Alternative power solutions for hyperscale projects

The Data Center Power Model Is Changing — Permanently For decades, hyperscale data centers were built around a simple assumption: The grid would supply primary power. Backup generators would only run during outages. UPS systems would bridge the gap. That model is changing fast. In 2026, operators are increasingly deploying behind-the-meter generation, large-scale battery energy storage systems (BESS), and hybrid microgrids as part of primary infrastructure strategy — not just backup resilience. Why? Because utility interconnection timelines in many major markets now stretch 4–7+ years, while AI demand is forcing deployments on much faster schedules. The result: Build the power plant with the data center. We are seeing a new architecture emerge: • Utility power as one source • On-site natural gas generation for dispatchable baseload • BESS for ramp response, peak shaving, and ride-through • Optional solar / renewable integration • Island-mode capability during grid instability • Intelligent controls managing dispatch in real time This is not “off-grid.” This is grid-optional, island-capable private utility infrastructure built for compute. And this trend is not limited to greenfield hyperscale campuses. It is also driving: • Legacy retrofit modernization • Brownfield industrial conversions • Repowering aging electrical infrastructure • New commissioning models that now include utility-scale integrated systems testing Commissioning is changing too. Testing now extends beyond switchgear, UPS, and generators into: • Microgrid synchronization • Black start sequencing • Battery dispatch logic • Load shedding automation • Relay coordination • Thermal runaway mitigation • Cybersecure controls integration The future data center is no longer just a building. It is becoming a self-managed power ecosystem that happens to process data. That is a fundamental shift in infrastructure delivery. #DataCenters #AIInfrastructure #Commissioning #Microgrids #EnergyStorage #NaturalGas #BatteryStorage #GridInfrastructure #Hyperscale #MissionCritical #Construction #DigitalInfrastructure #PowerGeneration #MEP #DataCenterDesign

Battery Energy Storage Systems (BESS): More Than Just "Big Batteries" The exploded-view hierarchy below highlights something often overlooked in discussions about grid-scale energy storage: A modern BESS is not simply a collection of battery cells—it is a highly integrated electromechanical, thermal, power-electronics, and software platform. At the plant level, the Power Conversion System (PCS) serves as the heart of the installation, converting power between the grid and battery system. Modern utility-scale deployments increasingly utilize 1500V DC architectures, medium-voltage PCS designs, and grid-forming inverter capabilities to improve efficiency, support black-start operation, and enhance grid stability. Inside the container, energy density continues to climb. While 2–6 MWh containers have become common, the industry is rapidly moving toward liquid-cooled 5–7+ MWh platforms. Advanced thermal management enables tighter battery packing, improved temperature uniformity, and higher continuous power capability. At the rack and module level, manufacturers are simplifying architectures through cell-to-pack designs, advanced compression systems, and integrated thermal propagation barriers that improve both safety and cost efficiency. At the cell level, LFP remains the dominant chemistry for stationary storage due to: - Long cycle life (6,000–8,000+ cycles) - Superior thermal stability - Reduced cobalt and nickel dependence - Lower total cost of ownership Emerging technologies such as LMFP and sodium-ion batteries are also beginning to appear in pilot deployments, particularly where cost and supply-chain resilience are priorities. Several industry trends are accelerating adoption: • Grid-forming inverters • DC-coupled solar + storage architectures • AI-driven energy management systems • Long-duration storage (4–12+ hours) • Second-life and recycling integration • Factory-built plug-and-play deployments For AI data centers, BESS is evolving beyond backup power. Hyperscalers increasingly use energy storage for demand response, renewable firming, peak shaving, and behind-the-meter energy optimization. As global storage deployments continue growing at more than 40% annually in many markets, the industry's key differentiators are no longer just battery chemistry, they are system integration, software intelligence, thermal management, safety performance, and long-term bankability. The future of energy storage belongs to the companies that can seamlessly integrate power electronics, batteries, thermal systems, controls, and software into a single scalable platform. ✅ Educational purpose only #BESS #EnergyStorage #BatteryTechnology #GridModernization #PowerSystems #LFP #EnergyTransition #RenewableEnergy #AIInfrastructure #DataCenters #ElectricalEngineering #BatteryStorage #GridScaleStorage #UtilityScaleEnergyStorage

