From the EditorEvery number in this issue is correct. Several are also useless on their own, and that is the week's thread.
Our cover story sets the world's electrolyser manufacturing capacity — roughly 33 GW a year — against the fleet actually installed, about 3 GW in total. Neither figure is disputed. What decides the industry is the one between them: announced projects reach final investment decision at about 11%, and the factories were sized against the announcements. The shape recurs all week. Published orbital radiator requirements span a factor of seventeen, and almost none of that spread is physics — it is the operating temperature nobody declares. Three rotating-detonation programmes reported run duration and thrust; none reported pressure gain, the only reason the cycle exists. An issuer grew stablecoin circulation 19% and still pays 46 cents of every revenue dollar to one distributor. Fusion's capital cost is four-fifths not-fusion. Lunar delivery fell 90% per kilogram of lander capacity while the price per mission rose.
The published figure is rarely the deciding one, and the deciding one is usually the figure that cannot be stated without a baseline. Twelve features and eight briefings, aligned onto one plane.
— The Ecliptic, Editorial Desk
Cover Story · Power & Energy · Joule
Joule — The Factories Were Built Against the Wrong Number — August 18, 2026
The world can manufacture roughly 33 GW of electrolysers a year and has installed about 3 GW in total. The overbuild ratio is not a forecasting error about hydrogen — it is the arithmetic consequence of sizing factories against an announcement pipeline that converts to final investment decision at about 11%.
Global electrolyser manufacturing capacity is roughly 33 GW per year, inferred from the IEA's statement that China holds nearly 60% of it with more than 20 GW per year. Cumulative electrolyser capacity installed worldwide is about 3 GW. The entire installed fleet is therefore roughly 9% of what the world's factories can build in a single year, and the recent build rate implies utilization of about 5%. This report argues the overbuild is not a mistaken bet on hydrogen but a category error about which pipeline number to build against: announced low-emissions hydrogen production for 2030 stands at up to 37 Mtpa, of which more than 4 Mtpa is operational, under construction or past final investment decision — a conversion rate near 11% — while the announced figure itself fell from 49 Mtpa in a single year. Capacity sized against the announcement pipeline and demand arriving at the FID pipeline produces exactly the utilization now observed. The second finding is a feedback that closes the loop: at 5-10% utilization, fixed factory cost is amortised over almost no output, so electrolyser prices rise rather than descend a learning curve, which raises the capex that made projects unbankable in the first place. The third is that trade friction has consumed roughly 85% of the Chinese cost advantage at the point of delivery, leaving buyers outside China unable to reach the cheap curve and Western factories undercut anyway.
Read the Graphic Edition →
Flagship Star Power
Star Power — The Gap List Is an Engineering List — August 18, 2026
The DOE's June 2026 roadmap names 41 open problems on the path to fusion power. Seven of them are plasma physics. The facility that would close the largest of the rest is still five to ten years away — inside the window when the first plants are supposed to run.
Fusion's public problem list has been rewritten, and the new one is not about plasma. The U.S. Department of Energy's Fusion Science and Technology Roadmap, published June 2026, enumerates 41 named science-and-technology gaps across six challenge areas on the critical path to a fusion pilot plant. This report…
Flagship Star Power
Star Power — The Other 80% — July 31, 2026
The fusion island is one fifth of a fusion plant's capital cost. The rest is buildings, cryogenics, a steam cycle and paperwork — and it is where the programmes are actually failing.
If a fusion plant's capital cost is dominated by the fusion island, then fusion is a physics problem with a cost tail. If it is not, fusion is a construction problem with a physics prerequisite — and should be managed accordingly. This report opens the ARPA-E-derived fusion costing…
Flagship Faraday
Faraday — The First Cycle — July 30, 2026
Hard carbon's irreversible first-cycle loss — not its reversible capacity — now sets sodium-ion energy density: the pore-mouth mechanism, the precursor hierarchy, and the pre-sodiation route around it
Sodium-ion cells are now shipping from mass-production lines, but the chemistry's cell-level energy density is still set by a loss that happens once, in the first hour of a cell's life. This report asks what actually consumes the sodium that never comes back out of a hard-carbon anode,…
Flagship The Maple Grid
The Maple Grid — The Five Interties — July 30, 2026
Ottawa has named five interprovincial links and a target of doubling grid capacity by 2050. The binding constraint is cost allocation and market design, not steel — and none of the five closes Canada's most important east-west gap
In May 2026 Ottawa announced a National Electricity Strategy built on doubling Canada's grid capacity by 2050; in June it named five interprovincial transmission projects for priority treatment. This report asks whether the named links are the ones that matter, what they are worth, and what actually stands…
Briefing Parameter
Parameter — The Memory Tax Arrives — August 4, 2026
A depth follow-on: the contract-price shock has decelerated from +95% to +15% a quarter, but the level is now permanent — SK hynix is earning a 76% operating margin, Gartner has PC and smartphone shipments falling double digits, and the capacity answer does not land until 2028
This Briefing revisits the memory chokepoint six issues on, and covers six developments: (1) the contract-price shock has decelerated hard — from +93-98% QoQ in 1Q26 to +13-18% in 3Q26 — which markets are misreading as relief; (2) SK hynix printed a 76% operating margin on 79.3 trillion…
Briefing Parameter
Parameter — Briefing — July 29, 2026
A four-lab release wave in three weeks, the largest open-weight model ever published, Alphabet's capex guidance up $15B, CoWoS sold out into 2027, memory's price surge cooling, and the EU's August 2 deadline
This Briefing covers six developments in artificial intelligence and the market that prices it: (1) four frontier labs shipped inside three weeks, and the release cadence itself is now the story; (2) Moonshot published Kimi K3 at 2.8 trillion parameters — the largest open-weight model released to date;…
Flagship Engines of Tomorrow
Engines of Tomorrow — The Pressure Gain Nobody Publishes — August 20, 2026
Three independent rotating-detonation programmes cleared hardware milestones in 2026. None disclosed the one number that determines whether the cycle is worth building.
