Hydrogen Economies

A hydrogen economy means using hydrogen as a general-purpose energy carrier: made from electricity by splitting water, or from natural gas by steam reforming, then compressed or liquefied, moved, stored, and burned or run through a fuel cell wherever energy is wanted. The promotional framing is universality — transport, heating, industry and long-duration storage, all served by one molecule. That framing is what this brief tests.

The vocabulary matters more here than in most subjects, because almost every disputed number in the field turns out on inspection to be a number about a different substance. Grey hydrogen is made from unabated fossil feedstock, overwhelmingly by steam methane reforming, and is essentially all of the hydrogen the world uses today. Blue is the same route with carbon capture attached. Green is electrolytic hydrogen made with renewable electricity, and it comes in at least four cell chemistries — alkaline, proton-exchange membrane, anion-exchange membrane and solid oxide — whose efficiencies, capital costs and tolerance of variable power differ enough that quoting any of them without saying which is a category error. Natural or geologic hydrogen is not made at all; it is drilled. A cost per kilogram with no route, no electricity source, no cell chemistry and no year attached carries no information, and this brief states all four wherever it gives one.

The organising distinction is announced against built. Hydrogen has more announced capacity relative to operating capacity than any other energy technology in this corpus, and the four rungs of the ladder — announced, final investment decision taken, under construction, operating — are not degrees of the same thing. They are separated by attrition rates that are themselves the most informative measurement available in the field. When the promotional literature says a number of gigawatts, it is almost always quoting the top rung; when the deployment record says a number, it is quoting the bottom one. The gap between them, and the rate at which projects fall out between rungs, is the substance of this brief.

Where this brief stops. Hydrogen for steelmaking is covered by the Carbon Problem programme's zero-carbon industry spoke, which reaches it from the emissions side; this brief treats steel only as a demand-side test of what hydrogen can be sold for. The marine fuel orderbook is carried in detail by future ports and shipping and is used here only as evidence about hydrogen-derived fuels. The competing storage technology is advanced battery technologies. What is left for this brief is the economy-wide claim: whether hydrogen becomes an energy carrier at all, and in which applications.

The brief is organised around the four legs of the universality claim — industry, storage, heating and light transport — because the evidence now separates them sharply, and the separation is not a matter of maturity. One leg holds, and holds for a reason that has nothing to do with hydrogen being a good energy carrier. Three do not, and each fails for its own measurable reason.

Established The single most important number in this subject is an attrition rate, and it is 4%. In the 2024 edition of the intergovernmental review series, of roughly 520 GW of electrolyser capacity announced for 2030, only 4% had reached a final investment decision or entered construction. Every other figure in the hydrogen literature should be read against that ratio, because it is the conversion rate between the rung the sector talks about and the rung it builds on. Installed electrolyser capacity at the end of that same year was 1.4 GW.

Established The announced rung has since fallen by nearly half, measured by the same agency using the same method three years running. Announced low-emissions hydrogen production for 2030 went from 49 Mtpa in the 2024 review to 37 Mtpa in 2025 to 27 Mtpa in 2026 — a 45% collapse in announced ambition in two years, with the agency attributing the most recent drop as ten megatonnes down on the previous year. Announcements are the cheapest thing in this industry to produce, which is exactly why a fall in announcements is significant: it means the sector has stopped being willing to bear even the cost of a press release.

Established The committed rung barely moved while the announced rung halved. Capacity with an investment decision taken, under construction or operating stands at about 4.3 Mtpa for 2030, described in the 2026 review as “just over 4 Mt”, against 4.2 Mtpa — 9% of the pipeline — a year earlier. The 2026 review also states, in a different chapter and on a different definitional basket, that committed projects declined from 10 Mt to just above 6 Mt. This brief reports the disagreement rather than reconciling it; the two figures are not comparable and the report does not say which basket governs.

Established Installed capacity is real and it is small. Electrolysers in operation went 1.4 GW at end-2023, 2 GW at end-2024, above 4 GW at end-2025. That is genuine growth, roughly a tripling in two years, and it is the honest thing to set against the collapsing pipeline. It is also, on the same three-year view, an industry whose entire operating fleet is smaller than a single large gas-fired power station.

Established And the commitment rate is now falling rather than accelerating, which is the change that dates this brief. New investment decisions in 2025 came in below 0.8 Mtpa, against roughly 1 Mtpa in each of the two preceding years. The agency's warning is explicit: more than 100 GW of announced electrolysis “could lose any chance of being in operation by 2030 if investment decisions are not taken before the end of 2027”, and separately that 22 Mt of announced production is at risk without a decision by early 2027, two-thirds of it in Europe, North America and Latin America. The first statement is established as the agency's position; the outcome is frontier.

Established The decisive market test involved signed money, and the bidders declined it. In the second European Hydrogen Bank auction, developers holding awards covering 2.3 GW walked away from 1.9 GW of it — 83% of awarded capacity — rather than sign grant agreements and post completion guarantees. The final signed outcome was 380 MW and €271 million against a €1.2 billion budget, under 22% of the budget deployed, with 6 of 15 original winners signing. The walkaways had bid subsidies of €0.20–0.60 per kilogram; reserve-list replacements needed €0.64–1.22 per kilogram. This is not a policy-uncertainty story and it is not a permitting story. These were bidders refusing subsidy they had already won, at prices they themselves had named.

Handwave Reading the third auction's awarded capacity as delivered capacity. Round three, resolved around May 2026, selected nine projects, over €1 billion, 1.1 GW, at subsidies of €0.44–3.49 per kilogram. On the second auction's record, awarded capacity is a weak predictor of signed capacity and a weaker one of built capacity. The spread of the third round's clearing prices — nearly eightfold between the cheapest and dearest winning bid — is itself evidence that the sector has no settled cost.

Established Attrition is not confined to production; it is uniform across every asset class the pipeline contains. Of more than 40,000 km of hydrogen pipelines announced to 2035, 9% is operational or committed. Of 11 TWh of announced underground storage, just over 7% is at investment decision or under construction. In Europe, a commercial analyst counted 23 projects and 29.2 GW cancelled or stalled by the end of 2024, 20.3% of the European pipeline, with high costs implicated in 32% of cancelled projects and 40% of cancelled capacity and failure to obtain funding in a further 18% and 29%. The same analyst's projection that only 17% of the European pipeline materialises by 2030 is a forecast and is flagged frontier; the 20.3% already measured is not.

Established Costs moved the wrong way, and the reason is not the one usually offered. A commercial price survey recorded a median 57% rise in Western installed capital cost for electrolyser systems in 2024, against the same house's own prior forecast of 8–10% annual declines, with Western systems around $2,500 per kilowatt and Chinese around $600. The intergovernmental figures for installed cost agree in shape: $2,000–2,600/kW outside China against $600–1,200/kW inside it, and $1,500–2,400/kW for Chinese equipment installed abroad once transport and tariffs are counted. The mechanism is the part that is routinely missed: the stack is only 15–20% of investment cost and more than half the total is engineering, procurement, construction and contingency, which is why installing Chinese equipment in Europe cuts total capital expenditure by only about 20%. Any cost projection built on stack learning curves is extrapolating from a fifth of the cost.

