From the EditorThis issue keeps returning to the same shape: a total that looks adequate until you ask who holds it.
Our cover story screens fusion's record funding year company by company rather than industry-wide, and the total dissolves. Two companies have raised more than the sector's own average stated requirement; the next thirty-nine share under one percent of it between them. The same pattern recurs across the issue in different units. One operator is party to ninety-four percent of the conjunctions in our orbital-debris feature — not because it is careless, but because it is nearly everyone else's counterparty by default. A gas-market feature finds flexibility priced as a single average when the marginal megawatt and the average megawatt are worth entirely different things. And our Frontier desks each found a technology's headline number holding up only against the wrong baseline — a detonation engine's efficiency, a droplet radiator's mass advantage — once somebody bothered to ask which comparison was actually being made.
None of these are stories about totals being wrong. They are stories about totals being true and uninformative at the same time, which is a harder thing to catch and a more useful one to publish.
Twelve features and eight briefings from across the Orrery network this week.
— The Ecliptic, Editorial Desk
Cover Story · Power & Energy · Star Power
Star Power — The Fundability Screen — September 2, 2026
Two of fifty-six fusion companies have raised more than the industry's own average stated requirement. The other fifty-four need between two and five more rounds the size of this year's largest, and thirty-nine of them share less than one percent of the capital.
TRP-0016 established the industry-level arithmetic — 56 fusion companies, an average stated requirement of USD 2.7bn each, USD 151bn implied against USD 14.24bn raised since 2021 — and left open which companies are actually within reach of their own number. This report answers that by screening the seventeen fusion companies that have disclosed raises above USD 100m against the industry's own average requirement. Method: cumulative disclosed capital per company, divided by the USD 2.7bn average, with the residual gap expressed in units of the survey year's four largest disclosed rounds (mean USD 574m). Finding: exactly two companies — Commonwealth Fusion Systems at roughly USD 3.94bn and Helion at roughly USD 3.2bn — have raised more than the average requirement, and no third company is within a single record-sized round of it. The median screened company needs between four and five such rounds. The seventeen disclosed companies account for approximately USD 14.12bn against the association's USD 14.24bn industry total, meaning the remaining thirty-nine surveyed companies share on the order of USD 125m in aggregate — under one percent of sector capital, about USD 3.2m each. The implication is that the field's funding problem is not a shortage but a distribution: at the record annual rate the sector-wide gap closes in roughly 31 years, at sustained 20% annual growth in about 10, and only at a sustained repeat of this year's 69% growth in about 5 — which is the sole scenario consistent with the 71% of companies expecting grid electricity in the 2030s.
Read the Graphic Edition →
Flagship Beyond Solar
Beyond Solar — What the 60–80 Metre Band Is Actually Worth — September 1, 2026
Converting the crossover shift into addressable capacity puts about 492 GW of US seabed and roughly 5,600 GW globally in the contested band — and an April 2026 analysis argues the boundary may now move back the other way
HEL-0021 established that floating offshore wind's expected crossover with fixed-bottom moved from roughly 80 metres of water depth in 2019 to roughly 60 metres by 2035, and argued that this 20-metre shift — not floating's own cost decline — is the measure of the sector's progress. This report…
Flagship Joule
Joule — The Cost That Does Not Learn — August 31, 2026
Moving the same Chinese electrolyser from China to a Western site adds about USD 1,075/kW. Buying the cheapest equipment on earth instead of Western equipment saves about USD 325/kW. The location premium is three times the equipment premium, and it sits in the one cost block with the lowest assumed learning rate.
TIB-0054 bounded the prize from fixing electrolyser factory utilization at roughly USD 350/kW against an installed cost near USD 2,000/kW, and left the remaining USD 1,100/kW between manufactured and installed cost unexamined. This report opens that block. Two agencies reach the same decomposition by independent methods: the IEA…
Flagship The Gas Gauge
The Gas Gauge — The Flexibility Ledger — August 29, 2026
Classifying all 119 Bcf/d of US natural gas demand by the mechanism through which it can physically be reduced — and finding that the entire addressable flexibility, 16 to 29 Bcf/d, is smaller than a single winter's seasonal swing
When the US gas balance tightens, how much demand can physically be turned down, by what mechanism, and at what price? This report builds a flexibility ledger: every block of US natural gas demand — 119.0 Bcf/d of domestic consumption plus exports — classified by the mechanism through…
Flagship The Power Bill
The Power Bill — The Slot Is the Asset — August 28, 2026
GE Vernova holds 100 GW under contract of which 56 GW is not an order at all but a reservation for a delivery position, gas turbine prices are running 10–20 points higher per kilowatt than three months earlier, and the scarce good in the power sector has stopped being the machine and become the place in the queue.
What is actually scarce in the gas turbine market, and what does it cost? GE Vernova's 100 GW under contract splits into 44 GW of firm orders and 56 GW of slot reservation agreements — meaning the majority of the largest backlog in the industry is not an…
Flagship Parameter
Parameter — The Custom-Silicon Economics — July 23, 2026
Hyperscalers are building their own accelerators to escape the Nvidia margin — but the money they save lands mostly on Broadcom, TSMC and the memory makers, and the club stays closed
In 2026 custom AI silicon crossed from science-project to structural force: application-specific accelerators are on track to make up roughly 27.8% of AI-server accelerator shipments this year and to surpass merchant GPUs in unit shipments by 2027-2028 [5]. Every hyperscaler now runs its own program — Google's TPU…
Flagship Parameter
Parameter — The Tokenomics of Inference — July 1, 2026
Why the price of a token is collapsing, the AI bill is exploding, and the margin gap between labs is the story that survives
The unit price of frontier-quality intelligence is falling faster than almost any input in the history of computing — between 9x and 900x per year to hold a fixed benchmark score, with a post-2024 median near 200x per year [1]. Yet enterprise AI bills are rising, not falling;…
Flagship Engines of Tomorrow
Engines of Tomorrow — The Number That Would Make Three RDE Programmes Comparable — September 3, 2026
An academic paper this year finally measured what three hardware programmes wouldn't disclose — combustion efficiency of 50-70%, well short of the ideal cycle. It isn't the number MACH-0015 asked for, and that gap is itself the finding.
