Ceres Power

Phase A — Understand the business

Lens 1 · Company Overview

What it is. Ceres Power is an asset-light intellectual-property licensor of solid-oxide electrochemical technology. It does not manufacture at scale. It develops the cell and stack IP, proves it at pilot/demonstrator scale, then licenses the manufacturing rights to large industrials — collecting (a) upfront technology-transfer and engineering fees, and (b) per-unit royalties as licensees sell product. Think ARM-for-fuel-cells: the design house, not the fab. Founded 2001 (an Imperial College spin-out); ~24 years old.

The core technology — one platform, two directions. The "SteelCell" (and its 2025-gen successor, the Endura stack, a 10.8kW steel-substrate stack operating ~450–630°C) is a reversible solid-oxide cell:

• Run one way → SOFC (solid-oxide fuel cell): generates electricity from natural gas, biogas or hydrogen at high electrical efficiency. This is the power-generation product — and the one now aimed squarely at AI data-centre / on-site power.
• Run the other way → SOEC (solid-oxide electrolyser cell): consumes electricity + steam to make green hydrogen at claimed <40 kWh/kg (~25% more efficient than low-temperature PEM/alkaline, because it exploits waste heat).

The steel-substrate design (a gadolinium-doped-ceria ceramic printed onto perforated steel) is the differentiator — it runs cooler than ceramic-only SOFCs, which the company argues means lower cost, better robustness and fuel flexibility.

How it makes money — the FY2025 shape:

• Revenue £32.6m (−37% YoY) — lumpy, because it is dominated by milestone-based licence + engineering fees that recognise in bursts (2024 was inflated by large Delta + DENSO tech-transfer recognitions that didn't repeat).
• Gross margin ~70% (£22.7m gross profit) — the signature of an IP model with negligible COGS.
• First-ever royalties: £110,000 — from Doosan beginning stack production. This single line is the whole investment debate (see Lens 5, 12, 13).

Customers/partners (the licensees ARE the customers): Doosan (Korea), Delta Electronics (Taiwan), DENSO + JERA (Japan), Weichai (China), Shell (India, SOEC/H₂), Thermax (India, SOEC), and — new in Mar 2026 — Centrica (UK, deployment channel). Bosch exited in early 2025 (Lens 3/8/13).

Contract structure. Upfront fees are one-time and front-loaded (good for near-term revenue, bad for durability); royalties are the recurring annuity that is supposed to compound as licensees scale — but which so far barely exists. There is no take-or-pay; Ceres bears none of the manufacturing capex (licensees fund the factories — e.g. Delta bought NT$6.95bn / ~£170m of land and plant) but also captures none of the unit margin beyond a royalty rate.

Lens 2 · Supply Chain

Ceres sits upstream of the value chain as a design/IP node, not a physical-goods node. The chain, named end-to-end:

Inputs → Ceres. R&D talent + IP (Horsham/Redhill UK), plus a small in-house pilot line for prototyping. Ceres consumes almost no bulk commodities itself — its "input" is engineering. The one physical dependency in its own stack design is steel substrate + gadolinium-doped-ceria materials, but at licensee scale that sourcing is the licensee's problem, not Ceres's.

Ceres → manufacturing licensees (the chokepoint layer). This is where the entire model concentrates:

• Doosan Fuel Cell (South Korea) — 50MW SOFC stack + system factory, opened Jun 2025, mass production ("start of production") Jul 2025.
• Delta Electronics (Taiwan) — global SOFC+SOEC manufacturing licence (~£43m deal, Jan 2024); pilot production targeted end-2026.
• DENSO (Japan) — SOEC stack manufacturing under licence; with JERA ran "Japan's first SOEC demonstrator," a ¥46bn (£220m) project with up to ¥35bn government subsidy.
• Weichai Power (China) — SOFC cell + stack manufacturing licence (signed Nov 2025), targeting AI data-centre power; revenue recognition begins H1 2026.
• Thermax (India) — SOEC stack manufacturing + pressurised-system licence.

