Claude Backgrounder. Doublecheck rverything before using. What it is: a long-duration storage IPP created by a battery manufacturer to buy its own product, controlled by that manufacturer’s secured creditor, and made financeable by a Lloyd’s insurance wrap standing in for an operating record the technology doesn’t yet have.
The disclosure that reorders everything
From the May 13 formation release, under “Anchored by Institutional Capital”:
Cerberus will receive Eos warrants, as well as controlling equity in Frontier Power USA in exchange for its commitment.
Cerberus put in $100 million. Eos put in roughly $113 million. Cerberus controls.
Layer in what Cerberus already held. Under the June 2024 strategic investment — a $210.5 million delayed-draw term loan plus a $105 million revolver drawable at Cerberus’s sole discretion — Cerberus received penny warrants and non-voting preferred stock. As of the January 2025 8-K, those securities were exercisable or convertible into 158,433,112 shares, an applicable percentage of 33.0% fully diluted, with a contractual path toward 49% tied to operational milestones. CCM Denali Debt Holdings is administrative and collateral agent. (Issuances in November 2025 and the mid-2026 equity raises will have moved that percentage; the current figure needs the latest 10-Q or proxy.)
So Cerberus is Eos’s secured lender, a large economic shareholder, and now the controlling equity holder of the affiliate that constitutes roughly half of Eos’s backlog — and it collected additional Eos warrants for the commitment.
Against that, Joe Mastrangelo in the same release:
…while ensuring project capital is governed independently and on arm’s length commercial terms.
Independent of whom is the question the release doesn’t answer. Cerberus sits on both sides of the transfer price, both sides of the take-or-pay, and both sides of any workout. That’s a description of a disclosed structure, not an allegation. But “arm’s length” is carrying more weight in that sentence than the facts support, and anyone quoting it should say so.
What the platform is
Frontier Power USA Parent, LLC closed August 4, 2026, with approximately $263 million of gross equity: ~$113 million Eos, $100 million Cerberus (CCM Frontier JV Holdco), $50 million Hudson Bay (HBC MSF Capital Solutions Blocker II). With project debt at roughly 75% LTV, Eos says that supports more than $1 billion of deployable project capital. KKR Capital Markets is engaged to structure the long-term project financing.
Four pieces make it a system.
A 2 GWh take-or-pay capacity reservation agreement. FPUSA is contractually obligated to buy 2 GWh of Eos Z3. Eos books firm backlog; FPUSA gets guaranteed manufacturing slots rather than merchant procurement.
Technology performance insurance from Ariel Green. The load-bearing innovation, and the May release describes it most precisely: 15-year non-cancellable, sized at the project level, multi-year aggregate capacity up to approximately $1.5 billion, written through a Lloyd’s of London consortium rated A+/AA-. The stated purpose is explicit — the wrap will allow project debt to achieve investment-grade characteristics at competitive terms.
Read plainly: an insurance balance sheet substituting for an operating track record. Lenders will not underwrite twenty-year degradation risk on zinc-halide chemistry with a thin field record at conventional spreads. They will underwrite a Lloyd’s consortium. How thin that record is: per DOE data cited by Utility Dive, most operational zinc battery deployments have power ratings of 1 MW or less. Eos’s largest committed single deployment is 200 MW.
Ariel Green has placed TPI on storage before — on a 2.2 GWh Fluence-equipment procurement by Excelsior Energy Capital alongside DNV technical validation, and in July 2026 on the first UK storage TPI policies, for Pulse Clean Energy’s Plymouth and Dowlais projects. The Eos framework is larger and structurally more central to the business model.
Direct OEM access. The May release claims access at the cell IP level, through module and system architecture to the proprietary software layer (DawnOS) — the pitch being that vertical integration eliminates the seams where third-party BESS integrators carry risk.
Cash-flow recycling. Operating project cash flow reinvested to originate more projects, which buy more Eos equipment.
Where the model comes from
The May release names its own genealogy:
Manufacturers in capital-intensive industries, from aviation to power generation, have long invested into or partnered with development execution vehicles to accelerate customer deployment.
That is vendor finance — a manufacturer financing acquisition of its own equipment.
Who’s running it matters. Joe Mastrangelo spent 25 years at GE: entered through the Financial Management Program and Corporate Audit Staff, a decade in GE Oil & Gas, corporate officer in 2008, CEO of GE Power Conversion, and President and CEO of Gas Power Systems from 2015, per his Eos board biography. Nitin Gupta, FPUSA’s Head of Capital Formation, came up through GE Energy Financial Services. The vehicle is a GE structure built by GE people, and the release says so.
