Managing Power Allocation in Data Centers

Data centers are rapidly becoming a major driver for DC power distribution and DC microgrids. Hyperscale facilities consume 20 to 100MW each, wit most of that power ultimately delivered at ~1V at the point-of-load for xPUs and memory. To improve efficiency and to reduce distribution cost, designers push voltage levels upward, with the conversion to 1V as close to the silicon as possible. Hereby, the power conversion system architecture is of crucial importance! Across the industry, facility-scale DC distribution is converging on either +/-400VDC (Google, Meta, Microsoft) or 800VDC (NVidia). In a future architectures, these DC buses will likely be fed from the medium voltage AC grid via solid-state transformers (SSTs). Today, the power delivery from 800V to 1V is envisioned to move from 800 V → 48 V → 12 V → 1 V. A rack-level conversion from 800V to 48V, is followed by a tray- or GPU card-level conversion from 48V to 12V. The final conversion from 12V to 1V happens on the GPU card, as close to the silicon as possible. Exact voltages may vary a bit. This structure evolved from traditional AC-fed architectures. However, it has two big drawbacks: it still requires substantial copper at rack- and tray-level and it has multiple conversion stages. Both add loss and cost. Skipping a stage—for example, jumping from 800 V directly to ~12 V—sounds attractive, but creates challenges for converter semiconductors and magnetics. A multi-module series architecture may be more promising! On the high-voltage side, modules connect in series, naturally dividing the input bus (e.g., 800 V into ~100 V segments). Each module converts directly to 12V, a much more favorable design point for both semiconductors and magnetics. These modules can be integrated directly on the GPU board, minimizing the amount of copper needed to transport power within a rack. A series architecture taps into low-voltage power devices which are more more efficient and more reliable than high voltage ones. Moreover, power converter transformer ratios are less extreme which simplifies magnetics. On the flip side, a series architecture requires a more complex communication and control. But embedded digital control, and high-speed communication are becoming inexpensive, making the control challenge solvable. Power system design is ultimately about managing the “conservation of misery”. Design challenges remain, but you can choose where the burden sits. The arrival of smart, all-digital power modules unlocks new possibilities to redistribute that burden more intelligently. #DC, #800V, #microgrids, #datacenters, #nvidia

⚡ Power: The First Language of Data Centers (Data Center Series – Part 2) If data is the new oil, then power is the pipeline that keeps it flowing. Behind every “Tier III”, “N+1”, or “99.99% uptime” claim, there is a power story that most people never see: 🔌 Utility feeds, transformers, and switchgear 🔋 Generators sized for IT load + cooling + growth 🧱 UPS topology, battery autonomy, and bypass strategy 📡 Smart PDUs and clean distribution down to rack level You can buy the best servers in the world, but if the power design is weak, everything sits on a fragile foundation. ⚠️ Common blind spots include: • Two feeds that are “dual” on paper but share the same upstream failure point • Poor phase balancing that stresses parts of the system • Underestimated cooling power, leading to overload during hot days • No real visibility of power use per rack, per room, or per customer 🧠 Non-technical leaders don’t need to read single-line diagrams—but they do need to ask the right questions: • Do we have true dual power paths all the way to the rack? • What is our real battery autonomy and generator runtime? • How do we test our backup systems under realistic load? • Can we see power consumption trends and capacity for the next 6–12 months? 💡 Good power design is not just about avoiding blackouts. It directly impacts: • Operational cost • Expansion plans • The ability to host higher density workloads • Confidence from customers and auditors 🚨 The most dangerous phrase in critical infrastructure is: “We’ve always done it this way.” In modern data centers, power is a strategic discipline, not just electrical wiring. If you work with critical IT, ask yourself: 👉 When was the last time you reviewed your power resilience end-to-end? #DataCenter #Power #UPS #Uptime #CriticalInfrastructure #ITRisk #BusinessContinuity

The electrical decisions made during facility design determine how much of the energy entering a datacenter becomes useful computation. Not the processor selection, the cooling system, or the software stack; the electrical architecture. Every stage between the generation source and the processor is a conversion event. Each conversion event has a loss, and those losses compound. By the time energy reaches the workload in a typical AI facility, a significant fraction of the fuel that entered the system has already left as heat; before a single inference or training runs. This is not a maintenance problem or an operational one. The loss profile of a facility is largely fixed at design stage. The topology of the electrical system, the number of conversion stages, the distribution voltage, the architecture of the power delivery chain; these decisions are made once, during design, and they determine the energy productivity ceiling of everything that runs inside that facility for the next two decades. An organization that treats electrical architecture as a cost-minimization exercise at design stage is not making a facilities decision, it is setting its AI productivity ceiling.