Bring Your Own Capacity (BYOC) for data centers may finally be getting the attention it deserves. It’s encouraging to see more voices highlighting the role distributed energy resources (DERs) can play in meeting the enormous power needs of the AI era. Ari Matusiak at Rewiring America has been working on it. Jigar Shah and Stephen Lacey just dropped an episode about it on the Open Circuit podcast. Others have also been talking about it. The idea is simple but powerful: Hyperscalers can “bring” capacity by helping deploy DERs in the communities where they build. Imagine a data-center developer offering households solar, storage, or heat pumps as part of a community benefit plan. Those distributed resources reduce local load and provide flexible capacity — which can count toward the data center’s power and flexibility needs. That creates a rare triple win: • Governors and communities benefit because real people see lower energy bills. • Hyperscalers benefit because it reduces NIMBY friction while creating a flexible capacity cushion. • Utilities and regulators benefit because DERs help address the affordability challenge on the grid. In other words: AI infrastructure helping finance the electrification of American homes. I’m eager to see the first major data-center announcement that actually includes BYOC through DER deployment. When that happens, it could become a blueprint for how we build the next generation of digital infrastructure — with communities, not just around them.

The Hidden Megawatt: Why We Are Engineering Our Own Gridlock ⚡🏗️ Everyone is racing to secure the next grid connection 🔌 Almost nobody is asking how much compute is being left on the table with the connection they already have. I call it the Hidden Megawatt. In a traditional data center power chain, electricity is converted about five times between the grid and the GPU. Each stage adds loss, heat, cost, and failure risk. By the time power reaches the chip, roughly 5 to 7 percent of total facility capacity is already gone, burned as heat before a single token is produced ♨️ At 100 MW, that equals an entire row of GPU racks you paid for but never use 🖥️ At gigawatt scale, it becomes a full building of stranded compute capacity 🏢 This would matter less if new grid capacity were easy to obtain. It is not. In major hubs, large connections can take many years to secure ⏳ While the industry waits for new megawatts, existing megawatts quietly disappear inside legacy conversion chains. This is an architecture problem 🧠 An 800 VDC architecture cuts the conversion chain down to a minimal path ⚡ Converting medium voltage AC directly to high voltage DC and distributing DC natively turns more incoming watts into usable compute instead of heat. The ecosystem is already shifting 🚀 Next generation AI racks, high density power shelves, and 800 VDC reference designs are entering deployment now. The most valuable megawatt is often not the next one you are trying to connect. It is the one you can recover inside your existing facility 💡 No permits 📄 No queue 🚦 No multi year wait ⏱️ Just better power architecture ⚡ #DataCenterDesign #800VDC #AIInfrastructure #EnergyEfficiency #DataCenterDC #DataCenterSST #SolarEdgeSST #DataCenter800VDC

