Energy Weekly Review 2026-08-21
Week In Review
The single most legible number of the week was a round one: the world crossed three terawatts of installed solar capacity. Solar crosses 3 TW threshold is the kind of milestone that is easy to file under “nice chart” and hard to internalize, because the intervals between terawatts keep collapsing — roughly a decade for the first, under three years for the second, under two for the third. What makes the milestone interesting is not the generation itself but the constraint it exposes. Photovoltaic panels are now the cheap part. Everything that moves, stores, shapes, and firms their output is the expensive part, and most of this week’s other news was about that second category.
Three items read as direct answers to the same bottleneck from different directions. The world’s most efficient solar panel goes on sale later this year attacks it by extracting more energy per unit of land, interconnection, and racking — the costs that do not fall when module prices do. Startup Heron Power to build 40GW power conversion manufacturing facility in California attacks it at the interconnection point itself, replacing the copper-and-steel transformer with solid-state power electronics that a semiconductor supply chain can actually build at pace. And Ausgrid’s PLUS Grid Storage breaks ground on 400MWh Steel River East BESS in Australia attacks it in time rather than space, on the site of a former industrial estate in a region where coal plants are retiring. Solar’s growth curve has effectively become a demand signal for three separate industries.
Nuclear’s week was quieter and more structural, which is arguably a sign of health. Kaleidos microreactor begins journey to Idaho moved a commercial-design microreactor onto a national laboratory test bed for a fueled campaign — a hardware milestone rather than a paper one. Urenco breaks ground for new enrichment capacity addressed the least glamorous and most binding constraint on all of it, fuel. Spain’s decision to keep its largest power station running, covered in Spain’s Almaraz operating license extended into 2030, showed a government revisiting a phase-out schedule under pressure from price volatility and supply security rather than from any change in nuclear technology. And Atomic Canyon launches virtual assistant for US nuclear fleet targeted the paperwork layer — the regulatory and operating-experience corpus that makes nuclear projects slow in ways physics does not require.
Fusion contributed two results that share a theme: the field’s remaining problems are increasingly materials problems and control problems rather than plasma-physics problems. New measurement of diamond phase change could mean increased ICF energy gain concerns how the capsule holding the fuel behaves under compression, a question about a solid rather than a plasma. New ML framework predicts shifts between shots at DIII-D concerns whether a machine can keep an accurate model of itself as its own hardware drifts. Both are the sort of unglamorous engineering work that separates a demonstration from a power plant, and both — like Atomic Canyon’s assistant — involve turning accumulated experimental and operational records into something a machine can reason over.
Items
Global Solar Capacity Passes Three Terawatts
BloombergNEF reported this week that cumulative installed solar photovoltaic capacity worldwide has passed 3 terawatts. The interesting feature of that number is not its size but the shape of the curve behind it. It took roughly ten years, from 2012 to 2022, for global solar to climb from 100 gigawatts to the first terawatt. The second terawatt arrived in under three years. The third took less than two.
The composition of that growth has shifted markedly across the three tranches. Developed economies dominated the first terawatt. China drove most of the second and third. What BNEF’s analysis highlights, though, is the widening base underneath: 74 countries had at least a gigawatt of installed solar by 2025, up from 42 in 2020, with notably fast expansion in markets including Pakistan, Nigeria, the Philippines, Cuba, and Lebanon. Solar is no longer a technology that requires a wealthy state’s industrial policy to deploy — it is increasingly something that spreads because it is the cheapest available option, particularly where grid service is unreliable and the alternative is a diesel generator.
BNEF projects global solar capacity will exceed 9 terawatts by 2036, with developing nations expected to hold more than a quarter of the total and wealthy nations’ share falling to roughly a fifth. That forecast comes with a caveat the analysis makes explicit: fully utilizing additional solar capacity will become increasingly difficult without matching investment in storage and grid infrastructure.
That caveat is the connective tissue for much of the rest of this week’s news. A terawatt of panels that cannot be absorbed at midday is worth considerably less than a terawatt that can be time-shifted or moved. The marginal value of the next solar installation now depends less on the panel and more on what sits between the panel and the load.
Source: pv magazine
Tandem PV Books Roughly $1 Billion in Letters of Intent for Perovskite-Silicon Modules
Tandem PV, a US perovskite manufacturer, told pv magazine it has secured close to $1 billion in letters of intent for its next-generation modules — roughly 2.5 gigawatts of signed capacity at the $0.40 per watt average selling price the company cites. The company shipped its first production-quality panel two weeks before the article and expects revenue-generating sales later this year, with paid pilot deployments at independent power producers to follow.