Optimizing Data Center Generator Systems: Bigger Generators and Paralleling Architectures When designing a data center, one of the most critical decisions is selecting the right generator system. Cummins’ White Paper, Diesel Generators in the Data Center—When is Bigger Better?, explores the advantages of using larger generators and paralleling architectures to optimize power systems for data centers. Key Insights: 1. Benefits of Larger Generators: - Reduced Footprint: Larger generators require less space, making them ideal for urban data centers or sites with limited real estate. Fewer, larger units can provide the same power as multiple smaller ones, reducing the overall footprint. - Lower Installation and Maintenance Costs: Larger generators reduce installation complexity and costs. Maintenance expenses are also lower since fewer units need servicing. - Increased Reliability: Data shows that systems with fewer large generators have a lower probability of failure compared to systems with many smaller generators. This is due to the reduced number of auxiliary components, which are often the source of failures. 2. Paralleling Generator Architectures: - Efficiency and Scalability: Paralleling generators allows for more efficient use of power capacity and easier scalability as data center demands grow. - Redundancy and Reliability: Paralleling systems provide redundancy, ensuring that if one generator fails, others can take over, maintaining uninterrupted power. - Concurrent Maintainability: Advanced paralleling designs, such as segmented or dual bus architectures, allow for maintenance without shutting down the entire system, ensuring continuous operation. 3. Design Optimization: - Combining larger generators with paralleling architectures can offer the best of both worlds: reduced footprint, lower costs, and enhanced reliability. For example, a 10MW data center using 3.5MW generators can save significant space and installation costs compared to using 2.0MW units. Why This Matters: Data centers are the backbone of modern digital infrastructure, and their power systems must be both reliable and efficient. By opting for larger generators and paralleling architectures, data center operators can achieve significant cost savings, improve reliability, and ensure scalability for future growth. These design choices not only optimize operational efficiency but also enhance the overall resilience of the data center. Conclusion: Cummins’ white paper provides valuable insights into the strategic advantages of using larger generators and paralleling systems in data center designs. By carefully considering these options, data center designers can create more efficient, reliable, and cost-effective power systems that meet the demands of today’s digital world. #DataCenter #PowerSystems #Generators #Cummins #EnergyEfficiency #Reliability #LinkedInPost

