Designing Cities That Never Lose Power: AI-Driven Microgrids at Three Scales

In March 2011, when the magnitude-9.0 earthquake struck Japan's Pacific coast, the city of Sendai lost connection to the national grid. But Sendai's microgrid—a locally controlled energy network serving the university district—kept the lights on, the water pumps running, and the emergency shelters heated. A decade later, in February 2021, Texas experienced what engineers politely call a "cascading grid failure" and what 4.5 million households experienced as four days without power in freezing temperatures. The difference between Sendai and Texas was not wealth, geography, or climate. It was architecture: Sendai had invested in distributed energy resilience; Texas had not.

These two events bookend a fundamental question in urban energy engineering: can we design cities where widespread blackouts are structurally impossible? Three 2025 studies, each operating at a different scale of analysis, converge on an answer that is cautiously optimistic—but reveal that the obstacles are as much institutional as they are technical.

The Research Landscape

Scale 1: Global Technology Survey — What Tools Exist?

Almihat and Munda (2025) provide a systematic review of smart grid technologies for urban energy planning, synthesizing evidence across four technology pillars and three governance models. Their contribution is not to propose new technology but to map what already works, what is scaling, and what remains experimental.

The four technology pillars they identify are:

• AI-based Energy Management Systems (AI-EMS): Machine learning algorithms that forecast renewable generation, predict demand, and optimize storage dispatch in real time. These systems have moved beyond research prototypes into commercial deployment, with companies like AutoGrid and Opus One Solutions operating AI-EMS platforms across hundreds of microgrids.
• Hybrid Energy Storage Systems (HESS): Combinations of batteries (for fast response) with longer-duration storage such as hydrogen, compressed air, or thermal mass. No single storage technology addresses all timescales—batteries handle minute-to-hour fluctuations; hydrogen or thermal storage handles multi-day or seasonal gaps.
• Peer-to-Peer (P2P) Energy Trading: Platforms that allow households and buildings to buy and sell energy directly, without routing transactions through a central utility. Brooklyn's Transactive Grid experiment demonstrated the concept; regulatory frameworks in Australia and Germany are enabling broader deployment.
• Edge AI: On-device intelligence embedded in inverters, meters, and controllers that enables real-time decision-making without relying on cloud connectivity—critical for resilience when communication networks themselves are disrupted by the same events that threaten the grid.

Equally important are the three governance models Almihat and Munda identify as enablers: public-private partnerships (PPPs) that share infrastructure investment risk, local energy markets that create financial incentives for distributed generation, and performance-based incentives that reward reliability outcomes rather than capacity installation.

The real-world cases they examine are instructive. Brooklyn's P2P trading experiment demonstrated that prosumers (households that both produce and consume energy) will participate in local energy markets when the price signals are transparent. Sendai's microgrid proved that islanding capability—the ability to disconnect from a failing national grid and operate independently—is not a theoretical concept but a life-saving one. California's wildfire-response microgrids showed that utilities facing existential liability risk (Pacific Gas & Electric's bankruptcy following wildfire liability) will invest in microgrids when the alternative is corporate dissolution.

Scale 2: 50-Year System Dynamics — What Happens Over Decades?

Rozhkov (2025) shifts from technology to temporal dynamics, modeling six energy transition scenarios for the Illinois ComEd service territory over a 50-year horizon using system dynamics simulation. System dynamics is a methodology that captures feedback loops—the way that decisions made today alter the conditions under which future decisions are made. In energy planning, these feedback loops are pervasive: renewable investment reduces electricity prices, which reduces the financial incentive for further renewable investment; carbon taxes increase fossil fuel costs, which accelerates renewable adoption, which drives down renewable costs through manufacturing scale, which further accelerates adoption.

The six scenarios Rozhkov models are:

Business-as-usual (BAU): Continuation of current policies and investment patterns.

Renewable-first (REN): Aggressive renewable deployment with carbon pricing.

Decentralized (DEC): Maximum consumer sovereignty—prosumer microgrids, P2P trading, minimal utility involvement.

Clean Energy Jobs Act (CEJA): Illinois's actual 2021 climate legislation, modeled as enacted.

Nuclear extension (NUC): Extending existing nuclear plant lifespans with modest renewable additions.

Electrification (ELEC): Rapid electrification of heating and transport with mixed generation.