Licensee → end customer. Doosan/Weichai/Delta manufacture stacks → integrate into power systems → sell to data-centre / commercial-industrial / grid buyers. Shell (India) is the hydrogen off-taker/deployer on the SOEC side; Centrica (UK) is a newly-signed deployment/EPC + services channel — origination, install, O&M, recycling — to route Ceres-based SOFC systems to UK/EU data-centre and C&I demand.

Chokepoints / single-source dependencies (critical).

1. The royalty depends entirely on the licensee's commercial success, which Ceres does not control. Ceres has done its job (transfer the tech); whether volume materialises is a Doosan/Weichai/Delta execution question.
2. Per-country exclusivity. The model appears to sign one manufacturing licensee per major geography — which caps competition for Ceres's IP within a region but also means if that one licensee stalls, Ceres has no fallback in that market.
3. Geographic concentration in Asia (Korea/Taiwan/Japan/China) — exposed to Asian industrial-policy subsidy regimes and, for China (Weichai), to geopolitical/trade friction. This lens is real, not generic: the names above ARE the supply chain.

Lens 3 · Competitive Advantages (moats)

The moat, stated fairly:

• Process IP + a genuine efficiency edge. The steel-substrate, ~500–600°C "Goldilocks" design is hard to replicate and delivers a claimed ~25% efficiency advantage over PEM/alkaline on the electrolyser side and high electrical efficiency on the SOFC side. The SteelCell won the MacRobert Award (2023) — UK engineering's top prize — which is real third-party validation of the technology (not the business).
• Reversibility (one platform, power AND hydrogen) is a structural product advantage — it lets a licensee amortise one manufacturing line across two end-markets.
• Embedded switching costs at the licensee level. Once Doosan/Delta tool a factory around Ceres's stack, ripping it out is a multi-year, multi-£100m decision — that is the stickiness that (in theory) turns fees into a royalty annuity.
• Blue-chip validation as a flywheel. DENSO, Bosch (formerly), Shell, Weichai, Centrica signing up is itself a marketing moat — each logo lowers the next partner's diligence cost.

Bargaining power — and why it's weaker than it looks (who needs whom more). Here the moat inverts. Ceres is a ~£1bn-cap company licensing to £10bn–£50bn industrials (DENSO, Bosch, Weichai, Shell, Centrica). In that relationship, the licensee holds the power: they fund the capex, own the customer, own the manufacturing know-how at scale, and can walk. Bosch did exactly that in early 2025 — exited after reportedly investing ~€400m, and sold its ~17% stake. A moat you can walk away from without owning anything is a weak moat. Ceres's power over its licensees is limited to IP exclusivity and the switching cost of a tooled line — real, but not dominant.

Versus a rival business model — Bloom Energy. Bloom is the anti-Ceres: vertically integrated, it makes and sells the boxes itself, keeps 100% of unit economics, and is now booking multi-GW AI-datacentre deals directly (Oracle up to 2.8GW; AEP up to 1GW; Brookfield $5bn). Bloom captures the whole margin and the whole customer relationship; Ceres captures a royalty sliver but with zero capital intensity. The bet is whether asset-light royalty scales faster than it dilutes — the model works spectacularly IF licensee volume compounds, and is a value trap if it doesn't.

Verdict on the moat: the technology moat is real and award-validated; the economic moat (durable pricing power, control of the customer) is thin and licensee-dependent. That gap is the entire risk.

Lens 4 · Segments

Ceres does not break out clean product-segment P&L in the way a diversified operator does — revenue is reported by type (licence / engineering-services / provision-of-technology / royalty), not by SOFC-vs-SOEC or by geography, and no segments.csv exists (research layer empty). What can be sourced:

• Revenue by type, FY2025 £32.6m: dominated by licence + engineering-services fees; royalty = £0.11m (0.3% of revenue). The other ~£32.5m is milestone/project work — inherently non-recurring.
• Directional trend: FY2024 revenue was materially higher (hence −37% in FY2025) because 2024 recognised big one-off Delta + DENSO transfers. This is the lumpiness signature — you cannot read a growth rate off any single year; you must look at contracted forward revenue (~£45m contracted for 2026 before new business) and the record £112.8m order intake booked around the Delta/DENSO/Thermax signings.
• By end-application: SOFC (power) is the near-term revenue + royalty engine (Doosan live, Weichai coming, Centrica channel); SOEC (hydrogen) is earlier-stage / demonstrator (Shell 1MW, DENSO-JERA, Thermax pilot) — real but pre-commercial.
• By geography: revenue and partnerships are Asia-weighted (Korea, Taiwan, Japan, China) with a UK/Europe deployment layer now added via Centrica.

