The United States needs to build a lot of solar, fast. Electricity demand is accelerating driven by data centers and electrification, while the construction labor market for solar farms has been constrained for years. The 100-megawatt milestone Maximo announced(PR Newswire) last week at AES's Bellefield complex in Kern County, California is evidence that the automation answer to that problem has moved past the demo stage.
Bellefield is a 2,000-megawatt solar-plus-storage project under a 15-year power contract with Amazon(Nasdaq). The first phase includes 500 megawatts of solar and 500 megawatts of four-hour battery storage. Maximo, the robotics startup AES incubated(CleanTechnica) to automate solar construction, was deployed there first(PR Newswire) — and the Bellefield result is now the benchmark for what field robotics can deliver at utility scale.
The numbers: four Maximo robots working in parallel(PR Newswire) with human crews installed 100 megawatts of solar modules. Version 3.0 of the system consistently beat one module per minute(PR Newswire). Crews using the robots installed as many as 24 modules per shift hour per worker(PR Newswire) — nearly double the output of traditional installation methods in the same region. The company is targeting a 50 percent reduction in module-installation time(AES Energy Insights) on scopes where the robots operate.
What makes that rate improvement meaningful is the physical reality it replaces. Installing panels on eight-foot torque tubes normally requires three workers on ladders on uneven ground(The Robot Report). The panels are aluminum and glass, the sites are dusty, windy, and off-grid, and the glare off the modules can fool computer vision systems. Deise Yumi Asami, founder of Maximo, told The Robot Report(The Robot Report) that the California site forced the team to solve for conditions that factory-style automation assumptions don't account for.
The four-robot fleet that completed the Bellefield work started as a single unit. Scaling from one to four(PR Newswire) taught the Maximo team how to stage robot movements, coordinate with human crews, and minimize workflow changes — operational lessons that show up in the upcoming v4.0 release.
Maximo used NVIDIA's Isaac Sim robotics simulation framework(The Robot Report) and AI infrastructure to develop and validate fleet behavior before deploying changes in the field. AWS provided the compute and data pipelines for real-time construction intelligence and fleet telemetry. The company also built a mobile microgrid so the robots can operate without being connected to site power — a practical requirement for a construction site that doesn't have power until the solar panels are installed(AES Energy Insights).
The 100-megawatt milestone marks the shift from early deployment validation to sustained commercial production. Maximo has pipeline alignment across 5 gigawatts(AES Energy Insights) and multi-megawatt deployments already completed. This isn't a research project anymore. It's a deployment program with a contracted power buyer, demonstrated throughput, and a fleet learning curve.
The labor constraint driving Maximo's development isn't going away. Andrés Gluski, CEO of AES, told The New York Times(New York Times) in 2024 that labor shortages on U.S. construction sites were a bottleneck to solar farm buildout. The Bellefield result doesn't eliminate that constraint, but it defines the parameters of what robotic augmentation can do about it: not replacing human crews entirely, but roughly doubling their output on the most physically demanding and repetitive part of the installation.
The 5-gigawatt pipeline means the Bellefield learning gets applied at scale. What Maximo proved at 100 megawatts is being designed into projects orders of magnitude larger.
† † Source-reported; not independently verified. Consider: 'Deise Yumi Asami of Maximo' or verify her title in the original source.
†† † Source-reported; not independently verified. Either cite the source confirming AWS's role or rephrase to 'Maximo used cloud compute infrastructure' without specifying AWS.
