Demand charges eat half the bill. Novele just raised $17M to attack them.

At 3:45 on a Tuesday morning, a closed retail store draws enough current to set the highest number on its meter for the whole month. The spike lasts eleven minutes. Days later the building's owner receives a bill in which that one burst can account for more than half the total — a demand charge, priced on the worst moment instead of the average one.

On September 30, Novele closed an oversubscribed $17 million Series A led by boisei labs to chase exactly that number. The Stamford and Irvine startup builds 2-inch-thick lithium iron phosphate boards that mount on the inside wall of an occupied commercial building, plus a control layer that learns the building's load shape well enough to step off the grid before the spike arrives. Locke Health, Arup Ventures and construction firm Barton Malow joined the round, alongside commercial real estate owners the company would not name.

CEO Charles Conwell frames the bet as stranded capacity rather than new capacity. The megawatts the grid needs through the hottest afternoon hour are already sitting inside the buildings people work in. Novele's own materials put more than 2,000 buildings in a 2.6-billion-square-foot pipeline, with live systems running at Fortune 50 customers it also declines to identify.

The eleven-minute peak

A commercial electricity bill has two parts. The energy charge is billed on kilowatt-hours, and it rewards using less. The demand charge is billed on the highest draw inside a set window during the billing period, and it punishes one thing only: spiking. BoardOS, Novele's control software, publishes a detail that explains the whole design. The peak that sets the demand charge, the company writes, can form and pass in eleven minutes — at 3:45 in the morning, in a store that is closed. The bill arrives a month after the event.

Commercial operators already know this, which is why most utilities run demand-response programmes paying buildings to cut load when the grid asks. What breaks in practice is the tenant. Someone has to stop a dishwasher, hold back an EV charge, or accept a dimmer lobby, at an hour the grid operator picked. Conwell called the alternative either ineffective or something that makes tenants unhappy.

Timing is not incidental. PJM's capacity price has moved further in three auctions than in the preceding decade. The 2024/25 base residual auction cleared at $28.92 per megawatt-day. The 2025/26 auction cleared at $269.92. The 2026/27 auction cleared at $329.17 in every zone. The December 2025 auction, which set the 2027/28 delivery year, cleared at $333.44 — a third consecutive record, and the first in PJM's history to leave the operator short of its reliability target across the whole footprint. It procured 145,777 MW of capacity and missed the 20% installed reserve margin requirement by about 6,500 MW. The operator raised its peak demand forecast by 5,250 MW for that auction, almost entirely data centres.

What five kilowatt-hours can and cannot shave

Each EnergyBoard holds 5 kWh and delivers 2 kW continuous, 2.64 kW peak. The cells are lithium iron phosphate, chosen for a documented safety record rather than for maximum energy density, which is a deliberate trade in a product meant to hang inside an occupied room. The enclosure meets UL 50 Type 1 — the same standard as the electrical equipment already in the building. UL 1973 and UN 38.3 cover the battery and its transport. UL 9540 is the system listing. UL 9540A is the fire test fire marshals rely on when deciding whether a battery may go indoors.

In May 2026 Novele completed installation-level large-scale fire testing to the sixth edition of UL 9540A, published March 13 that year. The result supports cutting the required separation space between units by more than 60%, and it covered an alternative cell from a second qualified supplier. That edition matters more than the length of a certification list. UL 9540A is the only fire and explosion test method referenced in the 2026 edition of NFPA 855, and the revised Section 10 demands an explicit large-scale fire test that reports written to the fifth edition do not satisfy.

Now the arithmetic the round does not fix. Demand charges are set by transients: a compressor starting, a bank of elevators, a server hall spinning up. Covering one means supplying its power for its duration, and that is a C-rate problem rather than a capacity problem. A board carrying 5 kWh at 2.64 kW covers roughly half an hour at nameplate. It handles the ordinary bumps of a commercial load. It does not handle a building's sharpest event, and the capacity tag that sets an industrial building's bill is built from five specific hours rather than from an average.

The boards are modular and parallel with no fixed maximum, so a building installs enough of them to hold its own worst hour. That design choice pushes the real question into the tariff. Where the demand charge is set on a 15-minute window, shaving 400 kW for half an hour is worth real money. Where the demand charge is flat, the same hardware is expensive wall decoration.

The company's payback figures need attribution rather than adoption. boisei labs, the lead investor, puts demand-charge reduction at 25–45% with payback of four to seven years, dropping under three once utility or government incentives are counted. Conwell told Heatmap separately that payback is typically 20 to 40 months. Both describe the same product and the same deployments. The gap is the incentive layer, and incentives are the variable that changes most by jurisdiction. Novele is targeting California, New York, New England and parts of PJM for that reason.

What the $17 million is actually buying

$17 million does not buy a factory. It buys deployment crews, engineering headcount and more training data for BoardOS. Novele builds vertically from the cell up, which is an expensive way to own the decision loop. Its argument is that software bolted onto someone else's battery crosses a vendor boundary on every dispatch, and that decision and action have to land in the same millisecond.

The capacity the grid needs is already sitting inside the buildings we work in every day. We just had to build something smart enough to use it.— Charles Conwell, co-founder and CEO, Novele

The competition is not thin. Tesla has sold Megapack systems to commercial customers for years, and Voltus — a demand-response platform that raised $225 million of its own in October — aggregates commercial load while owning no hardware at all. Novele's wedge is the indoor envelope. A Megapack lives outdoors in a container. Thin panels can go inside a hospital corridor or an office and be distributed across the floor, running as one coordinated system instead of a single fenced asset.

As we wrote in September, the version of this problem that drew the most attention this year was Google's: paying residential battery owners to release grid capacity for its data centres. That model reaches megawatt scale by recruiting households. A commercial building is already wired, already metered and already paying the demand charge, which makes it the better place to shave a peak. It is also a slower sale, because the buyer is a facilities manager with a capital request rather than a consumer with a subscription.

A second caution sits inside the region's accreditation rules. The operator has cut the accredited capacity of demand resources and of battery storage in consecutive auctions, so a megawatt nominated into demand response earns less credited capacity each year even as the need for flexibility climbs. The roughly $325/MW-day collar agreed with Pennsylvania and extended by FERC through the 2029/30 auction holds that price signal down at the same time. A stable price makes a facilities budget easier. It is a poor signal to build anything against.

Does shaving commercial peaks change what PJM actually builds?

Where this lands