MACH-0015 asked what minimum disclosure would make two rotating-detonation engine programmes comparable, and noted that the programme-level record — Astrobotic, L3Harris, Pratt & Whitney — contained no pressure-gain figure at all, and predicted the answer would have to come from the academic literature instead. It has. A January…
Flagship The Reality Gap
The Reality Gap — The Droplet Radiator Beats the Worst Case, Not the Best One — September 3, 2026
A Georgia Tech magnetic droplet radiator hit 500 W/kg in April 2026 — a headline 15-30x improvement. Against TAC-0015's own published range for conventional radiators, it beats only the coldest, most Earth-IR-loaded case. Against the hottest ideal-environment case, it is roughly parity.
Builds on TAC-0015 [4], which found conventional orbital-compute radiator sizing spans a 17x range because operators don't disclose operating temperature, and flagged liquid droplet radiators as the one technology that could break the resulting areal-mass floor entirely — naming a November 2025 demonstration and a 450 W/kg figure…
Flagship Delta-V
Delta-V — One Operator Is Party to Ninety-Four Percent of Conjunctions — August 31, 2026
SpaceX's registration statement discloses 8,900 satellites, roughly 75% of every active maneuverable object in orbit, and over 1,000 automated collision avoidance maneuvers a day. At that share, space traffic coordination has quietly become one company's internal scheduling problem.
SpaceX's Form S-1 registration statement, submitted to the SEC in March 2026 and now public, contains an operational disclosure that has not been read for what it implies about space traffic management: as of December 31, 2025 the company operated over 8,900 Starlink satellites accounting for approximately 75%…
Flagship Chain Reaction
Chain Reaction — The Bid That Is Strongest When It Reverses — August 31, 2026
A $3.5bn stablecoin inflow moves the 3-month bill about 5 basis points at its trough. The same research finds the effect two to three times larger under Treasury-market stress — which is the only condition in which a stablecoin run actually happens.
SAT-0019 flagged stablecoin float as a possible macro instrument — whether a USD 300bn supply backed by short-duration Treasuries now behaves as a marginal buyer in the bill market. A BIS working paper revised in June 2026 answers it with daily data from January 2021 to March 2026…
Flagship Chain Reaction
The Wrapper Is the Product: Inside the $2.2 Billion Tokenized-Equity Market
Tokenized stocks roughly doubled since March and 6×'d their on-chain volume — but the tokens confer no shareholder rights, run almost entirely offshore, and settle 95%+ on one chain. What the market is actually buying, and who captures the rent.
Does putting a stock on a blockchain create a better claim on a company — or merely a more tradable one? This report examines the tokenized-equity market as it crossed a record ~$2.2B aggregate market cap on 21 July 2026, roughly doubling since March, while on-chain volume 6×'d:…
The Power Bill · 2026-09-03The Power Bill — Briefing — September 3, 2026
1. **The RBP as a second, parallel capacity price.** Once bids clear in the September 30 window, compare the implied $/MW-day against the $325 cap that produced the shortfall it exists to cover, and…
Chain Reaction · 2026-09-03Chain Reaction — Briefing — September 3, 2026
1. **The September 15 vote as a dated resolution point.** SAT-0019 found the market pricing a probability revision on a bill whose only observable data point was a failure. A second failure, on a…
Joule · 2026-09-02Joule — Briefing — September 2, 2026
1. **LNG export growth is now the marginal claim on US gas production, not power burn.** Exports averaged 17.4 Bcf/d in the first half and are forecast at 18.7 Bcf/d by early 2027, against…
The Maple Grid · 2026-09-01The Maple Grid — Briefing — September 1, 2026
1. Ontario's demand forecast moved, and the data-centre term moved most 2. Alberta's market reform is law, minus two of the three markets it started with 3. Hydro-Québec is buying wind on criteria that…
The Gas Gauge · 2026-09-01The Gas Gauge — Briefing — September 1, 2026
1. The record storage forecast is an average-build forecast 2. EIA cut third-quarter Henry Hub by 50 cents in a month 3. Supply is forecast to grow four times faster than LNG can take…
Faraday · 2026-09-01Faraday — Briefing — September 1, 2026
Builds on FAR-0012 — that report built a bottom-up mass budget for lithium–sulfur and concluded that the electrolyte-to-sulfur ratio, not sulfur loading, sets cell-level specific energy; this briefing follows the ratio out of the…
The Power Bill · 2026-08-31The Power Bill — Briefing — August 31, 2026
Builds on CAL-0022 — that Briefing left an open thread asking "whether the six RTO large-load filings converge on one cost-allocation model or fragment into six, and what a fragmented outcome does to siting…
Parameter · 2026-08-31Parameter — Briefing — August 31, 2026
Builds on PAR-0009 — "The Inference Cost Collapse" established that inference prices were falling across a 9× to 900× range depending on capability tier, and left open the question of what that does to…