Two Basketball Courts of Batteries Could Replace a Power Plant
How NineDot Energy is using data to solve NYC's dirtiest energy problem
The South Bronx at 3 PM on a July Afternoon
On those 100°F days when every air conditioner in New York City runs at full blast, the grid hits its breaking point. Power demand spikes from 5,500 MW to over 10,000 MW. That's when 15 aging peaker plants scattered across the city fire up, burning fossil fuels to keep the lights on. They run just 500 hours annually—about 6% of the year—yet New Yorkers pay $4.5 billion per decade to keep them ready.
The data tells a specific story: 750,000 New Yorkers live within one mile of these plants. Among them, 78% are either low-income or people of color. In neighborhoods like Hunts Point and Mott Haven, childhood asthma rates run triple the national average. The correlation between zip code, air quality, and respiratory illness is so consistent it might as well be causal.
Three physicists who worked at the Department of Energy looked at these numbers and saw an engineering problem with a data-driven solution.
From Government Lab to Energy Startup
Adam Cohen spent his days at the Department of Energy running complex models and simulations. His conclusion was counterintuitive: "The really hard parts about making clean energy get deployed at scale have almost nothing to do with science and technology. They have to do with human decision-making and regulations and permitting."
Cohen teamed up with David Arfin, who'd developed SolarCity's financial modeling for solar panel leasing, and Nalin Kulatilaka, whose academic research focused on quantifying risk in clean energy investments using Monte Carlo simulations and real options analysis. They named their company NineDot Energy after the classic lateral thinking puzzle—nine dots arranged in a square that must be connected with four straight lines without lifting the pencil.
Their first model assumed community solar could work in NYC. The spreadsheets looked good until someone asked where they'd put 30 acres of panels in Manhattan. They pivoted to fuel cells, running feasibility studies on converting natural gas to electricity at 60% efficiency. When that hit regulatory roadblocks, they turned to battery storage. Each failed model taught them something about New York's energy market that no amount of theoretical analysis could have revealed.
The 10,000 Square Foot Algorithm
By 2020, after three years of iterations, NineDot's data revealed a critical insight: New York City's zoning code had a loophole. Energy storage systems under 10,000 square feet—about two basketball courts—could be built in commercial and manufacturing districts as-of-right, meaning no special permits required. In December 2023, the city expanded this to residential neighborhoods.
The math was elegant. Each 10,000 square foot installation could house 5 MW of batteries, storing 20 MWh of energy. That's enough to power 5,000 homes for four hours during peak demand. NineDot's software tracked electricity prices in real-time, charging batteries when rates dropped below $30/MWh overnight and discharging when prices spiked above $200/MWh during peak hours.
Their financial models showed internal rates of return exceeding 15% based on three revenue streams: capacity payments from ConEd, energy arbitrage, and New York State Energy Research and Development Authority (NYSERDA) incentives. The algorithms optimized charge-discharge cycles to maximize revenue while maintaining battery health—never charging above 90% or discharging below 10% to extend lifespan to 15 years.
Institutional Capital Meets Community Need
Carlyle Group's infrastructure fund ran their own models on NineDot's projections. Their analysts saw what the founders saw: predictable cash flows, regulatory support, and scalable technology. They invested $100 million in 2022. When NineDot's first projects performed within 3% of projections, Carlyle returned with Manulife Investment Management for another $225 million in 2024.
First Citizens Bank structured a $65 million equipment financing deal based on Tesla Megapack performance data showing 92% round-trip efficiency and 0.05% monthly degradation rates. The financing covered nearly 100 MW across 20 sites—enough data points to prove the model at scale.
New York's Climate Leadership and Community Protection Act requires 35% of clean energy investments to benefit disadvantaged communities. NineDot's site selection algorithm weights census tract data on income, pollution burden, and health outcomes. Their first installation went into Pelham Gardens in the Bronx, where PM2.5 levels exceed EPA standards 30 days per year.
Real-Time Grid Management
NineDot's operations center runs on a tech stack built for high-frequency trading, adapted for energy markets. Their systems process:
ConEd's five-minute pricing signals
Weather forecasts at 1km resolution
Grid frequency deviations to 0.01 Hz
Battery state-of-charge for every cell
Predictive maintenance alerts based on temperature anomalies
During the July 2024 heat wave, when temperatures hit 103°F, their algorithms predicted peak demand would hit at 5:17 PM based on historical patterns adjusted for day-of-week and weather conditions. The system began discharging at 4:45 PM, ramping to full output by 5:15 PM—two minutes before actual peak.