The findings are sobering in their complexity. No single scenario optimizes all metrics simultaneously:

• REN achieves the deepest CO2 reductions at -78.9% by 2070, but at the highest infrastructure cost and with periods of grid instability during the transition when fossil plants retire faster than storage scales.
• DEC produces the best consumer outcomes—lowest electricity bills, highest energy independence—but drives utility revenue to zero, creating a utility "death spiral" where the infrastructure owner cannot finance grid maintenance. If the grid is needed as backup for microgrids during extended low-renewable periods, someone must pay for it.
• CEJA achieves a -64.1% CO2 reduction, which is neither the deepest cut nor the cheapest path, but represents a realistic balance that maintains utility viability, achieves meaningful decarbonization, and does not require technologies that are still pre-commercial.

The central insight is that optimizing for environmental outcomes, consumer outcomes, and institutional stability produces fundamentally different system architectures. Policymakers who promise to achieve all three simultaneously are either confused or dissembling.

Scale 3: NEOM City Design — What Does 100% Renewable Look Like?

Alharbi, Ali, and Diab (2025) move from simulation to engineering design, using HOMER (Hybrid Optimization of Multiple Energy Resources) to design the microgrid for NEOM, Saudi Arabia's planned $500 billion megaproject in Tabuk Province. NEOM is an unusual case study because it is a greenfield city with no legacy infrastructure, abundant solar and wind resources (6.0+ kWh/m2/day solar irradiance, 6-8 m/s average wind speed), and a stated commitment to 100% renewable energy.

The study evaluates four configurations:

• C1: Photovoltaic (PV) only with battery storage.
• C2: PV + wind turbines + battery storage (BSS).
• C3: PV + wind turbines + battery storage + hydrogen electrolyzer and fuel cell.
• C4: PV + wind turbines + battery + hydrogen + diesel backup.

Key results:

• C2 is cost-optimal at $0.0968/kWh levelized cost of energy (LCOE), achieving 100% renewable supply through the complementarity of solar (daytime) and wind (often stronger at night in the Tabuk region). For context, $0.0968/kWh is lower than the average retail electricity price in most countries.
• C3 adds hydrogen at a premium, producing green hydrogen at $2.35/kg. This is significant because the Saudi government's hydrogen export strategy targets $2.00/kg by 2030; the HOMER optimization suggests that a city-scale microgrid could produce hydrogen as a byproduct of its energy system at near-target costs.
• All configurations achieve zero CO2 emissions (C4 includes diesel but the optimizer does not dispatch it because renewables are cheaper).
• Net present cost (NPC) savings are approximately 85% compared to a diesel-only baseline, entirely eliminating fuel import dependency.

The hydrogen dimension deserves emphasis. C3's slightly higher electricity cost ($0.1035/kWh vs. C2's $0.0968) buys two things: enhanced multi-day storage resilience (hydrogen can store energy for weeks, unlike batteries that are economical for hours) and a revenue-generating exportable commodity. For a nation whose primary export is fossil fuel, building cities that produce exportable green hydrogen is not just an energy strategy—it is an economic transition strategy.

Critical Analysis: Claims and Evidence

<

Claim	Source	Evidence Type	Verdict
AI-EMS, HESS, P2P, and Edge AI form a complete smart grid technology stack	Almihat & Munda (2025)	Systematic literature review + real-world cases	✅ Well-supported — multiple deployed examples cited
No single energy transition scenario optimizes environment, consumer cost, and institutional stability simultaneously	Rozhkov (2025)	System dynamics simulation, 6 scenarios, 50 years	✅ Supported — but model assumptions about feedback loop parameters are debatable
Decentralized (DEC) scenario drives utility revenue to zero	Rozhkov (2025)	Simulation with utility financial model	⚠️ Plausible but extreme — assumes no regulatory adaptation over 50 years
NEOM microgrid achieves $0.0968/kWh 100% renewable	Alharbi et al. (2025)	HOMER optimization with local irradiance and wind data	✅ Technically sound — but HOMER assumes perfect component availability and no construction delays
Green hydrogen at $2.35/kg is feasible as microgrid byproduct	Alharbi et al. (2025)	HOMER techno-economic analysis, C3 configuration	⚠️ Promising — electrolyzer costs and degradation rates carry significant uncertainty at city scale
Sendai microgrid maintained power during the 2011 earthquake	Almihat & Munda (2025), citing prior literature	Historical case documentation	✅ Verified — well-documented in IEEE and government reports

Open Questions

The utility death spiral: Rozhkov's DEC scenario reveals a structural problem: if microgrids allow consumers to exit the grid, who maintains the transmission infrastructure needed as backup? This is not a technical question but an institutional design problem that no jurisdiction has fully solved.