The technical approach differs from the perovskite-silicon tandems that most large manufacturers are pursuing. Rather than a two-terminal monolithic stack, where the perovskite layer is grown directly onto the silicon cell and the two must be current-matched, Tandem PV uses a four-terminal mechanically stacked design: a perovskite-on-glass layer sits above a conventional silicon cell, each operating on its own electrical circuit. The arrangement sacrifices some optical elegance but decouples the two subcells, which simplifies both manufacturing and the problem of matching their behavior across temperature and spectrum.
The company reports a demonstrated module efficiency of 30.4% and projects roughly two percentage points of improvement per year, reaching 37% by 2030. Those figures should be read as company claims about a development trajectory rather than as certified records, but the underlying physics has headroom: a two-layer perovskite architecture has a theoretical ceiling near 45%, and a three-layer version near 50%, against the roughly 29% limit for single-junction silicon.
Tandem PV targets gigawatt-scale production in 2028 at an estimated $100 million of capital expenditure per gigawatt, using off-the-shelf semiconductor equipment — a deliberately unexotic manufacturing story meant to make the capital raise tractable. Chief executive Scott Wharton framed the competitive stakes bluntly: “It’s an order of magnitude change. If there’s a 30 to 32% panel… you’re kind of in trouble.” The claim to watch is bankability, which the company expects only after six to twelve months of field data. Perovskite durability, not perovskite efficiency, has always been the hard part.
Source: pv magazine
Radiant’s Kaleidos Microreactor Ships to Idaho National Laboratory
Radiant’s Kaleidos microreactor — a 1 MWe transportable unit designed to be moved by land, sea, or air — began a journey of more than 1,000 miles from California to Idaho National Laboratory on August 12, arriving for testing at the DOME facility, the laboratory’s Demonstration of Microreactor Experiments test bed.
The detail that matters is that this is not a scaled-down test article. Radiant says the unit is the same design and scale intended for commercial production and customer delivery, fueled with TRISO coated particle fuel fabricated by Standard Nuclear. TRISO encapsulates fuel kernels in layers of carbon and silicon carbide that retain fission products at high temperature, which is the basis for the safety case that lets a reactor this small be sited without a large exclusion zone.
The planned campaign has five phases: zero-power criticality testing to confirm the reactor behaves as modeled, then heat and power generation, and finally a demonstration of 150 continuous hours of operation without operator assistance. That last item is the commercially decisive one. A microreactor whose economics depend on a full-time licensed operating crew at a remote site is not a microreactor in any useful sense; autonomous operation is what makes the form factor pay.
Radiant is targeting completion of the testing campaign in the third quarter of this year, with Nuclear Regulatory Commission approval for its R-50 production facility in Oak Ridge, Tennessee expected by the fourth quarter, and a first commercial delivery to Buckley Space Force Base planned for 2028. Remote military installations and off-grid industrial sites are the natural early market — places where the incumbent alternative is trucked diesel at a cost per kilowatt-hour that makes even first-of-a-kind nuclear look reasonable.
Source: World Nuclear News
Urenco Breaks Ground on Major Enrichment Expansion in New Mexico
Urenco began construction this week on an expansion of its National Enrichment Facility in Eunice, New Mexico, adding 24 centrifuge cascades and 2.1 million separative work units of annual capacity. The company puts total investment at the site above $8 billion. Initial production from the new cascades is scheduled for 2032, with the remainder installed through 2036, and the project is expected to support 300 to 600 jobs during peak construction and about 70 permanent operations positions.
Separative work units are the standard unit of enrichment effort — a measure of how much isotope-separation work is required to turn natural uranium into fuel-grade material, independent of how much uranium goes in. Adding 2.1 million SWU per year is a substantial increment to Western capacity, and the long build-out schedule reflects the reality that centrifuge cascades are precision manufacturing rather than construction.
The strategic context is straightforward. Every advanced reactor program in the news — the microreactors, the molten salt designs, the sodium fast reactors — ultimately depends on a fuel supply chain that has been thin outside Russia for two decades. Chief executive Boris Schucht framed the expansion as supporting the US Department of Energy’s goals to accelerate deployment of new domestic capacity and strengthen the US nuclear fuel supply chain.
Timing is the honest caveat. Fuel arriving in 2032 does not help a reactor that needs a first core in 2028, and the gap between announced reactor schedules and announced enrichment schedules remains one of the sector’s more persistent arithmetic problems. What the groundbreaking does establish is that the capital is now moving into the least visible part of the stack, which is generally where shortages bite first.
Source: World Nuclear News
Spain Extends the Almaraz Nuclear Plant’s License to 2030
Spain’s Ministry for the Ecological Transition and the Demographic Challenge has renewed the operating license for the Almaraz nuclear power plant by three years, extending it through June 8, 2030. The decision was issued on August 12 and posted publicly on August 14. Almaraz had been scheduled to begin shutting down in 2027 under the country’s phased nuclear exit.