Hard requirement honoured: no segments.csv to source from → all figures ``, and I am not fabricating a SOFC-vs-SOEC revenue split the company does not disclose — n/a — not disclosed by segment.

---

Phase B — Measure performance

Lens 5 · Earnings Result (FY2025, reported 26 Mar 2026)

All `` unless noted:

vs consensus: UK small-cap; no clean quarterly consensus beat/miss line as with US names — revenue is milestone-driven so "the print" is less about a consensus surprise than about order intake and royalty inception. The market's focus was (correctly) the £110k first royalty and the ~£45m contracted-2026 number, not the headline revenue decline.

What drove it. Revenue fell purely on timing (2024's big transfers didn't repeat). The structurally important events were (1) Doosan reaching production → first royalties, and (2) burn cut from £35.9m → £20.1m via restructuring (op-cost −20% target for 2026).

Balance-sheet flags. No debt stress (net cash); the concern is the direction — EBITDA loss widened even as burn fell (the difference is non-cash / exceptional items and revenue mix). Post-year-end, the company raised £103m (Jun 2026, see Lens 8) — so pro-forma cash ≈ ~£180m. That is a ~5–7 year runway at ~£20m burn — genuinely long, and the single strongest bear-defeater.

Market reaction / context. The stock had already run ~450–970% over the prior 12 months into these results; results themselves were a "solid but not transformational" print against very high expectations. The volatility is in the expectations, not the current fundamentals.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts/ on disk (research layer empty); reconstructed from `` coverage of the FY25 call + prior updates:

• Management focus, most recent (FY25 call, Mar 2026): "cash, partnerships, long runway" was the explicit framing. The narrative has shifted decisively toward AI data-centre power — every new SOFC partner (Weichai, Centrica) is now framed around data-centre / on-site power demand, riding the same wave as Bloom. Two years ago the story was primarily green hydrogen (SOEC); today SOFC-for-datacentres is front and centre. That pivot is a tell — management is following the money (and the multiple) into the AI-power narrative.
• What they now say a lot: "asset-light," "royalty inflection," "multi-GW," "data centre," "one more major licence = breakeven."
• What they've stopped saying / gone quieter on: the grand hydrogen-economy TAM language of 2021–22, and specific near-term hydrogen volume timelines (SOEC slipped to demonstrator-stage reality).
• Tone trajectory: from "transformational hydrogen future" (2021, promotional) → "disciplined path to breakeven, cash-generative by 2027" (2026, more grounded). The de-hyping of hydrogen and re-hyping around AI power is the dominant sentiment shift. Credibility improved on cost discipline; still promotional on royalty timing.

Lens 7 · Comps

Ceres is a pre-profit IP licensor, so earnings multiples are meaningless (loss-making) — the honest frame is EV/Sales and a qualitative business-model contrast. Peers are the global solid-oxide / fuel-cell set.

Lens 8 · Stock-Price Catalysts (moves >5% over recent years)

Ceres is a high-beta narrative stock; the >5% movers reveal what the market actually trades on [all web]:

• ↓ 2024: Bosch collaboration termination (announced early/Feb 2025; Bosch ceased SOFC development to focus on PEM). Shares had collapsed to <60p on the loss of the anchor partner + a 17% overhang holder. The market reacts hardest to partner departures — because the partners are the business.
• ↑ Jul 2025: Doosan mass-production start → shares surged (~44% on the print; the broader run associated with Doosan chatter was far larger — some sources cite a ~250% move over the surrounding period). Partner commercialisation is the single biggest up-catalyst.
• ↑ Apr 2026: Goldman Sachs target 530p→670p → ~23% jump. Broker upgrades move it materially — a thin-float, high-conviction name.
• ↑ Nov 2025: Weichai China licence → up-move on China/AI-datacentre optionality.
• ↓ Dec 2025: Grizzly Research short report → shares slid on the "flawed business model" thesis.
• ↑ Mar 2026: Centrica multi-GW UK partnership + ↑ 2026 UBS "breakeven next year"/970p target.
• ↓ Jun 2026: £103m placing at a 6.5% discount → shares dipped on dilution, then partly recovered as the market re-rated the AI-datacentre opportunity.