Established The demand side has a ceiling, and it is below the supply side's floor. The intergovernmental cost-acceptability analysis is the cleanest statement of the whole problem: absent policy support, the maximum acceptable hydrogen cost is below USD 2 per kilogram for most sector and region combinations. Refining and ammonia tolerate the most, because energy is a large share of product cost. Steel requires a negative cost — that is, an incentive — to reach parity. This is a willingness-to-pay ceiling, not a supply-side cost problem, and no learning curve reaches a ceiling that sits underneath it.

Established The honest positive case is real, narrow, and not what the promotional framing claims. Global hydrogen demand surpassed 100 Mt in 2025, consuming 290 billion cubic metres of natural gas and 90 Mtce of coal and emitting on the order of 1,300 Mt of CO₂-equivalent a year with, in the agency's phrase, “no progress in reducing them.” That demand is almost exclusively refining, ammonia, methanol and fossil-based direct reduced iron; new applications are under 1% of global demand. Low-emissions hydrogen reached about 1 Mt in 2025, up 20%, still under 1% of the total. And the committed money is going exactly there: 2.5 Mt of committed 2030 consumption, 60% of all committed production, is refining and industry, with about 80% of “strong potential” production targeting chemicals. This is substitution into existing demand for hydrogen-as-feedstock. It is not hydrogen becoming an energy carrier, and conflating the two is how a narrow, defensible industrial decarbonisation case gets sold as a universal one.

Established Two countries out of sixty-six are on track. Of 66 national hydrogen strategies, only the Netherlands and China are on track to meet their 2030 targets, and both Chile and the Netherlands have revised targets downward. Firm offtake — the contracts that convert an announcement into a bankable project — went 2.4 Mtpa in 2023 to 1.7 Mtpa in 2024 to roughly 1.7 Mtpa in 2025, a cumulative total under 2 Mtpa, or “merely 5% of the potential production.” The binding constraint on this industry is not electrolysers, electricity or permits. It is that almost nobody has agreed to buy the product.

Established The geologic hydrogen case turns on a distinction most coverage never makes: whether the resource is a stock or a flow. A stock behaves like conventional gas, drained and abandoned, and is valued on volume in place. A flow behaves like geothermal heat: a fixed rate per well, indefinitely, valued on the rate and nothing else. Frontier Nearly every headline number published here is a stock number and nearly every economic claim made from it is a flow claim, and the two are not convertible. Established The federal assessment behind the best-known figures is explicit that its output is hydrogen generated and retained over geological time, that most of it is inaccessible, and that the recoverable fraction is not estimated. Speculative An in-place figure with an unquantified recovery factor is compatible with an industry and equally compatible with nothing at all.

Established Exploration has no validated play concept, which is a sharper problem than the flow rates. Petroleum exploration works because source rock, migration, reservoir and seal form a system predictable before drilling. Established Hydrogen generation is understood — serpentinisation of iron-rich ultramafic rock, radiolysis of water — but the seal term is weak: hydrogen is the smallest molecule there is, and caprocks that hold methane for a hundred million years are not automatically caprocks for it. Frontier Exploration therefore proceeds from surface expression, soil-gas anomalies and subcircular depressions, indirect indicators with no established false-positive rate, and the global count of wells drilled to test a hydrogen play is in the tens. Speculative Until a seal model predicts where hydrogen is retained rather than where it is escaping, drilling results are evidence about holes, not about a resource.

Frontier Replenishment is the claim that would make the resource interesting and the one with least direct evidence behind it. Bourakebougou has produced since 2012 without measurable pressure decline, consistent with active recharge, and that single observation is what the self-replenishing thesis rests on. Established It is also recharge sustaining 547,500 m³ a year into a village generator — two orders of magnitude below commercial flow. Frontier The published flow analysis is blunt about the consequence: if a system genuinely replenishes, the replenishment rate is the production ceiling. Speculative The escape route the field prefers is producing stock and flow together, drawing an accumulation down quickly and then settling to the recharge rate, which is a coherent reservoir model that has never been demonstrated on a hydrogen field because only one has ever been produced.

Established Purity is not the constraint, and composition still sets the cost. Mali gas is about 98% hydrogen, which is why purity keeps being quoted; the wider occurrence literature is dominated by streams in which hydrogen is a minority component diluted in nitrogen, with methane, carbon dioxide and helium variously present. Established Wellhead separation plant, pressure-swing adsorption or membranes, is sized by total gas throughput rather than hydrogen throughput, so capital cost per kilogram scales roughly with the inverse of the hydrogen fraction: a 40% stream costs something like two and a half times as much to process as a pure one at equal hydrogen output. Frontier The commercially interesting corollary is that co-products may be worth more than the hydrogen, since helium sells for a price per unit energy hydrogen has never approached. Speculative If geologic hydrogen has a first commercial decade, its likeliest shape is helium production with hydrogen credited as a by-product, which is a different industry from the one being announced.

Frontier There is no audited produced cost for geologic hydrogen anywhere. Figures in circulation, typically well under a dollar a kilogram, are developer models assuming per-well flow an order of magnitude or more above anything measured, quoted without the assumption that generates them. Established The one operating site supplies a small generator in a single village, which is a demonstration and not a cost datum. Frontier Stated honestly: the resource exists, one field has been produced, observed flow is two orders of magnitude short, and every published cost is conditional on closing that gap rather than evidence it can be closed.

What is genuinely open in hydrogen is narrower than the field's volume of activity suggests, because most of what is described as an open question is in fact a settled question whose answer is unwelcome. The efficiency chain is not open. The cost structure is not open. The willingness to pay is not open. Four things are.

Frontier The first genuine unknown is the leakage rate, and it is unknown by two orders of magnitude. Hydrogen is an indirect greenhouse gas: it consumes hydroxyl radical, extending atmospheric methane lifetime, and perturbs tropospheric ozone and stratospheric water vapour. The warming potential itself is not the open part. A five-model intercomparison puts GWP100 at 11.6 ± 2.8 and GWP20 at 37.3 ± 15.1, and a 2025 robustness study finds GWP100 essentially independent of emission size across 0.1 to 100 Tg per year, varying only 10.2 to 14.2 with emission location relative to soil sinks and rising only 3.0–7.8% under 2050 atmospheric composition, within one standard deviation. What is open is the multiplier. Value-chain emission estimates span 0.2% to 20% — liquefaction 0.15–10%, liquid hydrogen transport and handling 2–20%, liquid refuelling 2–15%, transmission pipelines 0.02–5.0%, distribution 0.0003–5.0%. Since climate impact is the product of warming potential and leakage, and one factor is known to about 25% while the other is known to within two orders of magnitude, any specific claim about a hydrogen system's net climate benefit is currently unfalsifiable in the direction that matters.

Established The reason that range is wide is worth stating precisely, because it is not the usual reason. The peer-reviewed authors are explicit that the spread is methodological rather than empirical: “It is virtually unknown how much H₂ is emitted intentionally and unintentionally from hydrogen systems since, to date, these emissions have not been measured.” Commercial sensors cannot resolve site-level emissions — the requirement is roughly 10 ppb sensitivity at seconds response, which is not a product anyone sells. The 0.2–20% range is therefore not a distribution of measurements. It is a distribution of assumptions.

Frontier And the distribution runs against the promotional case in a specific way. Electrolytic green hydrogen production has a wider and higher estimated leak range, 0.03–9.2%, than grey at 0.5–1.0% or blue at 0.0–1.5%, and liquid-hydrogen handling at 2–20% is the leakiest step in the chain. The architecture that hydrogen-as-universal-carrier requires — electrolytic production, liquefaction, long-distance movement, distributed refuelling — is precisely the leakiest architecture available. This brief also notes, as the pack does, that a 2025 study of model hydroxyl biases could not be retrieved and may bear on whether the 11.6 figure is itself biased; it is named here as unverified and is not in the reading list.

