Interview dossier · prepared 30 Jul 2026

They sell the fix for the thing that broke Ashburn twelve days ago.

On July 22, a transmission fault in Data Center Alley knocked roughly 3 GW of load off PJM in seconds. ON.energy sells a medium-voltage system whose entire purpose is to make that not happen. You live twenty minutes from the epicenter and you are interviewing for the job that prices the contracts.

Role
Senior Manager, Commercial Negotiations & Deal Structuring
Team
Finance (per job board)
Location
Reston, VA
Interviewer
Andrea Marsiglia
Talent Acquisition Partner
Format
30 min · MS Teams
00

The brief

Logistics first, then the one picture that summarizes the whole dossier.

Check your clock before anything else

The invite reads 4:00–4:30 PM (GMT−05:00) Bogotá. Northern Virginia on August 3 runs on EDT, GMT−04:00. Your local start is 5:00 PM ET, not 4:00 PM.

Fig. 0 — Time zone conversionConfirm on your own calendar
16:00 BOGOTÁ · UTC−5 21:00 UTC 5:00 PM NORTHERN VA · EDT (UTC−4) The invite time → Your actual time
If your phone's calendar shows 4:00 PM ET, the invite was built without conversion — email Andrea today. A one-hour miss on a 30-minute screen is unrecoverable.
Fig. 1 — The whole story on one lineMemorize this chain
AI boom $1T+ capex 2025–26 Grid stress ±70% GPU swings 3 GW Ashburn drop Regulation NOGRR 282 · NERC ride-through mandatory AI UPS™ inline MV system 5 GW Crusoe deal Your role contracts big enough to sink the company Every answer you give on Monday should connect back to some link in this chain.
The pitch in one breath: AI demand broke the grid's assumptions → regulators now require ride-through → ON.energy has the patented hardware → the Crusoe deal made the contracts enormous → they need someone to price the risk in them. That someone is the role.
Reading the markers in this document

Verified independent press or a regulator — safe to state as fact.  Company claim ON.energy's own materials — say "they say."  Live moving right now; recheck before you dial in.

01

The industry

Four forces define this market in 2026. Narrate these four fluently and you sound like someone who already works in the sector.

Fig. 2 — Demand has outrun the gridEPRI via DOE · Dominion · Morgan Stanley
DATA CENTERS AS % OF US ELECTRICITY 4% 2023 up to 9% 2030 (EPRI) WAIT FOR GRID POWER, 100 MW+ SITE (DOMINION) 4–7 years 3–4 yrs previous norm now Some sites reportedly quoted up to 17 years.
Why it matters commercially: power, not capital, is the binding constraint. Anything that de-risks or shortens interconnection commands a premium — that is ON.energy's wedge. Big-tech capex commitments exceed $1T across 2025–26 (Morgan Stanley). Verified
Fig. 3 — The two-front warThe core mental model of this entire market
THE GRID voltage dips · faults frequency excursions AI CAMPUS GPU clusters ramp ±70% in milliseconds AI UPS™ inline · always in the power path FAULTS BLOCKED SWINGS ABSORBED smooth power in clean power out Most products defend one direction. ON.energy's claim is both, simultaneously, from one box.
Front one: a grid fault can kill a training run worth millions of compute-dollars, so campuses trip offline to protect themselves. Front two: that self-protection — gigawatts vanishing at once — is itself what destabilizes the grid. The CTO calls this the two-front war. It is the correct frame for the whole sector.
Fig. 4 — Regulation arrived, fastERCOT · PUCT · NERC · FERC/PJM
ERCOT · NOGRR 282 ride-through MANDATORY for computational loads ≥ 75 MW EFFECTIVE 1 AUG 2026 NERC Level 3 Computational Load Alert urges all operators to follow; mandatory standard in the works PJM · CAISO · MISO · SPP moving the same direction; FERC docket EL25-49 rewriting PJM large-load rules now "Compliance became a gating item, not a differentiator." Say this sentence once, deliberately. It changes who has pricing power.
Nuance to know: Texas Industrial Energy Consumers has intervened, arguing ERCOT lacks authority to bind retail customers directly. ON's own guide expects the technical requirements to survive via a different implementation path — the tailwind is strong but not guaranteed. Live
Fig. 5 — Capital is chasing the picks and shovelsFT via Yahoo · MarketsandMarkets
ENERGY IPO PROCEEDS $4.3B ALL of 2025 $12.6B H1 2026 alone DATA CENTER UPS MARKET $8.8B 2025 $12.5B 2030 forecast
Translation for the room: ON.energy isn't fighting for existence — it's racing to convert a demand surge into signed, financeable, margin-safe contracts before the incumbents catch up. That race is literally the job you're applying for.
01B

The megaprojects — who is building what

These five build-outs define the customer landscape ON.energy sells into; each made a different bet on power, and that bet is the commercial story.