Google just paid $4.75 billion for Intersect—a company that co-locates data centers with dedicated power generation. This isn't just an acquisition. It's a signal: hyperscalers are done waiting for utilities. They're becoming power companies. The grid can't keep up. Interconnection queues stretch 5+ years. Utilities are quoting 10-12 year wait times just to study requests. Google's solution: buy the capability to generate power in lockstep with new data center load. -The Shift Nobody's Talking About- Everyone's focused on 100+ MW and GW-scale facilities. But listen to Jensen Huang's GTC speech. Edge compute and inference are growing exponentially. The same power-first logic applies at smaller scale—10 to 25 MW facilities closer to population centers. The constraint is identical. Power access determines what gets built and when. -The Opportunity for Everyone Else- Not everyone can spend $4.75 billion on a power company. Not everyone can launch satellites or build nuclear plants. But there's a terrestrial playbook emerging for those who think about power differently. The thesis: find stranded or underutilized power and repurpose it for compute. • Cold storage facilities not using their full power allocation • Industrial companies or REITs with excess capacity they're not monetizing • Existing infrastructure with power lockups that could support edge deployment In the past month, I've had conversations with operators approaching this from multiple angles—partnering with existing infrastructure owners, deploying BTM gas or modular solutions, tapping into assets that have power but aren't fully utilizing it. The common thread: they're not waiting for new grid capacity. They're finding power that already exists. -The Roadmap- Power-first development is becoming the standard, not the exception. Secure power before you break ground. Lock up access before competitors realize it's available. The operators who figure out how to access stranded power—whether through partnerships, alternative technologies, or creative deal structures—will have a significant edge as compute demand scales. Google just showed what the biggest players are willing to pay for power certainty. The question for everyone else: where's the power that nobody's using yet? Who else is seeing creative approaches to power access?

The news out of New Mexico today regarding Project Jupiter is worth a second look. Oracle and BorderPlex just announced they are ditching planned gas turbines for up to 2.45 GW of Bloom Energy fuel cell microgrids. What’s driving a move of this scale? - The Backlog Reality: We know the hardware queue for turbines is stretching into 2030. Is this "pivot" less about a preference for fuel cells and more about the reality of equipment backlog? - Water Stewardship: In a water-stressed region, Oracle is betting on non-evaporative cooling and combustion-free generation to bypass the traditional "Water vs. Data" friction. They’ve even committed $50M to local water infrastructure to anchor the deal. - The Grid Exit: By building a self-contained, DC-native microgrid, they are skipping the multi-year interconnection queue entirely. Is this the new blueprint for Hyperscale? In a world shifting from fuel-risk to delivery-risk, massive pivots like this are becoming the only way to protect a necessary ROI. Image: Project Jupiter's original concept rendering.

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 systems face a significant challenge in maintaining grid stability while increasing the use of variable renewable energy (VRE) due to the inconsistent electricity generation from sources like wind and solar. This variability complicates the balance between electricity supply and demand. A promising solution, as highlighted by the NETL techno-economic analysis (TEA), involves integrating electrolysis-based hydrogen (H₂) production, compressed air energy storage (CAES), and hydrogen-fired combustion turbines. This approach addresses the intermittency of renewable energy sources effectively. By combining large-scale energy storage with flexible hydrogen-based power generation, this integrated system enhances grid reliability, supports higher renewable energy penetration, and contributes to a more resilient and sustainable electricity network. **Simplified Process:** The proposed hybrid system operates in two modes: charging and discharging. - During low electricity demand or when excess renewable energy is available, surplus electricity powers a PEM electrolyzer to produce hydrogen from water. - The hydrogen is compressed and stored for later use, while air is simultaneously compressed and stored underground in a salt cavern using a CAES system. - During high electricity demand, the stored hydrogen is released and expanded to generate electricity before being used as fuel in a hydrogen-fueled combustion turbine (CTG) for additional power generation. - The stored compressed air is released from the salt cavern, preheated using the turbine's exhaust heat, and expanded through air expanders to produce more electricity. The preheated compressed air is also supplied to the combustion turbine, enhancing its efficiency through heat recovery. This post is for educational purposes only. Reference: NETL Report https://lnkd.in/gFTFGJXv

