Liu Xuechao: Innovation of Diesel Generator Sets for Data Centers: Localization, Digitalization and Multi-energy Coordination

The Macro Background of AI-era Computing Revolution and Diesel Generator Innovation

The past year marked a historic turning point for the data center industry. Driven by the explosive growth of generative artificial intelligence, computing power demand centered on large model training and inference has risen exponentially. Today’s AI is far more than a suite of product applications. It delivers tremendous efficiency gains and cross-industry integration capabilities. For instance, recent product launches from Google have completely reshaped user experience, proving highly engaging in a way comparable to gaming addiction.

Supply Crisis amid the AI Industrial Revolution

Gartner forecasts that AI-optimized servers will account for 44% of global data center power consumption by 2030. This surge in demand has directly pushed total data center deployments to an all-time high. Yet beneath this boom lie severe supply chain risks. Between 2024 and 2025, the global diesel generator market suffered acute shortages. Lead times for high-power units from major international brands once stretched to 50 weeks or longer. For internet giants and operators racing to capture AI computing windows, such delays were unacceptable.

Supply tightness acted as a catalyst for the rapid development of domestic diesel generators. Forced by product shortages, local manufacturers gained valuable opportunities for field trials and qualification, quickly filling market gaps. Domestic units now stand as a core option for most end users.

Three Bottlenecks of the Traditional Diesel Generator Model

While supply shortages have eased, deep structural challenges persist for diesel generator deployment in data centers. Our field experience shows that traditional diesel generator architectures struggle to cope with the high-density, dynamic loads of the AI era, burdened by legacy design flaws and unreasonable specifications.

Mismatched Application: Disconnect between Standards and Real-world Operations

Current codes and test requirements largely inherit decades-old criteria designed for mechanical loads such as fire pumps and emergency lighting, and have not been optimized for IT load characteristics of modern data centers. A prominent example is the widespread industry misunderstanding surrounding the 0–100% step load test for data centers.

The Misinterpretation of NFPA 110: NFPA 110 (Standard for Emergency and Standby Power Systems) requires standby power systems to restore power and accept loads within 10 seconds. The industry commonly interprets this requirement as mandatory capability for direct 0–100% step load acceptance. In data centers, however, all IT loads are powered by UPS (Uninterruptible Power Supply). When utility power fails, UPS achieves seamless transfer, and the rectifier ramps up load linearly via soft start.

First-principles Analysis: From a physical perspective, diesel generators in data center environments almost never encounter a hard 0–100% step load. (The sole exception would be low-voltage generators with UPS switched to bypass for load feeding, which represents a stacked extreme fault. Even in this edge case, although the generator can remain online, voltage and frequency deviations far exceed G3 performance thresholds, rendering the power system unstable.) To meet this practically irrelevant benchmark for data centers, manufacturers often compromise steady-state performance by deploying highly aggressive torque response tuning. This leads to misallocation of equipment performance resources. The excitation and governor systems of diesel generators both adopt PID control. Tuning for extreme transient step load response inevitably yields softer parameters and degraded performance under normal operating conditions. Many manufacturers market this capability as a selling point. This is not purely a technical issue; it reflects a failure to evaluate equipment deployment and calibration from the end user’s perspective. One example: belt-driven battery chargers fitted on diesel generators cannot fully recharge batteries during the 5–10 minute monthly test runs typical for data center standby units.

Unstable Performance: Hidden Risks under Special Operating Conditions

Load profiles of AI data centers differ drastically from conventional industrial loads, characterized by high power density, heavy harmonics and potential capacitive features. A large share of data center outages triggered by diesel generator load failures stem from power quality (PQ) issues. Mature solutions exist for this problem in other power generation sectors, yet little progress has been made within the data center industry for years.

PQ Characteristics and Capacitive Loads: Generators are rated at 0.8 power factor, while data center power factor approaches unity. Under light load or specific operating modes, data center loads often become capacitive (leading power factor). When feeding capacitive loads, the generator excitation system must reduce excitation current to maintain voltage. Excessive capacitive reactive power may drive excitation current down to zero or even reverse it, pushing the generator into leading power factor operation. This can easily trigger loss-of-excitation protection or oscillation-induced shutdown. At low load levels, low diesel fuel injection volume weakens governor response, and active/reactive power sharing among paralleled units exhibits large deviations. This explains why ISO 8528 and GB 2820 specify active and reactive power sharing requirements only for load rates above 20%. Under heavy load, generator regulation becomes stiff with poor disturbance rejection. Furthermore, paralleled generator systems have far weaker transient performance than standalone units. Combined with complex operating profiles, operating points can readily fall outside the generator PQ capability curve.

Low Load Operation and Wet Stacking: Due to N+1 redundancy design and phased data center IT load ramp-up, diesel generators frequently operate under light loads (<30%). At low load, combustion temperatures inside large-bore diesel cylinders remain low. Injected fuel cannot fully atomize or combust. Unburnt fuel and lubricating oil condense in the exhaust system, a phenomenon known as wet stacking. This produces black tar-like liquid dripping, and forms a glazed layer on cylinder walls. Piston ring sealing is permanently degraded, leading to power derating, sharply rising lube oil consumption, and even exhaust manifold fires. Over time, tar buildup accumulates and may flow back into valve guides and turbochargers.