𝗕𝗮𝗹𝗮𝗻𝗰𝗶𝗻𝗴 𝘁𝗵𝗲 𝗚𝗿𝗶𝗱 𝗶𝗻 𝗥𝗲𝗮𝗹 𝗧𝗶𝗺𝗲 𝗧𝗮𝗸𝗲𝘀 𝗠𝗼𝗿𝗲 𝗧𝗵𝗮𝗻 𝗝𝘂𝘀𝘁 𝗟𝗼𝗮𝗱 𝗦𝗵𝗲𝗱𝗱𝗶𝗻𝗴 When power systems get tight, most people think of one thing: load shedding is turning things off. But that’s just one lever. 𝗧𝗼 𝘁𝗿𝘂𝗹𝘆 𝗯𝗮𝗹𝗮𝗻𝗰𝗲 𝗽𝗼𝘄𝗲𝗿 𝗶𝗻 𝗿𝗲𝗮𝗹 𝘁𝗶𝗺𝗲, 𝗲𝘀𝗽𝗲𝗰𝗶𝗮𝗹𝗹𝘆 𝗶𝗻 𝗮 𝘄𝗼𝗿𝗹𝗱 𝗱𝗿𝗶𝘃𝗲𝗻 𝗯𝘆 𝗔𝗜, 𝗵𝘆𝗽𝗲𝗿𝘀𝗰𝗮𝗹𝗲 𝗴𝗿𝗼𝘄𝘁𝗵, 𝗮𝗻𝗱 𝗿𝗲𝗻𝗲𝘄𝗮𝗯𝗹𝗲 𝘃𝗮𝗿𝗶𝗮𝗯𝗶𝗹𝗶𝘁𝘆, 𝘆𝗼𝘂 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗰𝗼𝗼𝗿𝗱𝗶𝗻𝗮𝘁𝗲 𝗺𝘂𝗹𝘁𝗶𝗽𝗹𝗲 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗲𝘀 𝘀𝗶𝗺𝘂𝗹𝘁𝗮𝗻𝗲𝗼𝘂𝘀𝗹𝘆: ✅ 𝗟𝗼𝗮𝗱 𝗦𝗵𝗲𝗱𝗱𝗶𝗻𝗴 The emergency break glass. Cut non-critical loads fast. ✅ 𝗟𝗼𝗮𝗱 𝗦𝗵𝗶𝗳𝘁𝗶𝗻𝗴 Move flexible demand to low-cost or high-supply windows. ✅ 𝗙𝗮𝘀𝘁 𝗦𝘁𝗮𝗿𝘁 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 Fire up assets like gas turbines or battery peakers. ✅ 𝗘𝗻𝗲𝗿𝗴𝘆 𝗦𝘁𝗼𝗿𝗮𝗴𝗲 Discharge reserves when the system is stressed. ✅ 𝗥𝗲𝗻𝗲𝘄𝗮𝗯𝗹𝗲 𝗖𝘂𝗿𝘁𝗮𝗶𝗹𝗺𝗲𝗻𝘁 Sometimes you have to dial back the sun and wind. ✅ 𝗥𝗲𝗮𝗰𝘁𝗶𝘃𝗲 𝗣𝗼𝘄𝗲𝗿 𝗮𝗻𝗱 𝗩𝗼𝗹𝘁𝗮𝗴𝗲 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 Stability isn’t just about megawatts. ✅ 𝗗𝗲𝗺𝗮𝗻𝗱 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗲 Pre-contracted users drop load on signal. ✅ 𝗜𝘀𝗹𝗮𝗻𝗱𝗶𝗻𝗴 Microgrids and self-generation facilities relieve the bulk system. We’re entering a world where balancing the system in real time isn’t optional. It’s essential. Those who understand how to orchestrate these tools will be the ones who keep operations stable, costs low, and sustainability goals within reach. What are you doing to prepare for this level of energy intelligence? #GridStability #DemandResponse #EnergyManagement #RealTimeEnergy #DataCenters

Why PUE Should Keep Every Data Center Engineer Awake at Night? . . Every Data Center Facility Engineer knows that keeping systems running isn’t enough, the real challenge is how efficiently we do it. That’s where PUE (Power Usage Effectiveness) comes in. Defined by The Green Grid, PUE = Total Facility Energy ÷ IT Equipment Energy. In simple terms, it tells us how much of our power actually runs servers, and how much gets lost to cooling, lighting, or inefficiencies. (Source: The Green Grid, 2024) Why PUE Matters More Than Ever? With data centers now consuming about 2% of global electricity and rising fast (IEA, 2023), every watt counts. A low PUE doesn’t just cut bills, it defines your data center’s sustainability, resilience, and performance. According to Uptime Institute (2024), the average global PUE is 1.58, but the best hyperscale data centers achieve below 1.2. That’s a huge competitive advantage in both cost and environmental impact. (Source: Uptime Institute Global Data Center Survey 2024) How to Improve Your PUE? Improving PUE is not about one big upgrade, it’s a series of smart moves that add up: Optimize cooling systems using hot/cold aisle containment and precision airflow. Leverage free cooling or liquid cooling where climate allows. Replace outdated UPS systems and CRAC units with energy-efficient models. Monitor power distribution closely through DCIM systems to identify hidden losses. Even small changes , like raising server inlet temperatures by just 1°C can improve efficiency by 2–4% (ASHRAE, 2023). (Source: ASHRAE Thermal Guidelines, 2023) The Bigger Picture Lowering PUE isn’t just a technical goal it’s a statement of engineering excellence and environmental responsibility. In an era where uptime, sustainability, and cost-efficiency define success, mastering your PUE means mastering your facility’s future. (Source: IEA Data Center Energy Outlook, 2024) 💭 Question for You: What’s the average PUE in your data center and what’s the one strategy that helped you improve it the most?