Adam Cohen monitored the event from his laptop in a DC hotel room. "When they got off their school buses, the system was fully charged at 98.5% capacity. As air conditioning load ramped up across the city, our batteries discharged 19.7 MWh over four hours, exactly as modeled."
The Ravenswood Test Case
Rise Light & Power, which owns Ravenswood Generating Station—NYC's largest power plant at 2,480 MW—is running an even bigger experiment. They're installing 316 MW of batteries where they just retired 500 MW of gas peakers. LS Power, their parent company, uses machine learning models trained on 20 years of NYISO data to optimize the replacement ratio.
The challenge: current lithium-ion batteries discharge in four hours, but summer peak demand can last six to eight hours. Rise Light & Power's solution involves two battery systems operating in sequence, with proprietary software managing the handoff. Their models show this configuration can meet 95% of peaker plant duty cycles.
The FDNY required 18 months of safety data before approving the project. Temperature sensors every 10 feet, infrared cameras with anomaly detection, and automatic suppression systems that can isolate individual battery containers. The safety systems generate 100 GB of data daily, processed by pattern recognition algorithms that can predict thermal events 20 minutes before they occur.
Scaling the Model
The numbers define both opportunity and challenge:
NYC has 15 peaker plants running 500 hours/year
Combined capacity: 6,000 MW
NineDot's current pipeline: 400 MW by 2026
NYS target: 6,000 MW of storage by 2030
National storage additions in 2025: 18.2 GW
Required to fully replace NYC peakers: 3,000 MW minimum
Pacific Northwest National Laboratory data shows grid-scale battery safety incidents declined 97% from 2018 to 2023, with mean time between failures now exceeding 100,000 hours. But perception lags data. Community board meetings still feature residents worried about "thermal runaway" and "toxic fumes," despite air quality monitoring showing no detectible emissions from modern lithium iron phosphate batteries.
The PEAK Coalition—UPROSE, THE POINT CDC, New York City Environmental Justice Alliance—uses their own data to push for faster deployment. Their analysis shows retiring all peaker plants would prevent 2.66 million tons of annual CO2 emissions, 1,655 tons of NOx, and 171 tons of SO2. At current pollution-related healthcare costs, that translates to $47 million in annual savings.
The Infrastructure Stack
NineDot's technology choices reflect Silicon Valley thinking applied to utility infrastructure:
Battery Management: Tesla Megapacks with Autobidder software
Grid Integration: SEL-751 relays with DNP3 protocol
Monitoring: Grafana dashboards pulling from InfluxDB time-series database
Optimization: Custom Python scripts using SciPy optimization libraries
Trading: API integration with NYISO's market system
Maintenance: Predictive algorithms based on NASA's battery degradation models
Each site runs autonomously but reports to a central network operations center in Long Island City. The system processes 10 million data points per hour across all sites. Machine learning models identify patterns humans might miss—like how battery performance correlates with barometric pressure changes before storms.
What the Data Says About the Future
The New York Independent System Operator will keep four peaker plants running past their May 2025 retirement date—the Gowanus and Narrows barges—as grid insurance. Their models show a 446 MW reliability gap without them. But that gap is shrinking by 200 MW annually as battery storage comes online.
By 2030, if deployment continues at current rates, NYC will have 4,000 MW of battery storage. That's enough to handle 80% of peaker plant duties. The remaining 20%—extended heat waves lasting over eight hours—might require different technology. Iron-air batteries from Form Energy, gravity storage from Energy Vault, or compressed air systems could fill that gap.
The transition won't be smooth. The Champlain Hudson Power Express transmission line, bringing 1,250 MW of Canadian hydropower to NYC, has been delayed from 2025 to 2027. Until more transmission arrives, batteries and peakers will coexist, with software orchestrating an increasingly complex dance between old and new infrastructure.
The Model Spreads
NineDot isn't alone anymore. Convergent Energy, Plus Power, and Key Capture Energy are all deploying similar strategies. Combined, they have 2,000 MW of battery storage in development across New York State. Each company uses slightly different algorithms, battery chemistry, and financing structures, but the core model is identical: use data and software to turn batteries into grid infrastructure.
Three physicists who met in a government office proved that thinking outside the box meant building actual boxes—battery boxes, two basketball courts at a time, using data to optimize every electron. The peaker plants still stand, ready for the hottest days. But for the first time in decades, there's a scalable alternative.
The nine dots are being connected, one battery installation at a time. It took $325 million in capital, thousands of hours of software development, and terabytes of data analysis. But the model works. And it fits in a space smaller than a Brooklyn playground.