Rotating detonation engines had a hardware year in 2026. Astrobotic ran its Chakram engine for a record 300 seconds continuously at NASA Marshall, accumulating more than 470 seconds across firings at over 4,000 lbf per engine and reaching thermal steady state; L3Harris completed two self-funded test series at…
Flagship The Reality Gap
The Reality Gap — The Smaller Aperture Is the Heavier One — August 20, 2026
Orbital-compute radiator sizing spans 17× across published analyses. The spread is not disagreement about physics — it is an undeclared choice about how hot the silicon runs.
Builds on TAC-0012, which named radiator area rather than solar area as the binding aperture constraint on orbital data centres above roughly 100 kW and left the sizing open. This report closes it, and finds the question was mis-posed. Published radiator requirements for orbital compute span 345 m²…
Flagship Delta-V
Delta-V — The Lunar Delivery Cost Curve Is a Capacity Illusion
Task-order prices per kilogram of lander capacity have fallen ~90% since 2019 — but NASA's science manifest is not growing fast enough to fill the landers it is buying, and the fix has concentrated the risk
Does commercial competition actually lower the price of landing mass on the Moon? This report rebuilds the Commercial Lunar Payload Services (CLPS) price history from the awarded task orders and normalises each award against the published payload capacity of the lander that carries it. The result is a…
Flagship Chain Reaction
Chain Reaction — Forty-Six Cents of Every Revenue Dollar — August 20, 2026
Circle paid Coinbase $324.6m in Q2 against $701m of revenue and reserve income. Nineteen percent circulation growth bought no improvement in the split, and the terms just renewed to 2029.
Builds on SAT-0014, which opened the question of how much of a stablecoin issuer's reserve income is contractually paid away to the venues that hold the float. For USDC the answer is now measurable and larger than the sector generally assumes: Circle paid Coinbase $324.6m in the second…
Flagship Chain Reaction
Stablecoins After the GENIUS Act
The economics of a regulated, interest-banned, Treasury-backed dollar
Does federal legalization make stablecoins a bigger or merely a more durable business? This report builds a transparent reserve-income model for the US payment-stablecoin market after the GENIUS Act, triangulating issuer disclosures (Circle Q1 2026, Tether Q1 2026 BDO attestation), regulator filings, and on-chain data. The headline finding:…
The Power Bill · 2026-08-20The Power Bill — Briefing — August 20, 2026
1. **The procured-quantity series as the honest capacity metric.** Tracking MW procured rather than clearing price, since a binding cap makes price uninformative by construction (topic 1–2). 2. **Whether the six large-load filings converge…
Parameter · 2026-08-14Parameter — Briefing — August 14, 2026
1. A billion users, and most of them are talking 2. $240 million for 2,000 Blackwells puts a number on an accelerator-year 3. A 30-billion-parameter agent model under Apache 2.0, one day before DeepSeek…
Beyond Solar · 2026-08-12Beyond Solar — Briefing — August 12, 2026
A single corporate commitment is roughly **seven times** the combined value of the two most significant multi-day storage venture rounds of the period (*Beyond Solar estimate* from [1][2][3]). This report is general commentary published…
The Power Bill · 2026-08-11The Power Bill — The $6.2 Billion Nobody Has Built — August 11, 2026
1. $6.2 billion of the $6.5 billion is for data centres that do not exist yet 2. The execution gap is now a priced position, not a forecasting dispute 3. Cost causation gets a…
The Maple Grid · 2026-08-11The Maple Grid — The Capacity Arithmetic — August 11, 2026
1. Alberta's data-centre queue is 16 times the capacity it was allowed to allocate 2. The bring-your-own-generation rule turns a grid problem into a private-generation forecast 3. Wind West costs more per average kilowatt…
Faraday · 2026-08-11Faraday — The Pressure Is Sideways — August 11, 2026
1. Lateral pressure does at 15 MPa what stack pressure needs 100 MPa to do 2. The dangerous stiffness is in the middle, not at the top 3. LLZO's fracture toughness varies 3.4x with…
Chain Reaction · 2026-08-11Chain Reaction — Briefing — August 11, 2026
1. Riot signs 191 MW for $9.1 billion, and the bitcoin identity gets harder to argue 2. Bitdeer takes the model to Norway at $4.7 billion, with an extension worth $3.3 billion more 3.…
Joule · 2026-08-10Joule — Briefing — August 10, 2026
1. The US will add a record 86 GW this year, and half of it is solar 2. Lithium's six-month low sits inside a 94% annual gain — the supply response arrived 3. Pack…