Pattern: the market trades Ceres on (1) partner sign/exit events, (2) partner commercialisation milestones, (3) broker targets, and (4) the AI-datacentre-power narrative — NOT on reported earnings (which are lumpy and ignored). It is a story/catalyst stock with a ~10× 12-month range (87.8p–871.5p). Anyone holding it is trading catalysts, not a P&L.

---

Phase C — Judge people & books

Lens 9 · Management

• CEO — Phil Caldwell. ~18 years in the fuel-cell industry (previously Intelligent Energy; earlier ICI). Long tenure at Ceres; he is the architect of the asset-light licensing pivot — the strategy that took the company from a cash-burning would-be manufacturer to an IP licensor with blue-chip partners. Track record, fairly stated: he has signed the partners (Doosan, Delta, DENSO, Weichai, Shell, Thermax, Centrica) and cut the burn (£35.9m→£20.1m) — genuine, quantified execution. The unfinished part: he has not yet delivered material royalties — the model's whole point — after ~a decade of the strategy. He is a competent commercial/strategic operator; the jury is out on whether the model he built actually monetises.
• CFO — Stuart Paynter (appointed 1 Oct 2024). Strong CV: ex-CFO of Oxford BioMedica (7 yrs, grew revenue >5×), 8 yrs at FTSE-100 Shire, roles at Steris and De La Rue; Chartered Accountant, Physics degree (Imperial). A credible, scale-up-experienced finance chief — relevant that his OXB experience was taking a platform-IP/bio-manufacturing story to commercial revenue, which is precisely Ceres's challenge. Note the CFO churn: Eric Lakin left after ~3 years (Oct 2024) for "other interests"; Stuart Paynter is the replacement. One clean CFO transition — not a red flag by itself, but worth monitoring.
• Skin in the game / insider ownership: director participation was visible in the Jun 2026 raise (directors subscribed ~31k shares) — modest. No insider-transactions.csv on disk; I will not fabricate an ownership %. n/a on precise insider stake.
• Capital allocation. The strategy is deliberately capital-light — push capex onto licensees, spend Ceres's own cash on R&D (£48.6m) + pilot lines. Historically the company has funded itself through repeated equity raises (the Jun 2026 £103m is the latest; the register also absorbed the Bosch 17% stake exit). This is textbook pre-profit deep-tech: dilution is the funding model until royalties inflect. ROE/ROIC are negative and will stay negative until breakeven (UBS: EBITDA/cash breakeven possible 2026, cash-positive 2027 — conditional on one more major licence).
• Red flags (management-specific). (1) A long pattern of high-profile partnerships that generated fanfare but little/no product — Grizzly names KD Navien (2013), Cummins (2016), Nissan (2016), Honda (2016) as historical examples; this is the most damning management critique and it is about the announcement-to-revenue conversion rate. (2) Promotional framing — the pivot from "hydrogen future" to "AI data centre" tracks whatever narrative supports the multiple. Balanced view: Caldwell has clearly professionalised and de-risked the company (real partners, real factories now exist — Doosan's 50MW plant is not vaporware); but the team's credibility on converting logos into royalties is unproven and is the crux.
• Archetype: professional-manager-led (not founder-led), appropriate for a licensing/commercialisation stage — this is a "sign partners and cut costs" management, which is what the stage needs, provided the royalties come.

Lens 10 · Forensic Red Flags

Forensic lens — every figure `` (no filings on disk).