What the Q2 numbers say
Eos reported record Q2 2026 revenue of $68.8 million, up 351% year over year. The income statement splits it:
- Third-party revenue: $13.7 million (Q2 2025: $15.2 million)
- Related-party revenue: $55.0 million (Q2 2025: zero)
Third-party revenue declined year over year. All of the reported growth came from a single pre-existing project financed by a Cerberus affiliate and contributed into FPUSA at closing. Eos discloses this directly — approximately 80% of Q2 revenue — and most coverage skipped it.
Backlog hit a record $807 million (3.4 GWh). As of June 30, that project plus FPUSA represented 49% of backlog volume.
Gross margin was negative 71%. Adjusted EBITDA loss $71.4 million. Operating cash burn of $191.8 million in the first half; cash down from $568 million to $305 million. The related-party volume isn’t margin — every cube shipped loses money. What FPUSA buys Eos is fixed-cost absorption at Thorn Hill, a reason to run Line 2, and an installed base that might convert third-party buyers.
That is a real industrial strategy. It is also, in the short run, unfalsifiable, because the customer is controlled by the creditor.
Two numbers the releases don’t foreground.
The rights offering. Targeted approximately $150 million. Raised $37.7 million gross. Eos reached $263 million by adding Hudson Bay’s $50 million and a separate equity offering. The July 23 release truthfully says the $250 million aggregate target was exceeded; existing shareholders declining to fund roughly three-quarters of what was asked of them appears nowhere in the headline. Note that the July 7 Stella conversion was expressly contingent upon the successful closing of that rights offering — it survived because institutional money backfilled retail’s shortfall.
The pipeline figures don’t reconcile. May 13: approximately 5 GWh under active development, with a further 20 GWh of identified pipeline. July 23 and August 5: approximately 16 GWh of pipeline opportunities, with ~5.0 GWh acquired, selected, or in diligence. The active tranche held. Depending on whether May’s 20 GWh was additional or total, the broader number either shrank or was redefined. No explanation given.
The asset base
Everything closed or selected is ERCOT, and all of it is four-hour.
- Redbird — 100 MW / 400 MWh. Project developed by Bimergen (NYSE American: BESS), and previously by Bridgelink. First PO under the CRA, June 2026. FPUSA funds 100% of construction equity; Bimergen retains a minority interest.
- Two Texas 10 projects — same Bimergen conversion; 80 MWh between them, 480 MWh with Redbird.
- Blanquilla — 200 MW / 800 MWh, from Stella. $100 million Phase I purchase order, post-quarter.
- Aransas Pass, Nash, Wallis — balance of the 230 MW / 920 MWh Stella selection, July 7.
- Wildfire — 400 MWh, second Bimergen selection, July 14.
Closed or selected: ~1.8 GWh, roughly 90% of the 2 GWh reservation.
Stella holds an unusual position: FPUSA has an exclusive Selection Right over Stella’s late-stage pipeline of more than 2 GWh, and Stella is designated execution partner through commercial operation. Stella states it has developed, built, and operated more than 2 GW and 4.4 GWh of utility-scale storage. Per FPUSA’s own release, Stella’s principals worked with Mark Klein and Nitin Gupta at Broad Reach — so the execution partner is a decade-old relationship with FPUSA’s development and capital leads.
The leadership slate, read against the record
Everything in the September 8 release about aggregate track record — the “over 20GW developed” figure, the Klein/Gupta–Stella relationship, the Broad Reach characterizations, Jay Legault’s Ontario Power Generation background — is company-stated and unverified here. Treat as directional.
Klein and Gupta are the substantive hires. Broad Reach Power was founded in 2019 with EnCap backing; ENGIE agreed in August 2023 to acquire its battery storage business at an equity value in excess of $1 billion, with the deal excluding 1.8 GW of solar and wind and 4 GWh of Mountain West storage. FPUSA’s release claims a substantial portion of ENGIE’s present-day North American storage business traces to that platform, and that it remains the largest BESS player in Texas; neither is verified here.
Jay Bellows is where the release and the record diverge.
The September 8 release credits him as “most recently” CEO of KORE Power, plus Northern Reliability and Nomad. Accurate as written. What it omits: Bellows became KORE CEO in late January 2025, the same week the company cancelled the $1.25 billion KOREPlex gigafactory in Buckeye, Arizona — a project carrying an $850 million DOE conditional loan commitment that never closed, a roughly $10.4 million contractor lien from Yates Construction, and 3,000 promised jobs. He pivoted toward retrofitting existing sites. T1 Energy signed a definitive agreement on June 3, 2026 to acquire KORE Power at an enterprise value of approximately $32 million, plus a $9.6 million earn-out, and completed the acquisition in July 2026; it plans to rebrand the business T1 NRI.