Almaraz is not a marginal facility. Its two units total 2,017 MWe, making it Spain’s largest power station of any type, and it supplies more than 7% of the country’s electricity — the ministry’s figures put that at the equivalent of powering over four million homes. Removing it from the system in 2027 would have meant replacing that output, and the associated grid inertia and firm capacity, on a short timeline.
The government’s stated reasoning is notable for what it does not invoke. The extension is justified not by a reassessment of nuclear technology but by energy security: geopolitical tension in the Middle East, elevated and volatile fuel prices, and the aftermath of the major Iberian blackout of last year. The ministry’s language was that “this sudden context of high prices and greater volatility and uncertainty increases the exposure of the country” to fuel-dependence risk. That is a system-operations argument, and it is the same argument being made in several European capitals.
Importantly, this is a deferral rather than a reversal. Spain’s long-term policy of retiring the entire fleet by 2035 remains formally in place. What the Almaraz decision demonstrates is that phase-out schedules set in calmer conditions are being renegotiated against present-day grid realities — and that a functioning 2 GW low-carbon plant is a difficult thing to switch off on principle when prices are high and the interconnectors are stressed.
Source: ANS Nuclear Newswire
Atomic Canyon Launches an AI Assistant Trained on the Nuclear Regulatory Corpus
Atomic Canyon, founded by Trey Lauderdale, released NIVA — the Nuclear Industry Virtual Assistant — a platform intended to let nuclear engineers and operators query the industry’s technical, regulatory, and operational record in plain language. It launched on August 20 with Constellation Energy, which operates 21 reactors across 12 plants, among the early adopters, alongside the Institute of Nuclear Power Operations, EPRI, and the Nuclear Energy Institute.
The product ships with two assistants and a third in development. The Knowledge Assistant answers questions against industry content including NRC Regulatory Guides, NUREGs, and NEI guidance. The Operating Experience Assistant retrieves prior operational incidents and lessons by meaning rather than keyword — the distinction matters, because the useful precedent for a given plant problem is frequently described in vocabulary the searcher would never think to type. A Troubleshooting Assistant is in progress.
Underneath sits Atomic Canyon’s Neutron AI workbench and its FERMI models, which the company trained on more than 53 million pages of NRC data in collaboration with Oak Ridge National Laboratory using the Frontier supercomputer. Funding has come from NVIDIA, Plug and Play Ventures, and former Vanguard chief executive Tim Buckley.
The case for this being more than a convenience tool rests on where nuclear projects actually lose time. Licensing, documentation, and operating-experience review consume enormous engineering effort, and much of that effort is spent locating rather than reasoning — finding the applicable guidance, the analogous event, the prior submission that answered a similar question. Compressing retrieval does not change any safety requirement; it changes how many engineer-hours the requirement costs. Given that regulatory throughput, not reactor physics, is the binding constraint on much of the current buildout, that is a meaningful place to apply the technology.
Source: World Nuclear News
Measuring How Diamond Melts Could Triple Inertial Fusion Energy Gain
Researchers at Lawrence Livermore National Laboratory, led by Marius Millot, published measurements in Nature Physics on August 14 resolving how diamond melts under the extreme pressures and temperatures of laser-driven compression. The experiments were conducted at the University of Rochester’s Laboratory for Laser Energetics, and tracked atomic structure, temperature, density, and optical reflectivity through the phase transition.
The relevance to fusion is direct. In inertial confinement fusion, the fuel is held in a small capsule — frequently made of diamond — that is compressed by laser-driven ablation until the fuel reaches ignition conditions. How that capsule material melts governs how evenly pressure is transmitted inward. Uneven melting introduces asymmetries that hydrodynamic instabilities then amplify, and asymmetry is the classic reason an implosion fails to ignite.
The measurements settled a discrepancy roughly twenty years old between theoretical predictions and earlier experimental results, and confirmed something counterintuitive along the way: solid diamond floats in molten carbon, much as ice floats in liquid water. Most materials are denser as solids than as liquids; carbon, like water, is an exception, and getting that relationship right matters for modeling what happens inside the capsule during the first moments of compression.
The practical payoff the team identifies is a change to the shock timing. A more accurate melting curve permits a slower first shock, which leaves the fuel more compressible, which in turn allows more of it to burn before the assembly flies apart. The laboratory’s estimate is that this could potentially triple energy gain. That is a modeling projection rather than a demonstrated shot result, and inertial fusion has a long history of the gap between the two being wide — but it is a rare case of a fundamental materials measurement translating into a concrete lever on a machine’s output.