• Revenue recognition — the central accounting question. Ceres's revenue is milestone/tech-transfer + engineering fees, recognised in lumps as contract milestones are hit. This is legitimate under IFRS 15 but inherently flattering in signing years and hollow in gap years (2024 high → 2025 −37%). The risk is not fraud; it is that headline revenue overstates the recurring, durable business, which is really just £110k of royalties. Watch: how much of "revenue" is one-off vs recurring each year.
• Royalty vs fee mix. With royalties at 0.3% of revenue, ~99.7% of the top line is non-annuity. A forensic analyst reads this as "the business has not yet been proven to exist" in the recurring sense — 24 years in.
• Cash flow vs earnings. Operating cash burn (£20.1m) is lower than the EBITDA loss (£32.5m) and far below operating loss (£47.6m) — the gap is working-capital timing + non-cash (SBC, exceptionals, D&A on the pilot line). No sign of earnings being flattered by aggressive capitalisation of R&D relative to the burn — if anything the reverse (heavy expensed R&D). This is cleaner than many pre-profit deep-techs.
• Related-party concern (specific). Grizzly alleges Doosan's initial 9MW order is with a related party, implying the "commercialisation" is partly internal rather than arms-length end-demand. This is the highest-value forensic flag — if true, the first royalties are less validating than they appear. Unverified against primary filings (none on disk).
• SBC / dilution. Serial equity raises + option programs mean dilution is structural (share count 195.8m → 213.8m after the Jun 2026 raise, +~9%). Non-GAAP/adjusted EBITDA excludes SBC and exceptionals — standard, but it means "adjusted" flatters the true cash economics less than usual here because the cash burn is separately disclosed and reasonable.
• Goodwill/intangibles: low risk — this is an organically-developed IP company, not a roll-up; minimal acquisition goodwill.

Regulatory findings (required sub-section; per regulatory/regulatory-findings.md):

• SEC (EDGAR LR + AAER): None possible — Ceres has no CIK and does not file with the SEC. total_sec_findings: 0 by construction, not by exoneration.
• Non-SEC enforcement (web search — FTC/DOJ/FCA/consent-decree/fine/penalty): No material regulatory enforcement actions found against Ceres Power in web search as of 2026-07-06. It is an LSE-listed (formerly AIM, now Main Market/FTSE 250) UK company under FCA jurisdiction; no FCA enforcement, fine, or investigation surfaced.
• Short-seller report (not a regulatory action, but material): Grizzly Research published a short thesis on 11 Dec 2025 (disclosed short position) alleging a "flawed business model" and misleading commercialisation claims. This is an analyst opinion, not an enforcement finding — but it is the most material public challenge to the accounts/narrative and Ceres had not issued a formal point-by-point rebuttal as of the available coverage.
• Item 3 / Legal Proceedings equivalent: no UK annual-report "material litigation" disclosure ingested (no filings on disk) → n/a from primary filing.
• Summary: No material regulatory or legal enforcement findings — verified via SEC EDGAR EFTS (n/a, no CIK), web search (clean), as of 2026-07-06. The one open forensic item is the Grizzly related-party allegation on the Doosan 9MW order, which is unverified against primary sources.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (EPS, FY2026–FY2028)

Ceres is loss-making and will remain so on a statutory basis near-term; the meaningful projection is path-to-breakeven + royalty trajectory, not EPS. I build to EBITDA/cash rather than pretend-precise EPS, and label every input.

Base inputs:

• FY26 contracted revenue ~£45m before new business. Assume modest new-business + Weichai H1-26 recognition → FY26 revenue ~£45–55m.
• Gross margin held ~70% → FY26 gross profit ~£32–38m.
• Op-costs −20% YoY on ~£70m base → ~£56m.
• Cash burn guided to ~£20m and falling.

Scenarios (statutory EPS on ~214m shares):

Anchor to third-party model: UBS explicitly models EBITDA/cash breakeven in 2026 IF Ceres signs ≥1 more major manufacturing licence, positive FCF 2026–28, a £50m cash buffer held in 2026 → cash-positive 2027, and cumulative SOFC shipments 1.8GW by 2030 → >£100m royalties in 2030 alone. That 2030 royalty line, if realised, is what justifies today's ~£1bn cap; it is a conditional, out-year number.

Base call (for the record, NOT logged per --watchlist rule): Ceres reaches EBITDA breakeven in FY2027 (one year later than the bull "2026-if-a-licence" case), on ~£60m revenue, contingent on Doosan/Weichai royalties scaling AND one further major licence. Statutory net loss persists through FY2026. `` — arithmetic above. Per skill --watchlist rule, no forecast.ts create run.