Inflated High-altitude Power Ratings: Major computing hubs are being built in western China including Guizhou, Gansu and Inner Mongolia. Nevertheless, some manufacturers publish inflated high-altitude power derating figures. Multiple conflicting derating curves may be provided for identical engine configurations, mostly derived from theoretical calculations rather than physical testing. In thin high-altitude air, smaller-displacement engines suffer insufficient intake and sharp rises in thermal load. Units often fail to deliver nameplate power, posing severe safety risks to data centers.

Lagging Development: Marketing Prioritized over R&D

Compared with rapid advances in liquid cooling, AI chip optimization and high-voltage DC technology, technical iteration in diesel generators has been extremely slow. The market is saturated with homogeneous competition, with greater emphasis placed on marketing than fundamental R&D. Many recurring issues are not bounded by high technical barriers. Instead, they arise from disconnection from real-world operating scenarios and disregard for long-standing pain points. More critically, as high-capital core assets, diesel generators sit idle for 99% of their service life, causing massive capital lockup and wasted potential value.

Large-scale Deployment and Breakthroughs of Domestic Diesel Generators

Supply Chain Restructuring and Market Penetration

We commissioned and brought online a large fleet of domestic diesel generators in 2025. This marks not merely a procurement volume milestone, but an inflection point in market confidence. Our qualified vendor list includes leading OEMs such as Yuchai, Weichai and Shangchai. Through market competition, domestic diesel generators support fast project delivery while cutting capital expenditure (CAPEX) and operational expenditure (OPEX). The market share of high-power domestic diesel engines in the data center sector is climbing rapidly.

Rigorous Testing: Building Trust

We recognize that current localization progress represents supply chain transformation rather than fundamental technological disruption. To offset gaps in brand legacy and operational track records and ease end-user concerns about reliability, we adopt stringent test protocols exceeding ISO 8528, referencing nuclear power plant and marine classification society standards.

• 100 consecutive no-fail start cycles For core domestic models including Shangchai (SDEC), we apply nuclear safety-grade power supply test criteria. Test requirement: The unit must complete 100 consecutive starts under hot or cold conditions. Pass criterion: Zero failures. Any failed start, voltage build-up timeout or false protection trip results in test failure. Significance: This test heavily stresses the durability of starting motors, starter relays and engine thermal stability. Passing validates that domestic units reliably achieve the most fundamental and critical requirement: guaranteed starting capability.
• Field testing under high temperature and high altitude To resolve inflated high-altitude rating issues, we reject purely theoretical manufacturer derating curves. Instead, we partner with diesel OEMs and authoritative domestic testing institutes for field extreme-condition performance validation. Test practice: At 1,500 m elevation under 40°C summer ambient temperature, standard-specification units undergo full-load and overload testing. Engine thermal load data serves as the core evaluation metric to confirm operation within design envelopes. Physical testing ensures nameplate power ratings are authentic and fit for harsh operating environments at western China computing hubs.
• Disassembly inspection and marine classification society standards Beyond long-duration runtime tests, we adopt acceptance philosophies from marine classification societies such as CCS and DNV. After completing all durability testing, core components including connecting rod bearings, piston rings and cylinder liners are disassembled for inspection. Test practice: Evaluate micro-wear patterns, carbon deposition distribution and physicochemical changes in lubricating oil. This approach identifies latent design or material risks earlier than raw performance data alone.
• 110% overload and continuous transient load cycling Additional stringent protocols include extended-duration overload testing and repeated transient load tests:

1. Long-duration operation at 110% overload: ISO standards only mandate 1 hour of overload within a 12-hour window. To verify engine margin, we implement prolonged continuous overload testing. In essence, the high-power unit is operated above rated capacity to stress-test thermal management headroom and ensure sufficient safety margin for extreme scenarios.
2. Repeated 0–100% transient load tests: Although we consider this test of limited relevance to real data center operation, it is deployed at the component strength validation stage to apply mechanical stress to crankshafts and couplings and verify structural integrity.

Collaboration and Customization for Domestic Diesel Generators

Domestic diesel generators have resolved supply constraints, yet the development journey continues. Whether the industry landscape undergoes permanent reshaping hinges on strengthened practical collaboration and resolution of fundamental technical pain points. Customization, modularization and digitalization will better align generator products with computing center requirements. As individual data center campuses scale toward gigawatt levels, diesel generators are deployed in large clusters. A single large campus may host hundreds of units. Traditional manual inspection and reactive response modes can no longer meet the ultra-high reliability requirements of AI power infrastructure.