Source: ANS Nuclear Newswire
An Adaptive Machine Learning Framework Cuts DIII-D Prediction Error by 80%
A team led by Kishan Rajput at Thomas Jefferson National Accelerator Facility, working with the University of Houston, General Atomics, and Pacific Northwest National Laboratory, published a machine learning framework in Machine Learning with Applications that tracks how the DIII-D tokamak’s hardware behavior shifts between experimental shots. Compared with static machine learning models, the approach reduced prediction error by 80%, with uncertainty quantification contributing a further 10% reduction.
The problem being solved is one that shows up across experimental physics but is acute in fusion: the data are non-stationary. A tokamak does not behave identically from one shot to the next. Components age, calibrations drift, configurations change between campaigns, and the machine that produced last month’s training data is not quite the machine running today. A model trained once and frozen degrades steadily against a moving target, and it degrades silently.
The framework’s response is online learning — continuously updating as new shot data arrives, so the model tracks the machine rather than a historical snapshot of it. The uncertainty quantification component is arguably as important as the raw accuracy gain, because it tells operators when the model’s prediction should not be trusted. In a control context, a confidently wrong prediction is considerably worse than an appropriately hedged one.
This is infrastructure rather than headline physics, but it points at a real requirement. A fusion power plant will need to run reliably for long stretches while its own components change underneath it, which means its control systems must adapt without a human retraining them between every campaign. The DIII-D result is a step toward machines that maintain an accurate self-model as they age — a prerequisite that gets far less attention than plasma temperature records.
Source: ANS Nuclear Newswire
Heron Power Picks Morgan Hill for a Solid-State Transformer Factory
Heron Power announced on August 18 that it has selected Morgan Hill, California as the site of Heron Factory One, its first US power electronics manufacturing plant. The company plans to invest over $100 million, create more than 600 jobs, and produce up to 10,000 units per year of its Heron Link product. Site preparation begins immediately, with commercial production slated for late 2027.
The Heron Link combines a bidirectional inverter with a solid-state transformer, allowing solar, storage, and data center developers to connect directly to medium-voltage transmission without a separate conventional transformer. A traditional transformer is a large assembly of copper windings and grain-oriented electrical steel operating at grid frequency — a mature, reliable, and profoundly supply-constrained piece of equipment, with global lead times now measured in years. A solid-state transformer instead uses semiconductor switching at high frequency, which shrinks the magnetics dramatically and shifts manufacturing from heavy electrical assembly toward power semiconductor production.
The consequences go beyond lead time. Because the conversion is actively controlled rather than passive, a solid-state device can regulate voltage, respond to grid disturbances, and provide grid-forming behavior — capabilities a passive transformer cannot offer at any price. As inverter-based resources displace synchronous generators, that active control becomes more valuable, not less.
The reason this belongs in the same week as a three-terawatt solar milestone is that transformer availability is now a live gating factor on interconnection queues in the United States, Europe, and Australia. Projects with signed offtake and financed panels sit idle waiting on grid equipment. Heron’s factory will not resolve that shortage on its own, and late 2027 is a while away — but it represents capital moving toward the specific component that has become the choke point, and doing so via a supply chain that can scale faster than one built on electrical steel.
Source: Energy-Storage.news
Construction Starts on a 400 MWh Battery at a Retired Industrial Site in Newcastle
Construction began this week on Steel River East, a 200 MW / 400 MWh lithium-ion battery energy storage system in Newcastle, New South Wales. The project sits in the Steel River Industrial Estate and will connect to Ausgrid’s 132 kV Mayfield West zone substation. SCEE Electrical is the delivery partner, and commercial operation is expected in late 2027. The NSW Independent Planning Commission approved the project in March.
The corporate structure is the genuinely novel element. The developer, PLUS Grid Storage, is a separate commercial entity within the Ausgrid Group, developing, owning, and operating distribution-connected batteries independently of Ausgrid’s regulated network business. Ausgrid is the largest electricity distributor on Australia’s east coast. The arrangement lets a network operator’s knowledge of where its own grid is constrained inform where merchant batteries get built, while keeping the merchant risk outside the regulated ring-fence — a structure regulators elsewhere have generally been cautious about, since a distributor that also trades in the market has obvious conflicts to manage.
The siting is deliberate. Steel River East sits within the Hunter and Central Coast Renewable Energy Zone, a region undergoing substantial change as coal-fired stations retire and renewable capacity replaces them. Placing storage at a distribution substation rather than at transmission level puts the capacity closer to load and can defer network upgrades — value that a transmission-connected battery of identical size does not capture.
The project is one component of a broader platform. In June, the NSW Energy Security Corporation committed AU$100 million (about US$70 million) to a 650 MW portfolio of PLUS Grid Storage projects, which includes Steel River East and a matching 200 MW system at Homebush in western Sydney. Four hours of storage duration reflects where the Australian market currently pays — long enough to move solar output through the evening peak, which remains the country’s defining grid problem.
Source: Energy-Storage.news