Fig. M1 — The gigawatt league tablePress figures · Jul 2026
TARGET CAPACITY (GW) — SCALE TO 10 0 5 10 Stargate program OpenAI+Oracle+SoftBank 10 GW target $500B committed program Meta Hyperion Louisiana 5 GW compute >$50B ON.energy × Crusoe AI UPS deal 5 GW the deal your role exists for AWS Project Rainier Indiana 2.2 GW $11B ph.1 + $15B ph.2 Microsoft Fairwater WI+GA network ~2 GW per campus at full build · $7B+ initial WI, reported >$100B total program xAI Colossus 1+2+3 Memphis ~1.2 GW+ est. $30–40B incl. funding rounds
Capacities mix IT load, facility load, and announced targets — never compare them as like-for-like in the room; say "announced scale" instead. Verified press figures as of Jul 2026.

Stargate (flagship: Abilene, TX)

Location: Abilene TX flagship (Lancium Clean Campus), plus Shackelford County TX, Doña Ana County NM, Lordstown OH, Milam County TX, Port Washington WI ("Lighthouse," $15B, Oracle+Vantage).

Parties: OpenAI, Oracle, SoftBank; Abilene developed/operated by CRUSOE; Nvidia $100B investment; announced Jan 2025 at the White House.

Scale: ~1.2 GW at Abilene; ~7 GW planned across sites; 10 GW / $500B program target by 2029; >$100B already deployed as of early 2026.

End user: OpenAI training + inference (Oracle Cloud Infrastructure).

Power strategy: West Texas cheap energy + behind-the-meter self-supply incl. up to ~300 MW on-site gas turbines to sidestep interconnection queues.

Why ON cares: Crusoe built Abilene — the same Crusoe that signed ON's 5 GW AI UPS deal in July 2026. This project is one degree from your desk.

Meta Hyperion (Richland Parish, LA)

Parties: Meta + Blue Owl Capital JV (~80% sold to Blue Owl, off balance sheet) + Entergy Louisiana.

Scale: 5 GW compute + 2.5 GW support/cooling; ~3,200 acres; up to 9 buildings; first 2 GW by 2030, full ~2032.

Cost: >$50B (was $10B at Dec 2024 groundbreaking, then $27B — quintupled in under two years).

Energy: Entergy building 10 gas plants (>7 GW, +30% to Louisiana's entire grid); Meta funds 2.5 GW renewables, ~240 miles of transmission, grid-scale BESS at three sites; "Ratepayer Protection Pledge"; Earthjustice challenging the financing at the PSC.

End user: Meta Superintelligence Labs.

Why ON cares: the single biggest example of "the customer pays for the grid now" — generation, transmission, storage all inside the deal. Every one of those line items is a negotiated contract.

xAI Colossus 1/2/3 (Memphis, TN)

Parties: xAI, MLGW/TVA (Colossus 1 grid power), own 1.2 GW gas plant across the Mississippi line for 2 and 3.

Scale: Colossus 1 ~300 MW, ~200k GPUs, built in 122 days in a former Electrolux factory; Colossus 2 ~946 MW IT power — largest known AI data center by IT power (Epoch AI).

Cost: est. $30–40B including recent funding.

End users: xAI's Grok — plus merchant compute: Anthropic leasing ~325k GPUs (~$1.25B/month reported), Google ~110k, the Pentagon.

Power strategy: mostly OFF-GRID behind-the-meter gas; 168 Tesla Megapacks smoothing GPU load swings. Environmental fight over unpermitted turbines.

Why ON cares: proof at scale that GPU swings need a battery buffer — Colossus solved with parallel Megapacks exactly the problem ON argues needs an inline system. That contrast is a sophisticated interview point.

