.
• An amplifier to provide a scaled-down voltage for the analog-to-digital converter (ADC) for analog monitoring and telemetry of the power rail.

Figure 2 High-voltage monitoring circuit building blocks. (Source: Texas Instruments)

Implementing this circuit with discrete components may present significant drawbacks. Individual component tolerances will add together, resulting in significant errors requiring costly, high-accuracy, low-temperature-drift components. Resistors are especially problematic, as each resistor’s uncorrelated error will sum to create a significant cumulative error in the resistor-divider. Discrete components consume significant board space, which is typically at a premium in data-center applications.

Figure 3 shows a reference layout with space requirements for high-voltage monitoring with discrete components.

Figure 3 A discrete high-voltage monitoring implementation. (Source: Texas Instruments)

An integrated solution

An integrated device for high-voltage supervision addresses these challenges by fully integrating the high-voltage resistor-divider, comparators, buffer, and additional features. The functional diagram in Figure 4 illustrates this approach, helping reduce total solution size while maintaining high performance.

By integrating the resistors, reference, and comparators, TI’s TPS371K-Q1 achieves an accuracy of 1% across the –40°C to 125°C temperature range, with a fast fault detection time of <5 µs, programmable glitch rejection and release delay time, as well as a 1% accurate high-bandwidth buffer that can directly drive 16-bit ADCs or downstream control circuits.

Figure 4 TPS371K-Q1 functional block diagram. (Source: Texas Instruments)

An integrated monitoring solution also provides significant board space savings in a compact package (Figure 5), requiring minimal external components.

Figure 5 Integrated high-voltage monitoring solution. (Source: Texas Instruments)

Application example

The implementation of a voltage monitoring system using the TPS371K-Q1 is straightforward. Figure 6 shows a basic schematic for monitoring the ±400V or 800V input to a DC/DC converter.

Figure 6 Voltage monitoring for a high-voltage DC/DC converter. (Source: Texas Instruments)

Using resistors on the ADJ OV and ADJ UV pins, designers can select under- and overvoltage thresholds to fit their system. The CTR and CTS pins allow the use of a capacitor to program a delay before assertion of a fault and a delay before deassertion once the voltage returns to normal. Open-drain outputs enable easy interface with logic levels other than the device’s own supply voltage. The VSENSE output pin provides a scaled representation of the SENSE input voltage for direct connection to an ADC. Designers can select voltage sense output factors with options ranging from 200 to 900.

Integrated monitoring solutions

The transition to high‑voltage DC architectures is reshaping design requirements for next‑generation data‑center power systems, especially as AI workloads continue to push rack‑level power far beyond the limits of today’s distribution schemes. Reliable voltage supervision becomes foundational, helping ensure high‑energy power-rail monitoring with the speed, accuracy, and reliability required to protect downstream converters and maintain system stability.

Integrated monitoring solutions such as the TPS371K-Q1 address these challenges by combining precise threshold detection, fast fault response, programmable filtering, and compact implementation into a single device optimized for the electrical and space constraints of modern data centers. By adopting advanced monitoring approaches, designers can confidently deploy ±400 V and 800 V architectures that deliver the efficiency, power density, and reliability needed to support the continued growth of AI‑driven computing at the gigawatt scale.

Henry Naguski is an applications engineer for Linear Power at Texas Instruments, working with voltage references and supervisors. He specializes in shunt voltage references and high-voltage supervisors. Henry holds a bachelor’s degree in computer engineering from Montana State University.

Masoud Beheshti leads application engineering and marketing for Linear Power at Texas Instruments. He brings extensive experience in power management, having held roles in system engineering, product line management, and marketing and applications leadership. Masoud holds a bachelor’s degree in electrical engineering from Ryerson University and an MBA with concentrations in marketing and finance from Southern Methodist University.

Related Content

• The transition from 54-V to 800-V power in AI data centers
• The shift to 800-VDC power architectures in AI factories
• TI launches power management devices for AI computing
• TI’s power devices focus on higher power density
• Power Tips

The post Power Tips #150: Overcoming high-voltage monitoring challenges in gigawatt-scale data centers appeared first on EDN.