Frontier The second genuine unknown is what operating electrolysers actually achieve, and there is no dataset. The best-documented case is the world's largest operating green hydrogen project, a 260 MW alkaline installation at Kuqa in Xinjiang, which produced 2,010 t in 2023 — approximately 20% of rated capacity. A commercial analyst attributed the shortfall to insufficient power supply and to electrolyser underperformance at low load, and noted that levelised cost could rise by up to 20% on reaching rated capacity. Even at the operator's own subsequent target of 20,000 tpa the plant would run at roughly 51%. That is one project, in one year, reported by one analyst in August 2024, and it is not a fleet statistic. Neither the 2025 nor the 2026 intergovernmental review publishes operating capacity factors, and the pack could not verify whether the operator's promised fixes landed in late 2025. Anyone quoting a global hydrogen utilisation figure is quoting an assumption.

Frontier The third is whether cheap electrolysers are equivalent electrolysers. A commercial analyst puts Chinese full alkaline systems at about $1.0 million per megawatt in 2024–25, falling to roughly $0.8 million in 2026, against Western alkaline at about $2.0 million per megawatt in 2025 — and then argues that a technologically equivalent, commercially viable Chinese system is closer to $1.3 million per megawatt, a 30–40% discount rather than the headline 50–60%. That adjustment is an analyst's judgement, not a measurement. The mechanism offered is testable and untested: Chinese systems are said to have a 30–40% minimum load against roughly 20% for Western equipment, reducing their ability to follow variable renewable output, with the warning that “frequent start-stop operations could accelerate degradation and compromise stack durability.” Fleet degradation data does not exist. This is a plausible mechanism whose outturn will be known in about five years, and it matters because the entire cost-down case for green hydrogen currently runs through Chinese equipment.

Frontier The fourth is geologic hydrogen, where the geochemistry is settled and the commerciality is not. Formation by serpentinisation and radiolysis is established. Exactly one commercial site exists on Earth: Bourakebougou in Mali, confirmed as the only operating well by the 2026 review. A peer-reviewed 2026 flow-rate analysis is the sharpest test available: observed natural hydrogen occurrences typically flow 10⁵ to 10⁷ m³ per year, commercially viable rates need to be at least an order of magnitude higher, and conventional gas wells run at 15 to 111 million m³ per year. Mali delivers 547,500 m³ per year at 98% purity — about two orders of magnitude too small for economic production, with the authors concluding that economically recoverable natural hydrogen from self-replenishing systems is unlikely. That conclusion is a published finding and is contested; it is flagged frontier. A learned-society assessment independently estimates continental geological hydrogen flux at under 0.74 Mt per year excluding volcanic sources, “significantly lower than some recently proposed values based on modelling”, and lists reservoir volumes, distribution, production costs and subsurface microbial impacts as unknown.

Handwave Any use of the “twice the energy in all proven natural gas reserves” figure without its caveat. That number is resource-in-place modelling, reported by the surveying agency alongside its own statement that much of the resource is “likely too deep, too far offshore or in accumulations too small to be economically recoverable,” with “considerable uncertainty.” A resource-in-place model is not a reserve. Meanwhile geologic hydrogen has attracted about 15% of all hydrogen-production-technology venture capital, with natural-hydrogen equities delivering threefold returns, while the same review states initial production is possible in the 2030s and large-scale deployment is unlikely before the 2040s. The divergence between the valuations and the production forecast is the tell.

Frontier What is not researched here, stated as a gap rather than answered. The pack did no work on hydrogen for trucks, buses, trains or aviation, which is the strongest remaining transport claim and the position an advocate will retreat to when light vehicles are conceded. The agency expects trucks and buses to be roughly 60% and 30% of hydrogen consumption in transport respectively, and that expectation is recorded here without endorsement. Heavy-duty economics, hydrogen heating policy outside the United Kingdom, salt-cavern storage beyond the announced-versus-committed figures, the sectoral megatonne split of the existing 100 Mt demand, and a verified 2025–26 levelised cost per kilogram are all outside what this brief can support.

Frontier Stimulated hydrogen is the answer to the flow-rate problem, and it moves the subject out of exploration and into chemical engineering. The proposal, sometimes labelled orange hydrogen, is to stop hunting accumulations and drive the generating reaction on demand: inject water, and where useful a catalyst, into iron-rich ultramafic rock at depth and produce what the reaction makes. Established The chemistry is not in doubt; serpentinisation of olivine liberates hydrogen and runs fastest at temperatures in the low hundreds of degrees, a reachable depth in the right terrane. Frontier The engineering objection is specific and severe: the products occupy substantially more volume than the reactants, so the reaction progressively closes the porosity the injected water needs to reach fresh mineral surface. Speculative A process that seals its own flow path is not obviously fixable by better well design, and until someone reports sustained production from a stimulated interval rather than an opening transient, this is a mechanism rather than a technology with a yield.

Established The first bottleneck is willingness to pay, and it binds before anything else because it caps the price at which any of the others could be solved. Absent policy support, the maximum acceptable hydrogen cost is below USD 2 per kilogram for most sector and region combinations, with refining and ammonia at the top of the range because energy is a large share of their product cost, and steel requiring a negative cost — an incentive — to reach parity. Every supply-side improvement in this brief is chasing a ceiling that is set by what buyers can pass on, and the two large European steelmakers that walked away in 2025 walked away with subsidy on the table.

Established The second is the cost structure of the plant, which is not where the learning is. The stack is 15–20% of investment cost and more than half is engineering, procurement, construction and contingency. That is why Chinese equipment installed in Europe cuts total capital expenditure by only about 20%, and why a survey found Western installed capital cost rising by a median 57% in a single year against a forecast of 8–10% annual declines. The regional pattern in the investment data says the same thing from the other end: China holds more than 60% of electrolysis capacity but only 25% of investment, while Europe holds under 20% of capacity and 45% of investment, explicitly attributed to higher capital expenditure per unit of capacity.

Established The third is thermodynamic and it does not move with engineering maturity. Power to hydrogen to power delivers 24–48% round trip against 60–98% for batteries; hydrogen heating delivers 70% against 278% for a heat pump. This is a constraint on the addressable market rather than on hydrogen itself: it means that wherever electricity can do the job directly, hydrogen must be two to four times cheaper per unit of input energy simply to draw level, and electricity is not getting more expensive. The deployment record follows: 108 GW of batteries added globally in 2025, about 87 GW of it utility-scale and roughly 40% year-on-year growth, against about 4 GW of cumulative installed electrolysers — a ratio near 27 to 1 comparing one year of one technology against the entire history of the other. Battery costs fell more than 90% between 2010 and 2025.

Established The fourth is offtake, which is the bottleneck the industry itself names. Firm offtake agreements ran 2.4 Mtpa in 2023, 1.7 Mtpa in 2024, roughly 1.7 Mtpa in 2025 — cumulatively under 2 Mtpa, “merely 5% of the potential production”, and much of what exists is non-binding. A project without an offtaker cannot reach a final investment decision, which is the mechanism converting an announced pipeline into a cancelled one.