Scaling NEOM's design to existing cities: NEOM is greenfield—no existing buildings, no legacy grid, no incumbent utility with stranded assets. Retrofitting a microgrid architecture into Chicago, Lagos, or Mumbai involves orders-of-magnitude more complexity. How much of the HOMER-optimized design translates to brownfield contexts?

Hydrogen storage economics at scale: Alharbi et al.'s $2.35/kg assumes current electrolyzer costs. Electrolyzer manufacturing is scaling rapidly, but degradation rates at 15+ years of operation in desert conditions are not yet empirically established. The gap between modeled and realized performance could be significant.

Cybersecurity of AI-EMS and Edge AI: Almihat and Munda identify Edge AI as critical for resilience, but intelligent devices connected to energy infrastructure are high-value cyberattack targets. The 2015 and 2016 attacks on Ukraine's power grid demonstrated that state-level actors will exploit grid control systems.

Social equity in P2P markets: Peer-to-peer energy trading benefits those who can afford solar panels and batteries. Without deliberate policy design, microgrid-enabled P2P markets could create a two-tier energy system where affluent prosumers enjoy low costs and high resilience while renters and low-income households remain dependent on a degrading centralized grid.

What This Means for Your Research

These three studies, read together, outline a progression from technology catalog (Almihat & Munda) to system behavior over time (Rozhkov) to concrete engineering design (Alharbi et al.). For energy systems researchers, the key gap is the middle ground: most published work either surveys technologies or optimizes specific installations, but the 50-year feedback dynamics that Rozhkov explores—where today's investments alter tomorrow's incentive structures—remain undermodeled. For urban planners and policymakers, the central lesson from Rozhkov's work is uncomfortable but important: you cannot simultaneously optimize for environment, consumer cost, and institutional stability. Choose two, and design governance to manage the third.

Explore related work through ORAA ResearchBrain.

In March 2011, when the magnitude-9.0 earthquake struck Japan's Pacific coast, the city of Sendai lost connection to the national grid. But Sendai's microgrid—a locally controlled energy network serving the university district—kept the lights on, the water pumps running, and the emergency shelters heated. A decade later, in February 2021, Texas experienced what engineers politely call a "cascading grid failure" and what 4.5 million households experienced as four days without power in freezing temperatures. The difference between Sendai and Texas was not wealth, geography, or climate. It was architecture: Sendai had invested in distributed energy resilience; Texas had not.

These two events bookend a fundamental question in urban energy engineering: can we design cities where widespread blackouts are structurally impossible? Three 2025 studies, each operating at a different scale of analysis, converge on an answer that is cautiously optimistic—but reveal that the obstacles are as much institutional as they are technical.

The Research Landscape

Scale 1: Global Technology Survey — What Tools Exist?

Almihat and Munda (2025) provide a systematic review of smart grid technologies for urban energy planning, synthesizing evidence across four technology pillars and three governance models. Their contribution is not to propose new technology but to map what already works, what is scaling, and what remains experimental.

The four technology pillars they identify are:

• AI-based Energy Management Systems (AI-EMS): Machine learning algorithms that forecast renewable generation, predict demand, and optimize storage dispatch in real time. These systems have moved beyond research prototypes into commercial deployment, with companies like AutoGrid and Opus One Solutions operating AI-EMS platforms across hundreds of microgrids.
• Hybrid Energy Storage Systems (HESS): Combinations of batteries (for fast response) with longer-duration storage such as hydrogen, compressed air, or thermal mass. No single storage technology addresses all timescales—batteries handle minute-to-hour fluctuations; hydrogen or thermal storage handles multi-day or seasonal gaps.
• Peer-to-Peer (P2P) Energy Trading: Platforms that allow households and buildings to buy and sell energy directly, without routing transactions through a central utility. Brooklyn's Transactive Grid experiment demonstrated the concept; regulatory frameworks in Australia and Germany are enabling broader deployment.
• Edge AI: On-device intelligence embedded in inverters, meters, and controllers that enables real-time decision-making without relying on cloud connectivity—critical for resilience when communication networks themselves are disrupted by the same events that threaten the grid.

