The editorial argues the crossover is fundamentally a cost-curve story, not an environmental one. Solar's 90% LCOE decline since 2010 came from unglamorous industrial grind — Chinese polysilicon overcapacity, standardized modules, tracker maturation, and balance-of-system process learning — that compounded faster than coal operators believed possible.
The editorial emphasizes that this isn't a single-month blip from a heatwave or coal outage but a sustained inversion telegraphed for years by capacity additions. Utility-scale solar PPAs now clear at $20-35/MWh — below the marginal cost of running existing coal plants — and each coal retirement is a one-way door.
By submitting the Guardian piece framing solar overtaking coal as a 'first time' historic milestone, the submitter signals this is a watershed moment in US energy generation worth marking. The 236-point score reflects community endorsement of the framing as significant rather than incremental.
The Guardian reports that solar generated more electricity in the United States than coal for the first time in recorded history. The crossover, drawn from EIA generation data, isn't a single-month anomaly driven by a heatwave or a coal plant outage — it's a sustained inversion that's been telegraphed for years by capacity additions, and has now shown up in actual megawatt-hours delivered to the grid.
Coal's share of US generation peaked above 50% in 2007. Solar's share was a rounding error then — well under 0.1%. The two curves have been converging for a decade: coal retirements accelerated after 2015, gas filled most of the gap, and solar quietly compounded at roughly 25-30% annual capacity growth. The crossover happened not because anyone legislated it, but because utility-scale solar PPAs now clear at $20-35/MWh in sunny markets, well below the marginal cost of running an existing coal plant.
The numbers behind the headline matter. Coal generation in the US has fallen from roughly 2,000 TWh/year at peak to under 600 TWh today. Solar — utility-scale plus distributed rooftop — crossed that line on a trailing basis. Wind passed coal years ago. The remaining coal fleet is mostly old, paid off, and running as a hedge against gas price volatility. Each retirement announcement is a one-way door.
The instinct is to read this as an environmental story. It isn't, primarily. It's a cost-curve story with second-order effects that practitioners are already living with.
Solar's levelized cost of energy has fallen roughly 90% since 2010. The drivers were brutally unglamorous: Chinese polysilicon overcapacity, standardized module form factors, tracker maturation, and balance-of-system process learning. There was no breakthrough. There was a 15-year industrial grind that did to coal what flat-panel manufacturing did to CRTs. Coal didn't lose to climate policy; it lost to a learning curve that compounded faster than its operators believed possible.
The community reaction on Hacker News (236 points, lively comments) splits along predictable lines: one camp celebrating the milestone, another pointing out that solar's capacity factor is ~25% versus coal's 50%+, so nameplate parity arrived years before energy parity. Both are right, and the more interesting observation is what neither emphasizes: the grid's hard problem is no longer "how do we generate enough electrons" — it's "how do we orchestrate intermittent generation, storage, and flexible load in real time." That's a software problem.
The duck curve — net load dipping at midday when solar peaks, then ramping hard at sunset — has become the dominant operational reality in California, Texas, and increasingly the Southeast. CAISO routinely curtails gigawatts of solar at noon and pays gas peakers to ramp 13 GW in three hours at dusk. Batteries are absorbing more of that ramp every quarter: California's battery fleet went from ~250 MW in 2020 to over 13 GW in 2025. ERCOT is on a similar trajectory. Every battery added to the grid is, functionally, a compute problem: state-of-charge optimization, price arbitrage, frequency response, and degradation modeling all running on tight control loops.
This is where the story stops being abstract and starts touching code. Grid operators, utilities, IPPs, and a generation of well-funded startups (Tesla's Autobidder, Fluence's Mosaic, Gridmatic, Tyba, Modo Energy, dozens more) are racing to build the software layer for a grid where the marginal generator is variable and the marginal asset is a battery. The control-system stack — historically the domain of slow-moving SCADA vendors — is being rewritten by people who think in Kubernetes, Kafka, and reinforcement learning.
The AI angle compounds this. Data center load is projected to add 100-200 TWh of US demand by 2030, depending on whose forecast you trust. Hyperscalers are signing PPAs faster than the grid can build interconnections. The same week solar passed coal, multiple hyperscalers were reported to be co-locating gas turbines and batteries with new data centers because waiting in the interconnection queue is no longer competitive. The result: a private, semi-islanded grid layer is emerging alongside the public one, and most of its dispatch logic will be software.
If you write code adjacent to energy — utility software, EV charging, building controls, industrial IoT, data center ops — the assumptions you inherited are stale. Electricity is becoming time-variable in price and carbon intensity at hourly or sub-hourly granularity. Workloads that don't care when they run (batch training, video encoding, scientific compute, backups) have real economic incentive to follow the sun. Carbon-aware scheduling went from greenwashing to a line item on the AWS, GCP, and Azure roadmaps; expect region-and-time-aware APIs to keep multiplying.
For backend engineers more broadly: forecasting is the new bottleneck. Solar and wind output are predictable on minutes-to-hours horizons with the right models, and every percentage point of forecast accuracy translates directly into reduced reserves, lower curtailment, and tighter battery dispatch. The teams doing this well are running ensembles of numerical weather prediction, satellite imagery, and ML on commodity GPU infrastructure. If you've been looking for a domain where applied ML still has obvious unsolved problems and real economic stakes, grid forecasting is one of the cleanest.
For startup-watchers: the picks-and-shovels layer for grid software is being built right now, and most of it is still pre-Series-B. Battery optimization, virtual power plants, demand response, interconnection queue tooling, transmission planning, EV-grid integration. The pattern that worked for cloud infra in 2010-2015 — open-source primitives, API-first products, replace clipboard workflows with code — is repeating in energy with a 10-year lag.
Coal will keep falling. The interesting question isn't whether solar's lead grows — it will — but how fast storage and software catch up with the operational complexity that variable generation creates. The next milestone to watch isn't a generation crossover; it's the first full week in a major US ISO where batteries provide more peak-shoulder capacity than gas peakers. That's the moment the grid becomes structurally software-defined, and a lot of currently-boring code suddenly matters.
The growth of solar is astounding. I dug into data a while back and tried to do some visualizations of it, mainly for my own understanding:https://torkeldanielsson.se/solar-energy-forecasts/Solar is already by far our cheapest source of energy. As solar expands, the learning rate
+1 to the Guardian for mentioning their data source, but -1 for not linking to it.+2 for EMBER for having a data source AND being able to link to the parameters that show solar overtaking coal for the month in the US.https://ember-energy.org/data/electricity-data-explorer/?e
Question for those in the know... See lots of press about balcony solar in Germany, and California recently introduced a bill to allow it (I'm guessing other states already allow it; not sure if the CA bill has a chance of becoming law). But how far are we from a more plug and play home solar s
The US currently is at per capita GHG emissions approximately at the the same level as it was in 1910.https://ourworldindata.org/profile/co2/united-statesDespite not being in the paris treaty, the us needs only a 10-12% reduction to meet the paris accord requirements on sche
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
This is more from a lot of coal power plants being converted to gas over the past 20 years than solar overtaking the outputs of those power plants. Coal output shrinking, solar output rising, the lines have crossed.Coal is unpopular in all but a few areas where coal mining is still a part of the loc