Frontier The fifth is utilisation, on which there is almost no public data at all. The one well-documented operating project ran at about 20% of rated capacity in its first full year, attributed to insufficient power supply and to poor electrolyser performance at low load. On the manufacturing side, electrolyser factories were reported operating at roughly 10% of capacity on average — a finding whose publication date the pack could not recover, so its currency is uncertain. Both figures are single observations. Neither of the last two global reviews publishes fleet capacity factors, and this brief does not generalise from one plant.

Established The sixth is that the infrastructure a hydrogen economy requires does not exist and is not being built. Nine per cent of announced pipeline kilometres are operational or committed; just over 7% of announced underground storage is at investment decision. Storage and transport are not downstream conveniences here — without them, hydrogen produced from variable renewables cannot be time-shifted or moved to where the demand is, which is the whole premise of the carrier claim.

Established The seventh binds on the supply side rather than the demand side and is now visible in accounts. A listed electrolyser manufacturer reported fourth-quarter 2025 revenue of NOK 330 million, down 20% year on year, with a net loss of NOK 870 million including an impairment of NOK 799 million split between its membrane and alkaline lines, and an order backlog down 18%. The company's own explanation is the pipeline story in a sentence: final investment decisions took longer, project milestones shifted, and revenues declined. Capital spending on low-emissions hydrogen worldwide reached only about $7 billion in 2025, with roughly $10 billion projected for 2026.

Frontier The eighth is an accounting gap rather than an engineering one, and it will bind later. Nobody can measure hydrogen leakage at site level, and no regulator requires it. A carrier promoted on climate grounds whose warming contribution depends on an unmeasured parameter will eventually face a rule about that parameter, and the sector currently has no instrumentation with which to comply.

Established One constraint here is not geological at all: the resource is frequently not legally a resource. Mineral and petroleum tenure systems enumerate the substances they cover, and in most jurisdictions hydrogen is not among them, so there is no licence to apply for, no tenure to hold and nothing to borrow against. Established South Australia amended its petroleum and geothermal legislation to name hydrogen as a regulated substance and became, on that account rather than any geological one, where exploration licences were actually issued. Frontier It is the classic institutional bottleneck: cheap to fix, blocking until fixed, and invisible in every technical assessment of the resource.

Frontier The second is leakage, inherited from the rest of this brief. Hydrogen is an indirect greenhouse gas, and a produced molecule escaping at the wellhead, in gathering, or through the same permeable overburden that made the accumulation hard to seal carries a climate cost the zero-emission framing omits. Established The measurement regime that would settle it is the same one this brief already lists as a requirement, so no new capability is being asked for — only that the gap be applied to a production route usually exempted from it rhetorically. Speculative A resource whose distinguishing property is migration through rock that holds other gases is not an obvious candidate for low fugitive emissions, and nobody has measured it.

Established Nothing here waits on a physics result, which is worth saying plainly because the field is often discussed as though it did. Alkaline electrolysis is more than a century old. Fuel cells, ammonia synthesis, methanol synthesis and direct reduction all operate at industrial scale today, and the one genuinely novel piece of metallurgy in the story — hydrogen-reduced sponge iron — has already been demonstrated at pilot scale. What the economy-wide claim waits on is not a discovery.

Established It waits first on a buyer. Firm offtake stands at under 2 Mtpa, about 5% of potential production, and much of that is non-binding. Every other dependency in this list is downstream of somebody agreeing to pay, and the willingness-to-pay ceiling of under $2 per kilogram absent policy is the number that governs whether they will.

Established It waits on electricity that is both cheap and available for enough hours of the year. The best-documented operating plant underperformed in substantial part on insufficient power supply and on poor electrolyser performance at low load, and the cell chemistry that dominates the cheap end of the market is reported to have a 30–40% minimum load against roughly 20% for Western equipment. Electrolytic hydrogen is therefore a derivative of the generation question rather than an independent technology — it competes with direct electrification for the same electrons, and loses wherever direct electrification is possible.

Established It waits on transport and storage that do not exist. Nine per cent of announced pipeline kilometres and just over 7% of announced underground storage have passed a commitment threshold. Without them the carrier claim has no mechanism.

Frontier It waits on a measurement nobody is taking. Site-level hydrogen emissions have never been measured, the instrumentation requirement is around 10 ppb at seconds response, and no regulator mandates it. This is an accounting and metrology dependency rather than an engineering one, and it is the one most likely to arrive as a shock.

Established What depends on hydrogen is narrower than the sector claims and is not nothing. Ammonia, methanol and refining depend on it absolutely, today, at roughly 100 Mt a year and 1,300 Mt of CO₂-equivalent emissions. Hydrogen-based zero-carbon industry depends on it in the specific case of direct reduced iron. Ammonia and methanol as marine fuels depend on it, and the orderbook says the shipping industry has not yet decided to take the dependency. Nothing else does.

Established The heating experiment was run, twice, and both instances were cancelled before they produced data. The United Kingdom's hydrogen village trials were eliminated at Whitby in July 2023 and cancelled at Redcar in December 2023, the latter on the stated grounds that the trial “cannot go ahead as designed, as the main source of hydrogen supply will not be available.” The government letter recording this is itself formally withdrawn as out of date. The 2026 development is quieter and more damaging than either cancellation: the promised 2026 decision deadline on hydrogen's role in heating has passed with no decision, only the H100 Fife neighbourhood trial remains, and the national climate advisory body recommends that government “rule out 100% hydrogen for heating.” The same source records a 10 GW 2030 national target against 2 GW shortlisted or approved for funding, and an estimated £118 per year household bill increase from hydrogen levies.

Established The literature experiment has been run twice by independent groups and returned the same answer. A peer-reviewed meta-review of 54 studies found that none supports widespread hydrogen for building heat, with a median hydrogen contribution in cost-optimal pathways of 1% of final energy demand (range 0–10%), energy-system costs up a median 24%, consumer costs up a median 86% with a range from +27% to +650%, and zero studies in which hydrogen was cheaper for consumers than electrification. An independent 2026 meta-analysis of 50 studies by a different author group reached the same conclusion: heat pumps remain most efficient and most cost-effective, direct hydrogen heating is 5–30% more expensive, hydrogen system efficiency runs “as low as 25%”, and 20–30% volumetric blends deliver only about 7–10% CO₂ reduction because hydrogen carries so little energy per unit volume. Two reviews agreeing is a materially different evidentiary position from one, because a single-author meta-review invites a single-author rebuttal.

Established California is a natural experiment on the chicken-and-egg argument, and it came out the other way round. The joint agency assessment published 1 May 2026, on data as of 2 September 2025, records 50 stations open to the public and 11 temporarily non-operational, with network availability around 60% because of maintenance and equipment failures, and 14,128 registered light-duty fuel-cell vehicles against network capacity for about 34,300 — the regulator's own phrase is that “statewide fueling capacity exceeds demand.” In the one jurisdiction that built the stations first, the stations are not the constraint. A March 2026 gaseous-supply disruption reportedly took most of the network offline; the pack could not retrieve either report and this brief asserts no number for it.