Equally important are the three governance models Almihat and Munda identify as enablers: public-private partnerships (PPPs) that share infrastructure investment risk, local energy markets that create financial incentives for distributed generation, and performance-based incentives that reward reliability outcomes rather than capacity installation.

The real-world cases they examine are instructive. Brooklyn's P2P trading experiment demonstrated that prosumers (households that both produce and consume energy) will participate in local energy markets when the price signals are transparent. Sendai's microgrid proved that islanding capability—the ability to disconnect from a failing national grid and operate independently—is not a theoretical concept but a life-saving one. California's wildfire-response microgrids showed that utilities facing existential liability risk (Pacific Gas & Electric's bankruptcy following wildfire liability) will invest in microgrids when the alternative is corporate dissolution.

Scale 2: 50-Year System Dynamics — What Happens Over Decades?

Rozhkov (2025) shifts from technology to temporal dynamics, modeling six energy transition scenarios for the Illinois ComEd service territory over a 50-year horizon using system dynamics simulation. System dynamics is a methodology that captures feedback loops—the way that decisions made today alter the conditions under which future decisions are made. In energy planning, these feedback loops are pervasive: renewable investment reduces electricity prices, which reduces the financial incentive for further renewable investment; carbon taxes increase fossil fuel costs, which accelerates renewable adoption, which drives down renewable costs through manufacturing scale, which further accelerates adoption.

The six scenarios Rozhkov models are:

Business-as-usual (BAU): Continuation of current policies and investment patterns.

Renewable-first (REN): Aggressive renewable deployment with carbon pricing.

Decentralized (DEC): Maximum consumer sovereignty—prosumer microgrids, P2P trading, minimal utility involvement.

Clean Energy Jobs Act (CEJA): Illinois's actual 2021 climate legislation, modeled as enacted.

Nuclear extension (NUC): Extending existing nuclear plant lifespans with modest renewable additions.

Electrification (ELEC): Rapid electrification of heating and transport with mixed generation.

The findings are sobering in their complexity. No single scenario optimizes all metrics simultaneously:

• REN achieves the deepest CO2 reductions at -78.9% by 2070, but at the highest infrastructure cost and with periods of grid instability during the transition when fossil plants retire faster than storage scales.
• DEC produces the best consumer outcomes—lowest electricity bills, highest energy independence—but drives utility revenue to zero, creating a utility "death spiral" where the infrastructure owner cannot finance grid maintenance. If the grid is needed as backup for microgrids during extended low-renewable periods, someone must pay for it.
• CEJA achieves a -64.1% CO2 reduction, which is neither the deepest cut nor the cheapest path, but represents a realistic balance that maintains utility viability, achieves meaningful decarbonization, and does not require technologies that are still pre-commercial.

The central insight is that optimizing for environmental outcomes, consumer outcomes, and institutional stability produces fundamentally different system architectures. Policymakers who promise to achieve all three simultaneously are either confused or dissembling.

Scale 3: NEOM City Design — What Does 100% Renewable Look Like?

Alharbi, Ali, and Diab (2025) move from simulation to engineering design, using HOMER (Hybrid Optimization of Multiple Energy Resources) to design the microgrid for NEOM, Saudi Arabia's planned $500 billion megaproject in Tabuk Province. NEOM is an unusual case study because it is a greenfield city with no legacy infrastructure, abundant solar and wind resources (6.0+ kWh/m2/day solar irradiance, 6-8 m/s average wind speed), and a stated commitment to 100% renewable energy.

The study evaluates four configurations:

• C1: Photovoltaic (PV) only with battery storage.
• C2: PV + wind turbines + battery storage (BSS).

• C3: PV + wind turbines + battery storage + hydrogen electrolyzer and fuel cell.
• C4: PV + wind turbines + battery + hydrogen + diesel backup.

