45 Days, 18MW, 300 Cabinets: Huawei Moves SenseTime AIDC Power Supply onto the Rooftop

In just 45 days, Huawei delivered an 18MW power supply and distribution system for SenseTime's AI Data Center (AIDC) in Lingang, Shanghai — enough to support 300 IT cabinets. What used to take at least three months from approval to commissioning was completed in just one and a half months from contract signing. Instead of occupying indoor space, the entire power system was pre-integrated into standardized containers and lifted onto the rooftop.

18MW: Enough to Power 300 Cabinets

A typical commercial office building draws about 100W per square meter, so 18MW equals the full-load consumption of a 180,000-square-meter office complex. In SenseTime's Lingang facility, at 48kW per cabinet, 18MW provides complete power infrastructure for approximately 300 cabinets.

Behind these cabinets is far more than sockets and cables: UPS systems, intelligent busways, power distribution cabinets, lithium battery backup, fire protection, and environment monitoring systems. Traditional electromechanical retrofits need design drawings, factory scheduling, on-site assembly and wiring, testing, and full-system commissioning — at least three months even in the best case. Power distribution is often the bottleneck: repairs take the longest and impact operational continuity the most. SenseTime's requirements were even more demanding: the machine room was already in operation — no downtime allowed, no existing cabinets relocated, and a strict 45-day delivery deadline.

The Secret to 45 Days: Moving the Construction Site into the Factory

Huawei's answer was to minimize on-site work by moving processes into the factory. Power modules and SmartLi lithium battery systems are pre-integrated into standardized containers — UPS, lithium backup, power distribution units, air conditioning, fire protection, and enclosure systems are assembled and fully tested before shipping. Once on site, lifting, docking, and simple wiring are all that is needed. Factory prefabrication and site infrastructure thus run in parallel, instead of the traditional serial flow of drawings, construction, finishing, and commissioning. Lifting the containers onto the rooftop also freed indoor space for IT cabinets and physically isolated backup power from the main business.

Rooftop Deployment of 18MW: More Than Just Lifting

• Load bearing: the rooftop must support the full weight of the power containers;
• Heat dissipation: power modules and lithium batteries must run stably in outdoor high temperatures;
• Integration: UPS, distribution, lithium battery, and fire protection are pre-integrated and commissioned before shipping;
• Track record: Huawei has delivered more than 3,000 power module sets globally.

Lithium as Primary Backup Power: AI Monitors Every Cell

The project also marks the large-scale deployment of SmartLi lithium batteries as primary AIDC backup power. In the past, backup power was passive: when grid power failed, diesel generators started and battery banks held for a few minutes until synchronization; fault alarms relied on manual inspection. Huawei's solution turns backup power into proactive early warning — on top of five-layer safety protection (cell, module, pack, cabinet, system), AI algorithms monitor voltage, temperature, and internal resistance in real time, detecting anomalies 7 to 30 days in advance. Lithium also offers roughly 70% higher energy density than lead-acid, and intelligent BMS management reduces routine O&M costs.

Not Just Huawei: The Whole Industry Is Switching

These cases point to one clear trend: AIDC construction is shifting from "project customization" to "product delivery."

Why Now: Demand, Localization, and O&M Costs

In 2026, large model training and inference demand keeps growing exponentially. The three major telecom operators' intelligent computing capex grew more than 60% year-on-year for the first time, and leading AI companies are expanding while operating — the traditional three-to-six-month delivery cycle can no longer keep up.

High-end GPU supply constraints mean domestic GPUs and localized infrastructure must advance together. Meanwhile, traditional power O&M is labor-intensive, accounting for about 15%–20% of annual TCO. AI-driven predictive maintenance, remote monitoring, and automatic fault location are key to cutting this cost — lower power costs create room for lower per-token and per-compute costs.

The Future: A Plug-and-Play Computing Power Station

SenseTime Lingang's 45-day, 18MW delivery is just one example. In the future, customers may buy a "plug-and-play computing power station" — equipment prefabricated and tested in the factory, then lifted, connected, and commissioned on site. Equipment vendors will become more concentrated, division of labor clearer, and large model companies will no longer see launch timelines delayed for months. The AIDC construction site is becoming a factory workshop.