Established The industrial experiments are the ones that mattered, and two large ones returned negative results in 2025. A German steelmaker suspended its green hydrogen tender in March 2025 because “proposed hydrogen prices were significantly higher than the company expected”, having sought 104,000 t by 2028 and 143,000 t a year from 2029 to 2035, and will start its Duisburg direct-reduction plant on natural gas instead, claiming roughly 50% CO₂ avoidance against a blast furnace. In June 2025 another cancelled direct-reduction and electric-arc projects at two German sites, forfeiting €1.3 billion of federal and state subsidy, on the grounds that the business case was not strong even with the subsidy, that green hydrogen viability was slower than anticipated, that natural-gas direct reduction was not competitive either, and that German electricity prices were uncompetitive. The pattern across the sector is that the shaft furnace is being built and it is being fed natural gas. The metallurgy works; the hydrogen does not clear at any price steel can pay, which is exactly what the cost-acceptability finding predicts.

Frontier The largest green steel experiment is still running and is in doubt on schedule rather than on principle. A Swedish greenfield plant designed around 700–800 MW of on-site electrolysis and 2.5 Mt a year of steel, which raised €6.5 billion, was about 60% built by autumn 2025, hit rising costs and a cash crunch, and closed a €1.4 billion rescue package announced in April 2026. Its original late-2026 completion target is “under review”, with production start put at 18 to 24 months after facility completion. That it was rescued and delayed is established; any specific start date is not. The Swedish pilot consortium that first produced hydrogen-reduced sponge iron achieved a genuine technical first, and this brief cannot report its current status: the consortium's own page appears to date from 2023 and still promises fossil-free steel to market in 2026.

Frontier Two decisive experiments are already scheduled and need no new apparatus. More than 100 GW of announced electrolysis either commits before the end of 2027 or loses any chance of operating by 2030, with 22 Mt of announced production at risk without a decision by early 2027 — a clean test of whether the 2024–26 attrition was a shakeout or a trend, resolving on a date already fixed. And the third European auction awarded 1.1 GW across nine projects at €0.44–3.49 per kilogram, where the second lost 83% of awarded capacity between award and signature; the third round's signature rate is a direct replication of that measurement on a fresh cohort, and it will be known well before any of the projects would have been built.

Established One positive result, recorded because negative findings dominate this brief. China took nearly three-quarters of new electrolyser installations in 2025 and holds about 65% of installed-plus-committed capacity with roughly 60% of global manufacturing capacity. Whatever else is true, the thing is being built somewhere, at scale, into ammonia and methanol demand that already exists — which is the same finding as the industrial leg, arriving from the supply side.

Established The attrition ladder, asset class by asset class, because the ratios are the finding. Each row compares what was announced with what has passed an investment decision and what exists. The rows are drawn from different editions and chapters of the same intergovernmental review series and from one commercial analyst, and the sources are named alongside; the percentages are this brief's arithmetic on the sourced inputs.

Established The named withdrawal record for 2024 to 2026, because it is short enough to enumerate and because aggregate pipeline statistics conceal who actually left. Entries are from a wire-service factbox, trade press and government sources; stated reasons are the parties' own.

Established The conversion chain, which is where the physics enters and does not leave. Electrolysis efficiency on a higher-heating-value basis, by cell chemistry: alkaline 50.5–78.8%, proton-exchange membrane 47.5–78.8%, anion-exchange membrane 57.1–69.1%, solid oxide 71.6–87.6%. Compression alone then costs 7% to 27.8% of hydrogen's higher heating value before the molecule has moved anywhere. Reconversion to electricity runs at 40–48% in a hydrogen internal combustion engine, 35–39% in a gas turbine, 40–70% in a fuel cell. Multiplying through, power to hydrogen to power delivers 24–26.8% through a gas turbine, 27.4–33% through an engine and 27.4–48% through a fuel cell, against 60–98% round-trip for batteries. The multiplication is the source's; the summary range of roughly 25–48% against 60–98% is this brief's reading of it.

Established The same penalty appears in heating from an entirely independent direction, which is why it should be believed. A peer-reviewed meta-review computes green hydrogen heating system efficiency at 70% against 278% for a heat pump at a seasonal coefficient of performance of 3.0, with heat pumps needing four to six times less energy input. Two unrelated literatures, one on grid storage and one on domestic heat, produce the same factor of roughly two to three in delivered energy per unit of electricity in, and in heating the factor is nearer four.

Established What the brief cannot give, and why. There is no verified levelised cost of hydrogen in dollars per kilogram for 2025 or 2026 in this brief. The pack could not source one: the intergovernmental review publishes 2030 projections, and the summary that carried them was an interested party's. The only per-kilogram figures here are therefore of three other kinds, and they are not interchangeable — auction subsidies of €0.20–3.49 per kilogram, an intergovernmental willingness-to-pay ceiling below $2 per kilogram, and a Californian retail dispensed price that hit $34.11 per kilogram on 1 February 2024 for light-duty fuel, which is a pump price for mostly reformed hydrogen and not a production cost at all.

The closest neighbour is zero-carbon industry, which reaches hydrogen from the emissions side and owns the steel case in full. The boundary is deliberate and the two treatments meet at direct reduced iron: that brief asks how to decarbonise the process, this one asks what the hydrogen costs and whether anyone will buy it. On the evidence here the answer feeds back the other way — the shaft furnaces are being built and fed natural gas, so the emissions question is being resolved without the molecule this brief is about.

Advanced battery technologies is adjacent in the strong sense: it is the competitor, and the round-trip comparison of 24–48% against 60–98% is the single fact that decides the storage leg. Batteries added 108 GW in 2025 against roughly 4 GW of cumulative electrolysers, so this is not a contest between two emerging options but between a deployed technology and a proposed one.

Future ports and shipping carries the marine orderbook in detail and is the place to read the ammonia and methanol case properly. It appears here only as evidence: six ammonia-and-methanol orders out of 137 in the first half of 2026, against 128 for fossil LNG and LPG.

Small modular reactors and geothermal megaprojects are adjacent as suppliers of the firm, high-availability electricity that would raise electrolyser capacity factors above the 20% the best-documented plant achieved. Atmospheric management is adjacent because hydrogen is an indirect greenhouse gas with a well-constrained warming potential and an entirely unconstrained leak rate, which makes it a case study in a problem that brief treats generally: what to do about a forcing agent nobody measures.

One further neighbour is not a topic but a habit of reading. Every brief in this corpus that quotes an announced pipeline — whether of reactors, of capture plants or of transmission corridors — is quoting a number of the same kind as the ones this brief takes apart, and the attrition ratios here are the closest thing available to a calibration for them.

Established The institutional record of 2024 to 2026 is a sequence of withdrawals by parties who had already won public money, which is a stronger signal than a policy reversal. A government changing its mind tells you about a government. A developer abandoning 83% of awarded capacity rather than sign a grant agreement, or a steelmaker forfeiting €1.3 billion of committed subsidy, tells you what the people closest to the economics believe about the economics. Sixteen named corporate withdrawals, two state programmes cancelled outright, and a national hub programme placed on a termination list all fall within twenty-four months.

Established Who buys is the defining institutional fact, and it is not who the strategies assume. The buyers of hydrogen are refiners, ammonia producers, methanol producers and iron reducers, and they buy roughly 100 Mt a year of it as feedstock, almost all of it grey. They are the buyers with the highest tolerance for cost, because energy is a large share of their product cost, and they are where the committed money is going: 2.5 Mt of committed 2030 consumption, 60% of all committed low-emissions production, with about 80% of strong-potential production targeting chemicals. The utility, the household and the motorist — the buyers on whom the universal-carrier framing depends — are not in the market in any volume, and the sector's own contracting record shows it: firm offtake under 2 Mtpa, 5% of potential production.