I recently presented to Columbia Business School alumni on AI Growth, Power Demand, and Implications for Climate Change (Goldman Sachs forecasts US data center demand for power will rise 175% by 2030). Key takeaways: There are 7 leading solutions for AI data center demand for power (other solutions exist eg. repurposing bitcoin mining, extending or reopening coal or nuclear facilities, but have limited potential). o  Wind – low-cost , but a hostile administration is blocking permits, siting is often far from data centers requiring costly transmission lines, and intermittency. o  Solar – low-cost and the current default solution (solar = 75% of capacity additions in Q1-3 2025, >7x natural gas). Low-cost battery energy storage systems (BESS) are solving intermittency. However, solar suffers from permitting + interconnection delays, and siting often far from data centers. o  Nuclear – large-scale nuclear facilities provide reliable base load power but new build is very costly and takes decades to permit and build. Small modular reactors (SMR) nuclear has potential to be cost-competitive, co-locates with data centers, and generates 24/7 firm power. However, the first commercial-scale SMR facilities are likely to be operational >2030. Nuclear fusion faces even greater technological hurdles, commercial operations early to mid-2030’s. o  Geothermal – traditional geothermal is very limited but enhanced/advanced geothermal using established drilling and fracking technologies has potential in many states, and generates 24/7 firm power. The first commercial project will be online in 2026 (Fervo Energy), but uncertainty about generation costs. o  Combined-Cycle Natural Gas – currently generates 1/3 of US power. Advantages include abundant natural gas and low-cost, siting near data centers, and 24/7 firm power. Challenges include a 3-7 year wait for turbines, permitting for new gas pipelines, and CO2 emissions. o  Fuel Cells – converting natural gas to electricity using a fuel cell instead of combustion eg Bloom Energy. Fuel cells can be quickly sited next to data centers and provide 24/7 firm power. However, fuel cells are costly (2-3x the cost of power from wind, solar, and combined-cycle natural gas facilities), and emit CO2. o  Demand curtailment – research from Nicholas Institute for Energy, Environment & Sustainability demonstrates that curtailment of just 0.25% (ie. 99.75% uptime) could create 76 GW of new load capacity. Demand curtailment is technically a valid solution, but the willingness of hyperscalers to curtail remains unknown. Historically, the lowest cost solution would win. Today, data centers are taking an “all of the above” approach, contracting power wherever they can get it at any price, given that the AI race is also a race to access power. Given that, the immediate winners are solar with battery storage, and fuel cells, followed by CCGT. Wildcards are SMR nuclear, enhanced geothermal, and demand curtailment.

The grid-forming BESS sitting behind the data center fence line is either your most powerful compliance asset — or the source of your next oscillation event. There is no middle ground. When a hyperscaler installs a 200 MWh BESS for UPS replacement or backup, the conversation usually starts with capex and ends with runtime. The moment that BESS is configured in grid-forming mode, however, it stops being a load-side asset and becomes a voltage-source converter sitting parallel to the TSO's network. The implications run in both directions. The opportunity is real. A grid-forming BESS at the point of connection can ride through deep voltage dips (FRT compliance, not just survival), inject synthetic inertia in a system where rotational mass is disappearing, and deliver four-quadrant dynamic reactive power — exactly the capability EirGrid's MPID345 proposal is moving to codify for demand facilities. For a developer connecting in Dublin, Northern Virginia, or any weak-grid cluster, this is the difference between a connection agreement you can sign and one you cannot. The risk is equally real. Grid-forming inverters in weak grids are operating-point-dependent control systems coupled to the network impedance. The same hardware can be stable today and unstable tomorrow if a nearby line trips and the Thevenin impedance shifts. Sub-synchronous control interactions, harmonic resonance, forced oscillations — these are no longer textbook problems. They are documented operational events. Full EMT simulation across the operating envelope is now the minimum bar. White-box impedance models are increasingly being requested by TSOs. Reactive capability must remain available across the full BESS state-of-charge range, including the awkward zone where the unit is charging at high power and grid voltage is dropping. A grid-forming-capable BESS costs more. EMT validation adds 6–12 months and specialist fees. The alternative — discovering at the connection study stage that your site is unbuildable — is far more expensive. The hyperscalers who treat behind-the-meter GFM BESS as a strategic system asset, not a procurement line item, will close their connections in 2027 and beyond. The ones who don't will be reading the EMT report and wondering why. → Subscribe to GridStab News for technical deep-dives on grid-forming control, EMT validation, and the operational realities of hyperscale grid integration. #GridStability #GridForming #DataCenterPower