Lens 12 · Bull vs Bear

Bull case. Ceres is the ARM of solid-oxide power arriving exactly as AI data-centres create a step-change in on-site power demand that the grid cannot meet fast enough. It has assembled a portfolio of blue-chip manufacturing licensees across every major industrial geography (Doosan/Weichai/Delta/DENSO/Shell/Thermax) plus a UK/EU deployment channel (Centrica) — capturing the AI-power TAM with zero factory capex of its own and ~70% gross margins. Doosan's 50MW plant is live and producing; first royalties have inception; the fee-to-royalty flywheel is starting. Burn is halved, £180m pro-forma cash gives a 5–7yr runway, and management guides to cash-positive by 2027. If UBS is right about 1.8GW cumulative shipments → >£100m annual royalties by 2030 at ~90% incremental margin, the current ~£1bn cap is a fraction of intrinsic value — this is a 3–5× from here. Secular tailwind (AI power + industrial decarbonisation) + asset-light margins + optionality on the SOEC/hydrogen leg = the bull's dream setup.

Bear case (2–3 things that could permanently impair).

1. The royalty never scales — the £110k stays a rounding error because SOFC end-demand is subsidy-driven, lumpy, and small, and licensees (who own the customer) prioritise their own economics. 24 years, ~a decade of licensing, and £110k of royalties is the base rate; the burden of proof is on the bulls.
2. Licensee flight / the moat is walk-away-able. Bosch already left after ~€400m. If a second major licensee stalls or exits (Weichai "routine business," per Grizzly; or Doosan's flat 2026 volume), the model's premise breaks — Ceres owns no factories, no customers, only IP others can decline to scale.
3. Expectations are baked in at ~28× sales. The stock has run ~450–970% in a year on narrative, not delivered royalties. Any royalty disappointment de-rates hard (the Dec-25 Grizzly slide and Jun-26 placing dip show the reflexivity).

Pre-mortem (it's 18 months out, thesis broke — what happened?): Doosan's 2026 volume came in flat (as Doosan itself guided), Weichai's licence stayed "routine" with no meaningful production, no new major licence was signed (so the UBS breakeven trigger missed), royalties limped to a low-single-£m, and the AI-datacentre-power narrative rotated to Bloom/FuelCell (who actually ship boxes). Ceres burned through part of the £180m, guided to another raise, and the multiple compressed from ~28× toward the ~5.5× FuelCell level — a 60–80% drawdown. The Grizzly thesis was directionally right on timing.

Are multiples too high? On FY25 actuals, yes, extremely (~28× sales, ~99.7% non-recurring revenue). On UBS's 2030 royalty model, no. The stock is a pure bet on the slope of the royalty curve over 2026–2028. It is not a value stock at any conventional multiple.

Contrarian view (what the market refuses to see): Bulls extrapolating Bloom's AI-datacentre bonanza onto Ceres are conflating two different businesses — Bloom captures 100% of a box's economics; Ceres captures a royalty sliver and doesn't control whether the box sells. The market is paying a higher multiple for the weaker economic position because "asset-light + 70% margin" screens beautifully — but a 70% margin on £110k is worth nothing, and the licensing model's structural ceiling (one licensee per geography, licensee owns the customer) may cap royalties far below the £100m dream. The contrarian read: the technology works and will be deployed; Ceres just may not be the entity that captures most of the value from it.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case (channelling and stress-testing the Grizzly thesis):

