Seven Million Dollars and the Margin Migration

Stem Inc. posted its first profitable year on March 4. Adjusted EBITDA came in at roughly $7 million for 2025, a number that would barely register on most industrial balance sheets.

The company that spent years as a hardware integrator, buying battery cells and assembling systems for commercial buildings, made its money in 2025 by selling software. Revenue from software, services, and edge hardware grew 25 percent year over year to $141 million, now comprising more than 55 percent of total revenue. Non-GAAP gross margins expanded to 45 percent, up from 36 percent the prior year. Annualized recurring revenue hit $61 million, up 16 percent year over year. The 2026 guidance: $10 to $15 million in adjusted EBITDA.

The numbers are small. The direction is not.

The margin inversion. For most of the past decade, commercial battery storage economics centered on hardware. Cell procurement, pack assembly, power electronics, and installation drove the cost structure. Integrators competed on system price per kilowatt-hour. Software was an afterthought, bundled free or priced as a rounding error on the hardware contract.

Stem’s Q4 results invert that logic. Hardware integration, the business the company is deliberately shrinking, carries margins in the low teens. Software and services now generate margins above 45 percent. The company’s new products, PowerTrack EMS and PowerTrack Sage, launch broadly by the end of March, layered on top of a fleet management platform that already oversees gigawatt-hours of deployed assets. The pitch is no longer “buy our battery.” It is “let our software optimize whatever battery you already bought.”

The hardware floor is falling. On March 5, BYD unveiled its second-generation Blade Battery in Shenzhen. Energy density jumped 40 percent to 210 Wh/kg. The product targets electric vehicles, but BYD’s LFP cell improvements at this scale cascade into stationary storage within 12 to 18 months. BYD is already a major commercial and industrial BESS supplier globally.

Simultaneously, Redwood Materials’ energy division has deployed a 12 MW, 63 MWh system using repurposed EV batteries for a Crusoe AI data center, with more than a gigawatt-hour in its reusable battery pipeline. The company is designing projects at 100 MW scale. Second-life LFP cells at potentially sub-$50 per kilowatt-hour represent a new competitive floor for hardware costs.

When cell costs fall and form factors multiply, the differentiator shifts. A 40 percent improvement in energy density does not help a building owner who cannot dispatch that capacity at the right fifteen-minute interval. A sub-$50 cell does not help if the control system cannot stack demand charge reduction, time-of-use arbitrage, and demand response participation in real time. The optimization layer captures the value that cheaper hardware creates.

The market context. Bimergen Energy, trading under the ticker BESS on the NYSE, is building the opposite model. The company completed its acquisition of 79.2 MW of distributed battery projects in ERCOT South from Aggreko on March 3, all designed at 9.9 MW to stay under simplified interconnection thresholds. Bimergen’s thesis is pure asset ownership: frequency regulation, voltage support, energy arbitrage, and capacity revenue captured through owning and operating hardware. It is a legitimate model, but one where margins are governed by cell procurement costs, construction timelines, and RTO market rules rather than software economics.

Both models can work. The question is which scales faster and which captures more margin per deployed megawatt-hour as the fleet grows.

The fleet is growing. The EIA reported on March 5 that U.S. electricity generation hit a record 4.43 terawatt-hours in 2025, up 2.8 percent, the largest annual increase in over a decade. The agency projects 24.3 GW of new battery storage capacity planned for 2026 alone. Every megawatt-hour deployed is a potential software customer, whether or not the software vendor sold the hardware.

This is the dynamic that makes Stem’s $7 million notable. In a market adding gigawatts annually, with hardware costs falling from both new chemistry (BYD’s 210 Wh/kg cells) and second-life supply (Redwood’s repurposed packs), a company selling optimization software to a growing installed base has a fundamentally different cost structure than a company buying and reselling cells.

The competitive risk. Tesla, Fluence, and every major BESS manufacturer are building their own software stacks. Vertical integration could squeeze independent software providers out of the market entirely. Stem’s bet is that a multi-vendor, hardware-agnostic platform wins in a market where building owners do not want to be locked into a single manufacturer’s ecosystem, particularly as the variety of available hardware expands.

Whether Stem specifically sustains profitability matters less than what its margin structure reveals. Commercial battery storage is following the pattern of every maturing hardware market: value migrates from the physical product to the intelligence layer that operates it. Servers became commodity boxes; the money moved to cloud software. Solar panels collapsed in price; the money moved to project finance and grid services platforms.

Seven million dollars in adjusted EBITDA will not make anyone rich. But it is the first empirical proof that the same migration is underway in commercial battery storage. The companies that recognize this early will price their software accordingly. The ones that do not will keep competing on hardware cost per kilowatt-hour as BYD and its peers drive that number toward a floor.

Sources