𝗛𝗼𝘄 𝘁𝗼 𝗔𝗽𝗽𝗹𝘆 𝗤𝘂𝗮𝗻𝘁𝘂𝗺-𝗜𝗻𝘀𝗽𝗶𝗿𝗲𝗱 𝗔𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺𝘀 𝘁𝗼 𝗗𝗮𝘁𝗮 𝗖𝗲𝗻𝘁𝗲𝗿 𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻 (𝗔𝗜𝗢𝗽𝘀 𝗪𝗶𝘁𝗵𝗼𝘂𝘁 𝗮 𝗤𝘂𝗮𝗻𝘁𝘂𝗺 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿) Most leaders hear “quantum” and think of it as experimental, expensive, and years away. That’s a mistake. Quantum-inspired algorithms run on classical infrastructure today and solve the hardest problem you actually have: large-scale optimization under constraints. If you run data centers, this is immediately actionable. What they actually do They convert your environment into an energy minimization problem. Instead of brute forcing every possibility, they rapidly converge on high-quality solutions across massive decision spaces. Think: • Placement • Scheduling • Routing • Thermal balancing • Power allocation Where to apply first (high ROI use cases) 1. Rack and cluster placement Model racks, power domains, cooling zones, and network topology as constraints. Objective: minimize latency + cable length + thermal hotspots. 2. GPU scheduling and utilization: Encode job priority, SLA windows, GPU affinity, and network contention. Objective: maximize utilization while reducing idle burn and queue latency. 3. Thermal + power balancing: Integrate cooling capacity, airflow constraints, and power density. Objective: flatten hotspots without over-provisioning. 4. Network traffic shaping Model east-west traffic flows and oversubscription ratios. Objective: Reduce congestion and packet loss under peak load. How to implement (practical workflow) Step 1: Define variables • Binary: placement decisions, routing paths • Continuous: load, temperature, power draw Step 2: Define constraints • Power caps per rack and row • Cooling limits by zone • Network bandwidth ceilings • SLA requirements Step 3: Build the objective function. Combine into a weighted cost function: • Latency • Energy consumption • Thermal deviation • Resource fragmentation Step 4: Select a solver. Use simulated annealing or related heuristics to explore the solution space efficiently. Step 5: Iterate with real telemetry. Feed in live data: • DCIM • BMS • Scheduler metrics: Continuously refine the model. What “good” looks like • 10–25% improvement in GPU utilization • Lower east-west congestion without network upgrades • Reduced thermal excursions • Faster schedule generation cycles Where most teams fail • Overfitting the model before validating its impact • Ignoring real-time telemetry • Treating this as a one-time optimization instead of a continuous system Bottom line: You don’t need quantum hardware to get quantum-level thinking. You need a structured optimization model and the discipline to iterate it against real operating data. If you’re running >10MW environments and not doing this, you’re leaving efficiency and margin on the table. #DataCenters #AIInfrastructure #GPU #Optimization #HighPerformanceComputing #Cloud #Infrastructure #DigitalTransformation

The next wave of data center innovation isn't about choosing between efficiency and sustainability. It's about achieving both through intelligent automation. Three key trends are reshaping how data centers operate in 2025: Smart Resource Management Advanced #AI systems now handle complex resource allocation automatically, reducing energy consumption by up to 40% while improving performance. The technology continuously analyzes workload patterns and adjusts server utilization in real-time, ensuring optimal efficiency without human intervention. Predictive Maintenance Evolution AI-driven systems detect potential issues days or weeks before they occur, nearly eliminating unexpected downtime. This capability has reduced maintenance costs by 35% for early adopters while extending hardware lifespan significantly. Sustainable Operations Data centers are becoming increasingly self-sufficient through renewable energy integration. Leading facilities now combine AI-controlled cooling systems with on-site solar and wind power, cutting both costs and carbon emissions. Emerging markets are at the forefront of this transformation, with facilities in #India and #Brazil showing how local resources can be leveraged effectively. The Results: - 50% reduction in operational costs - 90% decrease in system downtime - 60% smaller carbon footprint - 75% less human intervention required for routine tasks The shift toward autonomous, sustainable operations isn't just an environmental choice - it's a competitive necessity. Companies that embrace this transformation are seeing substantial improvements in both operational efficiency and bottom-line results. #datacenters