• What structurally breaks the money-making? The model only works if royalties compound. After 24 years and a decade of licensing, cumulative royalties are £110k. The bull case requires a ~1000× scaling of the one line item that has never worked. That is not a de-risked annuity; it is a hope.
• Where is revenue concentrated, and what if it shifts? ~99.7% of FY25 revenue is one-off fees, and forward revenue is concentrated in a handful of licensees' willingness to keep paying milestones. If milestone signings pause (as in 2025's −37%), revenue falls off a cliff — there is no recurring base to cushion it.
• Why the moat is weaker than bulls think. Ceres licenses to industrials 10–50× its size who fund the capex and own the customer. Bosch — the anchor — walked, after ~€400m. A moat your biggest partner abandons at a ~€400m sunk cost is not a moat; it is a marketing relationship. Per-geography exclusivity cuts both ways: no upside optionality if the chosen licensee underperforms.
• Most dangerous competitor bulls underestimate: Bloom Energy — not as a technology threat but as the entity actually capturing the AI-datacentre-power TAM (Oracle 2.8GW, AEP 1GW, Brookfield $5bn), vertically integrated, keeping all the margin. Every GW Bloom ships direct is a GW that doesn't need a Ceres-licensed box. PEM electrolysis (where Bosch redeployed) is another flank on the hydrogen side.
• Worst capital-allocation / governance items: serial dilution as the funding model; a history of fanfare-then-nothing partnerships (KD Navien, Cummins, Nissan, Honda); Grizzly's related-party allegation on the Doosan 9MW order (if the first "commercial" order is internal, the validation is hollow). Korean Clean-Hydrogen Power Scheme cancellation removed a demand channel.
• Assumptions that must hold for today's price: (1) royalties inflect from £0.1m toward £100m by 2030; (2) ≥1 more major licence signs (UBS's breakeven trigger); (3) Doosan/Weichai actually produce at volume (Doosan guided flat 2026); (4) the AI-power narrative keeps the multiple aloft while the P&L catches up.
• If growth disappoints 20–30%: because the valuation is ~28× sales on optionality, a royalty/licence disappointment doesn't cut EPS 20–30% — it breaks the narrative and re-rates the multiple, plausibly a 50–80% drawdown toward peer EV/Sales.
• Single scenario that permanently impairs: a second major licensee exit or public repudiation (a "Bosch #2") — it would confirm the "logos don't convert" thesis, collapse the flywheel premise, and there is no owned asset to backstop the equity. Plausibility: moderate — Bosch already set the precedent; Weichai's own "routine business" characterisation is a yellow flag.

Lens 14 · Management Questions (ordered by information value)

1. Of FY2025's £32.6m revenue, exactly how much is recurring royalty vs one-off licence/engineering fees — and what is the FY2026 and FY2027 royalty-only forecast, partner by partner?
2. Is Doosan's initial 9MW order arms-length end-demand or a related-party transaction, and what is Doosan's contracted (not aspirational) production volume for 2026 and 2027?
3. What specific, dated milestone would trigger the "one more major manufacturing licence" that UBS/management say unlocks EBITDA breakeven — and what is the pipeline of named counterparties?
4. Why did Bosch exit after ~€400m, in your account, and what changed structurally so the next anchor licensee won't reach the same conclusion?
5. What is the per-unit royalty rate (or £/kW) you earn on a licensee's SOFC/SOEC sales, and how does it change at scale?
6. Doosan stated no new orders in 2025 and flat 2026 volume — do you dispute that, and if not, how does the royalty ramp to >£100m by 2030?
7. What is your honest reply to the Grizzly Research thesis — point by point, especially the "£110k after 24 years" and related-party claims?
8. Weichai reportedly called the licence "routine business" not worth exchange disclosure — how do you reconcile that with it being framed to your shareholders as a major catalyst?
9. What are the contractual minimums / take-or-pay (if any) protecting Ceres if a licensee under-produces, versus pure success-based royalties?
10. On durability/degradation for data-centre "five-nines" duty — what are the demonstrated stack lifetime and degradation rates at licensee scale, independently verified?
11. What is the cash runway with zero new licences and current burn, and at what point do you raise again?
12. How do you compete for the AI-datacentre TAM against Bloom, which captures 100% of unit economics — why does a hyperscaler route through a licensee-made Ceres box rather than buy Bloom direct?
13. What proportion of the SOEC/hydrogen pipeline is subsidy-dependent, and how exposed are you to schemes being cancelled (e.g. Korea's Clean-Hydrogen Power Scheme)?
14. What is insider ownership across the board/exec team, and what recent open-market (not placing) purchases have insiders made?
15. Over the next 3 years, what is the single metric by which you want to be judged — and what number on it constitutes success vs failure?

---