Fair reading: the $32 million doesn’t value nothing. T1’s stated centerpiece was KORE’s NRI division — the integration business Bellows built before the merger, roughly 1,100 BESS deployments, expected to contribute $15–20 million of EBITDA in 2027. What collapsed was the gigafactory ambition, not NRI. Running a domestic cell manufacturer through 2024–25 was brutal, and Northvolt’s ending was worse.
But a release built entirely on leadership credibility that omits the terminal outcome of the credential it leads with is doing a job. Note it — and note that Bellows is the “CEO with a deep operating track record” the May release promised.
The hyperscaler question
Google is buying Eos long-duration storage. On September 2, 2026, Google, MN8 Energy, and Eos announced that Google will purchase the energy, capacity, and clean energy attributes of the Mammoth Solar project in Kanawha County, West Virginia — 86 MW of utility-scale solar on a reclaimed coal mine, paired with 70 MW / 280 MWh of lithium-ion and 10 MW / 100 MWh of Eos Z3 zinc-based storage. MN8 owns and operates. The project interconnects to PJM and is intended to serve Google’s regional data centers, including a planned West Virginia facility. It is Google’s first use of Eos technology and the first project under the MN8–Eos master supply agreement signed in October 2025 for up to 750 MWh. (Not to be confused with the much larger, unrelated Mammoth Solar in Indiana.)
Three qualifications matter.
It isn’t FPUSA. Mammoth is an MN8-owned project under an Eos master supply agreement. FPUSA has no announced role. FPUSA’s own portfolio — Redbird, Blanquilla, Wildfire, the Stella assets — contains no announced hyperscaler or data center offtake. Every FPUSA asset closed or selected is four-hour ERCOT storage monetized through conventional merchant channels.
The zinc doesn’t discharge until 2030. Reported commercial operation: solar 2028, lithium-ion 2029, Eos zinc 2030. A 2026 announcement of a 2030 asset is a procurement signal, not deployed capacity.
It’s 10 MW. Against Eos’s 3.4 GWh backlog and FPUSA’s 1.8 GWh committed, the Google deployment is small — notable for who signed it, and because most operational zinc deployments are 1 MW or under.
So the accurate framing is narrower than the marketing: a hyperscaler has underwritten Eos zinc storage, at modest scale, at the end of the decade, in a project FPUSA doesn’t touch. The May release’s claim that FPUSA’s Z3 units are expected to be deployed across AI data centers, and its listing of hyperscalers among expected investment-grade offtakers, remains forward-looking.
Utility Dive frames the PJM context as data center growth tightening capacity, pointing to recent auctions clearing at the price ceiling and a one-time backstop auction this fall covering a shortfall of nearly 7 GW. That is their attribution, and it intersects directly with the Reliability Backstop Procurement docket.
The international thread
Underweighted in most coverage. Ariel Green underwrites through a Lloyd’s of London consortium, making a UK insurance market the effective gatekeeper on whether US zinc storage gets investment-grade project debt. Gupta came to FPUSA from Voltwise Power, a UK and Germany developer backed by Sandbrook Capital. Eos signed a binding master supply agreement with CAPAC Energy for exclusive distribution across Germany, Austria, and Switzerland — initial 750 MWh, scalable to 2 GWh through 2031. And the platform Klein built was bought by a French utility. The domestic-content politics are American; the risk capital and a meaningful share of the demand are not.
Open questions
- Transfer pricing. At what price does FPUSA buy Z3 cubes from Eos? With Cerberus controlling FPUSA and holding Eos’s secured debt and a large converted equity position, the intercompany price determines where value lands. No disclosure found.
- Ownership split. Percentage interests among Eos, Cerberus, and Hudson Bay aren’t in any release. Controlling equity to Cerberus is stated; the cap table isn’t.
- Accounting treatment. Eos books FPUSA sales as related-party revenue rather than eliminating them, implying FPUSA is not consolidated. Confirm equity-method treatment and profit-deferral mechanics.
- DOE consent. The July binding term sheet listed DOE approval as a closing condition, tied to the LPO facility under which DOE took penny warrants in November 2025. The August 4 close implies consent; not confirmed.
- TPI terms. Attachment points, exclusions, claim triggers. Whether project debt actually prices at investment-grade levels is the test, and KKR’s placement will show it.
- Third-party conversion. The metric that settles it. It shrank in Q2.
Edited and authored by Dave with close collaboration by Claude.