Google Embraces Flexible Loads, Demand Response in New Utility Deals In a meaningful step forward on grid flexibility, Google has signed deals with two utilities to adjust a data center’s electricity demand by shifting the timing and volume of AI workloads. “These capabilities, often referred to as demand response, have several advantages, especially as we continue to see electricity growth in the US and elsewhere,” said Michael Terrell, Head of Advanced Energy at Google. “It allows large electricity loads like data centers to be interconnected more quickly, helps reduce the need to build new transmission and power plants, and helps grid operators more effectively and efficiently manage power grids.” Google has signed new utility agreements with Indiana Michigan Power (I&M) and Tennessee Valley Authority (TVA) as its first step on delivering data center demand response by managing machine learning (ML) workloads. Recent research suggests that up to 100 gigawatts of additional headroom could be created on US grids if data centers can be flexible, and limit their power demands for a few hours each year. A key tradeoff is that adopting flexible workloads could allow faster access to power for data centers, a key opportunity in a capacity-constrained landscape. Google says it sees “a significant opportunity” for demand response as AI boosts demand for AI infrastructure. “By including load flexibility in our overall energy plan, we can manage AI-driven growth even where power generation and transmission are constrained,” Terrell writes. Here’s the blog post: https://lnkd.in/eY2B994V

♦What is PUE / DCiE? Power Usage Effectiveness (PUE) and its reciprocal Data Center infrastructure Efficiency (DCiE) are widely accepted benchmarking standards proposed by the Green Grid to help IT Professionals determine how energy efficient data centers are, and to monitor the impact of their efficiency efforts. ♦How to Determine PUE ? 1- Take a measurement of energy use at or near the facility’s utility meter. If the data center is in a mixed-use facility or office building, take a measurement only at the meter that is powering the data center. If it is not on a separate utility meter, estimate the amount of power being consumed by the non-data center portion of the building and remove it from the equation. 2-Measure the IT equipment loads after power conversion, switching, and conditioning is completed. According to The Green Grid, the most useful measurement point is at the output of the computer room power distribution units (PDUs). This measurement should represent the total power delivered to the server racks in the data center. ♦PUE = Total Facility Power / IT Equipment Power ♦PUE Example: Having a facility that uses 100,000 kW of total power of which 82,000 kW is used to power your IT equipment, would generate a PUE of 1.2. The 100,000 kW of total facility power divided by the 82,000 kW of IT power. ♦How to Determine DCiE ? DCIE is the reciprocal of Power Usage Efficiency (PUE). PUE is defined as the total facility power divided by the IT equipment power. That means that ♦DCiE = 1/PUE *100. ♦An example that will help you understand how to work out your data center energy efficiency: Total Facility Power = 100 kW IT Equipment Power = 82 kW DCIE = 82/100 x 100% = 82% PUE DCiE Level of Efficiency 3.0 33% Very Inefficient 2.5 40% Inefficient 2.0 50% Average 1.5 67% Efficient 1.2 83% Very Efficient ♦How to reduce PUE? ♦Cold Aisle Containment - Cold aisle containment counts as the largest contributor to the PUE improvement in combination with by-pass air flow avoidance (blanking plates, by-pass air etc.) ♦Enhanced cooling technology - Much of a data center’s energy is spent on cooling IT equipment. Whether it’s through enhanced airflow management, advanced cooling systems, or better layout, improving the cooling system can save a great deal of energy. ♦Make small improvements - Modest improvements add up. Using advanced power supplies, automatic lighting, and removing waste ensures that the whole facility contributes to a lower PUE. ♦Measure regularly - A Data center should measure its PUE regularly. Not only does this show when there is an issue, but it also provides a record of efforts and successes. ♦Why it’s important to reduce PUE? PUE and DCiE demonstrate how efficiently a data center uses energy. By understanding the amount of energy spent on different processes, companies can assess how to save money, improve service, and reduce waste. #DCDC Knowledge.