Established Who funds has changed direction inside two years. Sixty-six countries have national hydrogen strategies and two — the Netherlands and China — are on track for their 2030 targets, with Chile and the Netherlands having already revised downward. New public funding continues to be announced, and private capital expenditure on low-emissions hydrogen reached only about $7 billion in 2025 with roughly $10 billion projected for 2026 — small numbers for a technology described as a pillar of the energy transition. The geographic split of investment is itself an institutional finding: China holds more than 60% of electrolysis capacity but 25% of investment; Europe under 20% of capacity and 45% of investment, explicitly because European capital cost per unit of capacity is higher. Europe is buying less capacity with more money, and it took nearly three-quarters of new installations away from Europe in 2025 for that reason.

Established Who regulates is fragmented, and the fragmentation is doing real work. A maritime regulator's deferral of greenhouse-gas pricing to October 2026 is the stated reason alternative-fuel vessel orders were deferred. A tax code provision terminating for facilities beginning construction after 31 December 2027 now sets the American investment clock. A European auction mechanism has run three rounds and, on the evidence of the second, has no reliable relationship between what it awards and what gets built. A state agency in California publishes a station-by-station availability figure of roughly 60% and a vehicle count against network capacity. Each is competent within its remit and none of them regulates hydrogen as a system.

Frontier The institution that does not exist is a measurement authority. Nobody measures hydrogen leakage at site level; nobody publishes electrolyser fleet capacity factors; nobody audits the announced-versus-committed pipeline other than an intergovernmental agency doing it once a year as a review chapter, and even that agency's 2026 edition carries two non-comparable figures for committed capacity in different chapters. A sector this dependent on public subsidy has no independent statistical apparatus, which is precisely why the announcement figures have been able to circulate for a decade without correction.

Frontier The governance question the sector has not resolved is what to do about a technology whose case rests on the applications it keeps losing. Heating has been examined by two independent meta-reviews and a national climate advisory body and recommended against. Light transport is outsold roughly 1,250 to 1. Bulk storage loses on physics to a technology deploying 27 times faster. Meanwhile the industrial substitution case — narrow, real, worth roughly 100 Mt a year and 1,300 Mt of CO₂-equivalent — is underfunded relative to the attention the carrier framing absorbs. Continuing to fund a general-purpose hydrogen economy while the evidence narrows to feedstock substitution is a live allocation choice, and it is almost never stated as one.

The first ethical question in this subject is evidence quality, because the sources are unusually asymmetric. Established The negative findings in this brief come overwhelmingly from disinterested parties — an intergovernmental agency publishing the same method three years running, two independent peer-reviewed meta-reviews, a research institute's efficiency assessment, a state regulator's own capacity audit, a parliamentary research service. The positive claims come disproportionately from parties selling something: vendors, classification societies, industry platforms, developer disclosures and government press releases about awards. This brief marks interested parties in its reading list and uses them for facts they would be embarrassed to get wrong — a shipped order count, an impairment charge, a plant's nameplate — not for outlook.

Established The second is public money, and the sums are specific. A European auction deployed €271 million of a €1.2 billion budget because its winners walked away. A German steelmaker forfeited €1.3 billion of federal and state subsidy rather than build. An Australian state spent on a 200 MW, A$600 million programme it cancelled, redirecting the funds to rescue a steelworks and disbanding the office created to deliver it. The United States moved to terminate hydrogen hub awards drawn from an $8 billion infrastructure allocation and separately legislated that its clean hydrogen production credit terminates for facilities beginning construction after 31 December 2027. The pattern is not that governments backed a technology that failed. It is that governments repeatedly funded the announcement stage of projects whose developers subsequently declined to proceed, which is a procurement design question rather than a technology question, and nobody has been held to account for it.

Frontier The third is opportunity cost, which in heating is quantified and falls on households. Consumer cost increases for hydrogen heating have a median of 86% and a documented range to +650%, against a heat pump already on sale that needs four to six times less energy. A parliamentary estimate puts £118 per year of household bill increase from hydrogen levies. Where a public programme funded hydrogen heating trials while a superior appliance was already available, the cost of the delay fell on the people who waited, and it fell hardest on those least able to pay it. Both village trials also ended amid local resistance, which is a governance outcome rather than a technical one and deserves to be recorded as such.

Established The fourth is disclosure inside the climate case. A technology promoted on emissions grounds has a warming potential of 11.6 over a century and 37.3 over twenty years and a leakage rate that its own peer-reviewed literature states plainly “has not been measured.” Marketing that describes hydrogen as zero-emission at the point of use without addressing indirect warming or leakage is asserting a benefit without a mechanism. The obligation runs both ways: the leakage range is not evidence that hydrogen is dirty, it is evidence that nobody knows, and a brief that used the upper bound rhetorically would be committing the mirror image of the error it criticises.

Finally, the questions this brief cannot answer, stated as an obligation rather than an omission. Frontier It does not know the capacity factor of the world's electrolyser fleet, because no such dataset exists. It does not know a verified 2025–26 levelised cost per kilogram for any production route, because the pack could not source one and the available figures were an interested party's summary. It does not know the current status of the Swedish hydrogen-reduced iron demonstration, whose own page appears to date from 2023. It does not know the fate of five of seven United States hydrogen hubs after the October 2025 termination list, nor whether any funding was restored. It does not know the status of the 2.2 GW Saudi green hydrogen project, reported near completion but unverified. It does not know how many Californian stations survived the March 2026 supply disruption. It has done no work on hydrogen for trucks, buses, trains or aviation, which is the strongest remaining transport claim. And it has excluded a gigawatt-denominated announced-versus-committed series for 2025 and 2026 that the pack recovered but could not corroborate. Each of those is a place where a reader should not assume coverage.

Established The civilisationally significant fact about hydrogen is not the molecule; it is what the last three years demonstrate about announcement-driven industrial policy. A technology accumulated 520 GW of announced 2030 electrolyser capacity, 40,000 km of announced pipelines, 66 national strategies and tens of billions in public commitments, and converted 4% to 9% of it into anything with a signature on it. The general principle the case illustrates is that an announcement pipeline is a measurement of enthusiasm, not of capability, and that any planning system which treats the two as interchangeable will systematically over-invest in whatever is currently most fashionable to announce. The attrition ratios in this brief are the most transferable thing in it.

Established The second general principle is that an energy carrier must beat the wire. Electricity delivered directly, at 90 %-plus efficiency, is the incumbent in every application where a wire can reach. Hydrogen loses that comparison by a factor of two to three in storage and four to six in heating, and those factors are set by conversion physics rather than by engineering maturity, so they will not improve with scale. Where a wire cannot reach — a chemical bond in a fertiliser molecule, the reduction of iron ore — hydrogen has no competitor and the case is strong. The universality claim fails not because hydrogen is bad but because it was asked to compete where it has a structural handicap.

Established The prize that remains is genuinely large and is not the one the sector advertises. Roughly 100 Mt a year of hydrogen is already made and used, consuming 290 billion cubic metres of natural gas and 90 Mtce of coal and emitting about 1,300 Mt of CO₂-equivalent annually with no progress in reducing it. Displacing that is a real civilisational task on the scale of a mid-sized national economy's emissions. Low-emissions hydrogen is currently about 1 Mt of it, and committed 2030 production of roughly 4 Mtpa is about 4% of the target. Framing hydrogen as a future energy carrier has distracted attention from the existing hydrogen industry, which is a present emitter and a present opportunity.