Urgency of Digital Transformation

In clustered deployments, manual logging of oil pressure, coolant temperature, voltage and other parameters is inefficient and incapable of identifying trending incipient faults hidden within raw readings. For example, a subtle rising trend in coolant temperature may go unnoticed on local gauges, yet big data analytics can pre-alert thermostat faults. The digital diesel generator system follows a data-driven philosophy, converting dumb equipment into intelligent terminals and resolving the black-box limitations of conventional maintenance.

Application of Digital Systems

The digital diesel generator platform we are developing centers on data-driven operations to resiliency and intelligence of standby power infrastructure. It transforms isolated, silent generator assets into transparent, intelligent, predictive power assurance units.

For instance, regarding the generator PQ characteristic challenges outlined earlier, digital operational performance assessment visualizes unit operating status. When operating points approach dangerous boundaries of the PQ curve, the system triggers immediate alerts and provides guidance for unit start/stop adjustment. Another example: high-frequency failures of starting batteries are monitored continuously via equipment health scoring and start condition assessment modules, with predictive alarms. Fault grading and scenario-based alerts shift maintenance from reactive troubleshooting to proactive prevention, fundamentally improving system resilience.

Digitalization improves reliability while easing cost pressure. Automated inspection reports replace manual logbooks, and condition-based component replacement cuts consumable waste. At fault onset, the system automatically pushes diagnostic trouble codes (DTCs) and troubleshooting guidance, shortening MTTR (Mean Time To Repair). Energy management reduces auxiliary power draw in standby mode and fuel consumption during generator load runs.

Three digital product tiers have been defined for diesel generators. Beyond solving immediate maintenance pain points, they establish data and control foundations for the next evolution of generators, enabling deeper integration into broader energy systems.

Multi-energy Coordination: Redefining the Role of Diesel Generators

Looking ahead, we must re-evaluate the position of diesel generators in energy architectures. They should no longer remain dormant assets, but act as dispatchable flexible resources within new power systems.

Redefinition of Generator Roles

Historically, diesel generators typically log fewer than 5 operating hours per year, mostly for periodic no-load or loaded testing. Under multi-energy coordination frameworks, we are advancing participation of data center diesel generators in virtual power plant operations.

Use case: During summer peak demand periods or power shortages, data centers start diesel generators, disconnect from utility supply and operate off-grid to deliver peak shaving power. Through this demand response, data centers can earn substantial ancillary service subsidies.

Performance outlook: Annual operating hours for diesel generators can increase dramatically. Peak-valley arbitrage can offset fuel costs and convert generators from cost centers into profit centers.

Diesel generators are integrated into source-grid-load-storage (SGLS) integrated microgrids. Applicable business scenarios include:

• Supplement and balancing for direct green power supply
• Peak shaving resources for source-grid-load-storage integrated projects
• Aggregation assets for Virtual Power Plants (VPPs)
• Trading entities in electricity spot markets and ancillary service markets

From Dormant Assets to Value-generating Assets

This role transformation unlocks massive economic value. Participation in power markets may raise annual operating hours of diesel generators from the current sub-5 hours to over 1,000 hours.

This paradigm activates heavy assets that previously lay dormant. In particular, hybrid energy systems combining storage, green power and diesel generation strengthen the position of generators as critical backstop nodes within source-grid-load-storage integrated projects. While safeguarding on-site energy security for data centers, they also support grid stability. This model aligns with data center energy transition, addressing low green power utilization, duck curve challenges and power supply complexity for ultra-large campuses.

The evolving role of diesel generators greatly improves asset return on investment and opens new value anchors amid intense market competition.

Future Outlook and Closing Remarks

To conclude this presentation, we look ahead to the future landscape, technical directions and evolving core value of the data center diesel generator market.

Supply conditions have improved, yet competition and technological disruption coexist. Looking back, diesel supply shortages were a major industry pain point causing project delays and cost inflation. However, the market landscape has undergone profound changes. 2026 will see fierce competition in the data center diesel generator sector.

Supply-side expansion: Domestic manufacturers including Yuchai, Weichai and Shangchai have fully ramped up capacity, with emerging local vendors targeting the core data center market. Foreign brand rebound: International OEMs have restored full supply capacity through capacity expansion and supply chain optimization.

At the same time, fundamental technical challenges must be acknowledged. Constrained by power density, overseas projects for ultra-large computing hubs have begun exploring alternatives to diesel generators, adopting cleaner, more sustainable power generation and storage solutions with lower operating costs. In China, diesel generators will remain the mainstream standby power solution for a considerable period due to grid characteristics and cost structures. Nevertheless, the sword of Damocles of disruptive technology hangs over the sector.

Return to Fundamentals, Embrace the Future

Moving forward, we call for orderly and healthy market development. Industry practitioners should refocus on product and technical fundamentals, delivering better products to meet demands of high-density AI computing power and jointly guarantee reliable data center operation.

Tomorrow’s diesel generator sets will carry a dual identity never seen before: they serve as the last line of defense safeguarding computing power, and also operate as intelligent management nodes embedded within complex energy ecosystems.

No matter how their roles evolve, one core value remains unchanged. I will close with this statement: Reliability admits no compromise.