Key results:

• C2 is cost-optimal at $0.0968/kWh levelized cost of energy (LCOE), achieving 100% renewable supply through the complementarity of solar (daytime) and wind (often stronger at night in the Tabuk region). For context, $0.0968/kWh is lower than the average retail electricity price in most countries.
• C3 adds hydrogen at a premium, producing green hydrogen at $2.35/kg. This is significant because the Saudi government's hydrogen export strategy targets $2.00/kg by 2030; the HOMER optimization suggests that a city-scale microgrid could produce hydrogen as a byproduct of its energy system at near-target costs.
• All configurations achieve zero CO2 emissions (C4 includes diesel but the optimizer does not dispatch it because renewables are cheaper).
• Net present cost (NPC) savings are approximately 85% compared to a diesel-only baseline, entirely eliminating fuel import dependency.

The hydrogen dimension deserves emphasis. C3's slightly higher electricity cost ($0.1035/kWh vs. C2's $0.0968) buys two things: enhanced multi-day storage resilience (hydrogen can store energy for weeks, unlike batteries that are economical for hours) and a revenue-generating exportable commodity. For a nation whose primary export is fossil fuel, building cities that produce exportable green hydrogen is not just an energy strategy—it is an economic transition strategy.

Critical Analysis: Claims and Evidence

<

Claim	Source	Evidence Type	Verdict
AI-EMS, HESS, P2P, and Edge AI form a complete smart grid technology stack	Almihat & Munda (2025)	Systematic literature review + real-world cases	✅ Well-supported — multiple deployed examples cited
No single energy transition scenario optimizes environment, consumer cost, and institutional stability simultaneously	Rozhkov (2025)	System dynamics simulation, 6 scenarios, 50 years	✅ Supported — but model assumptions about feedback loop parameters are debatable
Decentralized (DEC) scenario drives utility revenue to zero	Rozhkov (2025)	Simulation with utility financial model	⚠️ Plausible but extreme — assumes no regulatory adaptation over 50 years
NEOM microgrid achieves $0.0968/kWh 100% renewable	Alharbi et al. (2025)	HOMER optimization with local irradiance and wind data	✅ Technically sound — but HOMER assumes perfect component availability and no construction delays
Green hydrogen at $2.35/kg is feasible as microgrid byproduct	Alharbi et al. (2025)	HOMER techno-economic analysis, C3 configuration	⚠️ Promising — electrolyzer costs and degradation rates carry significant uncertainty at city scale
Sendai microgrid maintained power during the 2011 earthquake	Almihat & Munda (2025), citing prior literature	Historical case documentation	✅ Verified — well-documented in IEEE and government reports

Open Questions

The utility death spiral: Rozhkov's DEC scenario reveals a structural problem: if microgrids allow consumers to exit the grid, who maintains the transmission infrastructure needed as backup? This is not a technical question but an institutional design problem that no jurisdiction has fully solved.

Scaling NEOM's design to existing cities: NEOM is greenfield—no existing buildings, no legacy grid, no incumbent utility with stranded assets. Retrofitting a microgrid architecture into Chicago, Lagos, or Mumbai involves orders-of-magnitude more complexity. How much of the HOMER-optimized design translates to brownfield contexts?

Hydrogen storage economics at scale: Alharbi et al.'s $2.35/kg assumes current electrolyzer costs. Electrolyzer manufacturing is scaling rapidly, but degradation rates at 15+ years of operation in desert conditions are not yet empirically established. The gap between modeled and realized performance could be significant.

Cybersecurity of AI-EMS and Edge AI: Almihat and Munda identify Edge AI as critical for resilience, but intelligent devices connected to energy infrastructure are high-value cyberattack targets. The 2015 and 2016 attacks on Ukraine's power grid demonstrated that state-level actors will exploit grid control systems.

Social equity in P2P markets: Peer-to-peer energy trading benefits those who can afford solar panels and batteries. Without deliberate policy design, microgrid-enabled P2P markets could create a two-tier energy system where affluent prosumers enjoy low costs and high resilience while renters and low-income households remain dependent on a degrading centralized grid.

What This Means for Your Research

These three studies, read together, outline a progression from technology catalog (Almihat & Munda) to system behavior over time (Rozhkov) to concrete engineering design (Alharbi et al.). For energy systems researchers, the key gap is the middle ground: most published work either surveys technologies or optimizes specific installations, but the 50-year feedback dynamics that Rozhkov explores—where today's investments alter tomorrow's incentive structures—remain undermodeled. For urban planners and policymakers, the central lesson from Rozhkov's work is uncomfortable but important: you cannot simultaneously optimize for environment, consumer cost, and institutional stability. Choose two, and design governance to manage the third.

Explore related work through ORAA ResearchBrain.