Frontier And the climate premise on which the whole enterprise rests is not closed. Warming potential is well constrained; leakage is constrained only to within two orders of magnitude, and the architecture required for a universal carrier is the leakiest one available. A civilisation that built a hydrogen distribution network at scale without first learning to measure hydrogen emissions at site level would be running an uncontrolled experiment on the methane lifetime of its own atmosphere, and would not find out for years.

Established 2023, the year the heating case ended in the United Kingdom. Whitby eliminated in July, Redcar cancelled in December for lack of hydrogen supply. In the same year the largest operating green hydrogen plant in the world produced about 20% of its rated output, and the peer-reviewed meta-review of 54 heating studies was completed.

Established 2024, the year the announcements peaked and the withdrawals began. Announced 2030 production stood at 49 Mtpa with only 4% of 520 GW of announced electrolysers at investment decision. A major retailer permanently closed all seven of its Californian light-duty stations in February, with retail hydrogen at $34.11 per kilogram. Western electrolyser installed capital cost rose a median 57%. Fortescue, Shell Norway, Woodside, Hy Stor, Neste, Origin and Kawasaki all withdrew between July and December. A commercial analyst counted 20.3% of the European pipeline cancelled or stalled by year end.

Established 2025, the year the withdrawals reached the parties holding public money. Repsol, Air Products, Trafigura, BP and LEAG exited; a German steelmaker suspended its hydrogen tender in March and another forfeited €1.3 billion of subsidy in June; South Australia cancelled a 200 MW, A$600 million programme in May and disbanded its hydrogen office; European Hydrogen Bank bidders abandoned 83% of awarded capacity in September; the United States moved in October to terminate hydrogen hub awards worth $8 billion, with two West Coast hubs firmly cancelled. New investment decisions fell below 0.8 Mtpa. Installed electrolysers passed 4 GW, batteries added 108 GW, and a Chinese integrated hydrogen-to-ammonia project came online in December.

Established 2026 to date. The second auction's final outcome was confirmed in January at 380 MW and €271 million of a €1.2 billion budget. The Swedish green steel plant was rescued with €1.4 billion in April. A third auction awarded 1.1 GW around May. The parliamentary briefing of 28 May recorded a missed decision deadline on heating and a recommendation to rule out 100% hydrogen. The sixth edition of the global review appeared on 18 June with announced 2030 production at 27 Mtpa.

Established Fixed forward dates, of which there are two that matter. The United States clean hydrogen production credit now terminates for facilities beginning construction after 31 December 2027. And more than 100 GW of announced electrolysis loses its 2030 window without an investment decision before the end of that same year. Two independent deadlines land in the same quarter, and the sector's response to them will be the clearest signal it has produced in a decade.

Frontier Late 2020s. Either committed low-emissions production grows materially beyond 4 Mtpa against 100 Mt of existing demand, or hydrogen settles as an industrial feedstock substitution route and the energy-carrier framing is quietly retired. On the announced-versus-committed ratios, the second is the way to bet, and the announced figure should be expected to fall again.

Frontier 2030s: geologic hydrogen at initial production. The review's own phrasing is that initial production is possible in the 2030s and large-scale deployment is unlikely before the 2040s, against one operating well today running two orders of magnitude below commercial flow.

Handwave Any date for hydrogen as a general-purpose energy carrier in heating, light transport or bulk storage. Two peer-reviewed meta-reviews, a national climate advisory body, a state regulator's own capacity assessment and a 1,250-to-1 sales ratio all point the same way. Dates in this category are marketing and this brief declines to supply one.

• Geologic hydrogen, now to 2030: Frontier The tenure question resolves in most prospective jurisdictions because it is cheap to resolve, and licence counts rise sharply while telling nobody anything about the resource. Frontier A second producing field appears or does not; one is an anecdote, two is the start of a distribution.
• Geologic hydrogen, 2030s: Frontier Initial production is possible on the review’s own phrasing, and what would justify it is a sustained per-well flow rate an order of magnitude above anything yet observed. Speculative Helium co-production is the likeliest thing to be financed first, and it will be reported as a hydrogen milestone.
• Geologic hydrogen, 2040s and beyond: Handwave Large-scale deployment is not expected earlier on any published view, and whether it happens depends on a stimulated-production result that does not yet exist in a real formation. Handwave If stimulation fails, this is a helium play with a by-product and a few fortunate local supplies — a real industry, and not the one described.

• Depends on Nothing on this map, and nothing in physics. Every conversion step in the hydrogen chain is industrially mature, which is why this topic's failures are commercial rather than technical and why no research result would rescue the applications it has lost. The dependency that actually dominates — abundant cheap electricity at high annual availability — is a property of the generation fleet rather than a result any brief can deliver, and it makes this topic a derivative of the generation question rather than an independent technology.
• Requires (not on this map) A buyer: firm offtake is under 2 Mtpa, roughly 5% of potential production, and the ceiling on what most buyers can pay is below $2 per kilogram absent policy. Electricity cheap enough and available enough hours that an electrolyser can run above the 20% capacity factor the best-documented plant achieved. Transport and storage infrastructure, of which 9% of announced pipelines and 7% of announced storage have passed a commitment threshold. A leakage measurement regime, which is an instrumentation and regulatory gap rather than an engineering one. And electrolyser factories with customers, given a reported 10% average utilisation. All five are market, industrial or institutional capabilities, and their absence is what the field is actually waiting on. A sixth belongs to geologic hydrogen specifically and is purely legal: a tenure regime naming naturally occurring hydrogen as a substance an explorer can be licensed to seek and hold title to. Where that instrument does not exist, which is most of the world, there is nothing to permit, nothing to lease and nothing a lender can take security over, and the constraint binds before any question about flow rates is reached. It costs nothing to supply and no research result produces it.
• Enables Decarbonisation of ammonia, methanol, refining and direct reduced iron — the roughly 100 Mt a year of existing hydrogen demand emitting about 1,300 Mt of CO₂-equivalent. The enabling relation to steelmaking is real but is carried by the Carbon Problem programme rather than by a Frontier Research brief, so no typed edge is claimed here. Speculative And, separately, hydrogen as a general-purpose energy carrier at economy-wide scale — the claim the phrase “hydrogen economy” actually makes. That outcome requires a demand-side willingness to pay which does not currently exist in most sectors, before any infrastructure is costed, and it is retained here as a flagged speculative enabling relation rather than dropped, because it is the proposition the topic is named for and the one the deployment record argues against.
• Adjacent Advanced battery technologies, the competitor that wins on round-trip efficiency by a factor of two to three; future ports and shipping, which holds the marine fuel orderbook; small modular reactors and geothermal megaprojects as candidates for the firm cheap power electrolysis needs; and atmospheric management, because hydrogen is itself an indirect greenhouse gas with an unmeasured leak rate.

“Announced project pipelines show hydrogen is scaling.” Handwave They show the opposite once the rungs are separated. Announced 2030 production fell 49 to 37 to 27 Mtpa across three consecutive editions of the same review while committed capacity stayed near 4 Mtpa, and in the 2024 edition only 4% of roughly 520 GW of announced electrolysers had reached an investment decision or construction. More than half of potential electrolyser capacity is now set to slip past its target operational date, and half of announced projects face deferred start dates. An announcement is not a project, a project is not a decision, and a decision is not a plant; the conversion rates between those rungs are 4% to 20% depending on the asset class, and quoting the top rung as though it were the bottom one is the defining error of writing on this subject.

“Electrolyser costs are falling on a learning curve.” Established Western installed capital cost rose by a median 57% in the 2024 survey year against the surveyor's own forecast of 8–10% annual declines. More importantly, the learning-curve argument is aimed at the wrong component: the stack is 15–20% of investment cost and more than half the total is engineering, procurement, construction and contingency. Even a stack that became free would leave most of the cost standing. Any projection that extrapolates cell-level manufacturing learning onto system cost is unsupported by the cost structure and should be treated as such.

“China has solved electrolyser cost, so the West will import the savings.” Established Installing Chinese equipment in Europe cuts total capital expenditure by only about 20%, because the cost is mostly not the equipment. Chinese kit installed outside China runs $1,500–2,400 per kilowatt once transport and tariffs are counted, against $600–1,200 at home and $2,000–2,600 for Western systems. The China–West gap is not primarily a manufacturing-learning gap and treating it as one will mislead.

“Chinese electrolysers are simply cheaper, full stop.” Frontier A commercial analyst puts Chinese alkaline systems at about $1.0 million per megawatt in 2024–25 against Western alkaline at about $2.0 million in 2025, and then argues that a technologically equivalent, commercially viable Chinese system is nearer $1.3 million — a 30–40% discount, not 50–60%. The mechanism offered is a 30–40% minimum load against roughly 20% for Western equipment, poorer at following variable renewables, with start-stop cycling warned to accelerate degradation. That is a live mechanism claim, not a finding: fleet degradation data does not exist, and many manufacturers are reported to be selling at a loss on subsidy.

“Hydrogen is the fuel of the future for heating.” Handwave Two independent peer-reviewed meta-reviews covering 54 and 50 studies found no study supporting widespread hydrogen for building heat. Median contribution in cost-optimal pathways: 1% of heating final energy. Energy-system costs up a median 24%; consumer costs up a median 86%, range +27% to +650%; zero studies in which hydrogen was cheaper for consumers than electrification. A heat pump delivers 278% system efficiency against roughly 70% and needs four to six times less energy input. Blending does not rescue it either: a 20–30% volumetric blend cuts CO₂ by only about 7–10%, because hydrogen carries so little energy per unit volume that a fifth of the pipe is a twentieth of the fuel.

“The United Kingdom hydrogen heating trials were cancelled recently, which shows momentum turning.” Established The cancellations are older than they are usually reported: Whitby in July 2023, Redcar in December 2023, and the government letter recording them is formally withdrawn. The 2026 fact is different and stronger: the promised 2026 decision deadline passed with no decision, the climate advisory body recommends ruling out 100% hydrogen for heating, and government now describes heating as “likely having limited applicability.” A non-decision after a deadline is a more revealing datum than a cancellation, because nobody has to defend it.

“Hydrogen car sales are collapsing.” Established Be careful, because the 2025 headline runs the other way and a hostile reader will produce it. First-half 2025 global fuel-cell vehicle sales were 4,102 units, down 27.2%. Full-year 2025 was 16,011 units, up 24.4% — same analyst house, opposite direction — driven by one manufacturer's recovery and a December surge in China tied to a tax-exemption transition, while the United States fell 37.7%, Japan 37.3% and Europe 23.1%. The defensible statistic is not the trend but the ratio: 16,011 fuel-cell vehicles against 20 million electric cars sold in 2025, roughly 1,250 to 1, with electric cars at 25% of all car sales and a projected 23 million in 2026. Global fuel-cell vehicle stock is about 130,000.

“The problem with hydrogen cars is that nobody built the stations.” Established In the one jurisdiction that did build them, the regulator's own assessment is that “statewide fueling capacity exceeds demand”: 14,128 registered light-duty fuel-cell vehicles against network capacity for about 34,300, at roughly 60% availability across 50 open stations with 11 more temporarily out of service. The retailer that closed all seven of its Californian light-duty stations in February 2024 cited supply complications and market factors, with retail hydrogen at $34.11 per kilogram. The chicken-and-egg framing has the causality backwards here; the vehicles did not arrive to use the stations that were built.

“Shipping is switching to green fuels.” Established First-half 2026 vessel orders were 73 LNG, 55 LPG or ethane, 4 ammonia, 2 methanol, 1 hydrogen out of 137 — so ammonia and methanol together are 6 orders, 4.4%, against 128 (93%) for fossil LNG and LPG. The detail the category name conceals: LPG and ethane are fossil fuels and grew from 15 orders to 55 year on year, so “alternative fuel orders” reporting is recording a shift toward fossil fuels. Two trade reports of the same classification-society dataset disagree on the prior-year comparator, 155 against 151, and split methanol from ethanol differently, 2+2 against 4; this brief states no year-on-year percentage and does not resolve the discrepancy. Full-year 2025 alternative-fuel orders were down 47% on 2024, and the maritime regulator's delay of a greenhouse-gas pricing mechanism to October 2026 is the stated deferral driver.

“A methanol-capable orderbook is hydrogen-derived fuel demand.” Handwave It counts hulls, not molecules. More than 60 methanol-powered vessels were operational with nearly 300 on order in the 2025 review, but a methanol-capable engine runs perfectly well on fossil methanol. Without matched green methanol supply contracts — which the pack did not verify and this brief therefore does not claim — the orderbook is evidence about engine optionality, not about hydrogen demand.

“Hydrogen is zero-emission at the point of use.” Established It is zero-carbon at the point of use and it is not climate-neutral. Hydrogen's hundred-year global warming potential is 11.6 ± 2.8 and its twenty-year potential 37.3 ± 15.1 through hydroxyl depletion, and value-chain leakage “has not been measured”, with estimates spanning 0.2% to 20%. The near-term framing is the one usually dropped: the twenty-year figure is roughly three times the hundred-year one, so a carbon-dioxide-equivalent accounting on a century basis understates hydrogen's forcing over exactly the decades in which the emissions would occur.

“Green hydrogen is the cleanest colour on every measure.” Established Not on leakage, on the only estimates available. Electrolytic production is estimated at 0.03–9.2%, against 0.5–1.0% for grey and 0.0–1.5% for blue, and liquid-hydrogen handling at 2–20% is the leakiest step in any chain. These are assumption ranges rather than measurements, which is the point: the colour with the widest uncertainty is the one being built.

“Natural hydrogen changes the economics.” Frontier Possibly, and not yet, and not on the numbers usually quoted. The headline comparison to proven natural gas reserves is resource-in-place modelling that the surveying agency itself qualifies as largely unrecoverable. Beware a specific conflation: reported concentrations of 96% or 98% are sample purity, not flow rate, and purity was never the constraint — Mali's producing well is 98% pure and still two orders of magnitude short of commercial flow. One independent critique totals all claimed discoveries at roughly 2.47 Mt a year, about 2.5% of current hydrogen demand, of which 93% is a single French basin that is modelled potential rather than a reserve; that critique is commentary with numbers rather than a measurement, and is flagged accordingly.

“One operating project ran at 20%, so global electrolyser utilisation is about 20%.” Frontier That is one plant, in 2023, reported by one analyst in August 2024, and the operator said the problems would not be resolved until late 2025 — an outturn the pack did not verify. No fleet capacity-factor dataset exists in either of the last two global reviews. The honest position is that the utilisation of the world's electrolyser fleet is unknown, and a brief that turned one project into a global figure would be making the same error as the pipeline announcements it criticises.