AWS Project Rainier (New Carlisle / St. Joseph County, IN)

Parties: Amazon Web Services, built exclusively for Anthropic.

Scale: 1,200 acres, ~30 buildings, ~910 MW IT power operational, building toward 2.2 GW; 500k+ Amazon Trainium2 chips — the first hyperscale non-Nvidia cluster.

Cost: $11B phase 1 (operational late 2025) + $15B phase 2 announced.

Power strategy: bet on the utility grid (AEP Indiana), natural-gas backed.

Why ON cares: single-tenant, single-chip, grid-dependent — the maximum-exposure profile for exactly the ride-through risk ON prices.

Microsoft Fairwater (Mount Pleasant WI + Atlanta GA)

Parties: Microsoft; serves OpenAI + Azure AI.

Scale: WI campus >350 MW scaling toward ~2 GW, ~half a million chips; Atlanta ~636 MW IT power, second building to >700 MW by 2027; the two linked by dedicated fiber into one "AI superfactory" training a single model across states.

Cost: $7B+ initial WI; total program reported to potentially exceed $100B at completion.

Power strategy: the shocker — the Atlanta site runs straight off Georgia Power with NO UPS and NO diesel generators; Microsoft did the math on what redundancy buys during a training run and cut it.

Why ON cares: the bear case for ON's product, stated by the most sophisticated buyer alive. Know it exists; have a view (ride-through mandates like NOGRR 282 are precisely the counterweight — a regulator can't accept "we did the math").

How to use these in the room

Don't recite; deploy. One sentence pattern: "Every megaproject made a different power bet — Rainier bet on the grid, Colossus went off-grid with Megapacks, Microsoft cut the UPS entirely, Meta is buying the utility's build-out — and each bet creates a different contract surface. ON's thesis is that regulation is about to punish the naked-grid bets."

Figures move

Figures are press-reported as of July 2026 and move monthly — Grok prompt section 6 covers the refresh.

02

Ashburn, July 22 — your opening move

The single strongest thing you can bring to the call. It happened twelve days before your interview, thirty minutes from your house, and it is a live demonstration of ON.energy's entire thesis.

Fig. 6 — Anatomy of the eventDominion · PJM · Reuters · RTO Insider
Line fault Ashburn, VA — Data Center Alley Self-protection campuses' OWN systems trip to backup diesel 3+ GW vanishes ≈3% of PJM demand, gone in seconds Voltage ripple anomalies detected from DC to Chicago Stabilized ~10 minutes — vs milliseconds normally Roughly double the July 2024 event (~1.5 GW) that made NERC stand up a task force. The utility did not shed this load — the data centers jumped.
Where AI UPS fits this picture: an inline system rides through the fault instead of tripping. The campus stays on, the 3 GW hole never opens, the ripple never happens. The event is ON.energy's sales pitch, performed live, in their new office's backyard. Verified · aftermath Live

"The data centres didn't get knocked off the grid. They jumped. Every campus made a rational decision to protect its own compute — and the sum of those rational decisions is a three-gigawatt hole in PJM."

Suggested phrasing — say it in your own words
If asked "why ON.energy?"

"Two weeks ago three gigawatts of load dropped off PJM after a fault in Ashburn — twenty minutes from my house. I read the coverage, then read ON.energy's material on the inline architecture, and it was the first time I'd seen a product that addresses both directions of that problem instead of one. I've spent nine years selling energy systems, and I've never seen a market where the regulatory deadline and the customer's own self-interest point at the same purchase order. I want to be on the commercial side of that."

02B

July 10, 2024 — the first warning shot

Two years before Ashburn 2026, the same failure happened at half the scale, in the same place — and everything ON.energy now sells into traces to that night. NERC's own incident review is public; these details are verified.

Fig. W1 — Anatomy of the July 10, 2024 eventNERC incident review · Reuters · DCD
7:00 PM thunderstorm Lightning arrestor fails on 230 kV Ox-Possum near Fairfax, VA (Dominion) — line locks out Auto-reclose misfire 6 successive faults in 82s each cleared in 42–66 ms voltage sags to 0.25–0.40 p.u. ~60 data centers jump Protection logic → backup ~1,500 MW vanishes ≈ Boston's entire demand ~50× faster than plant trip Grid overshoots Frequency → 60.047 Hz (~4 min to settle) Voltage hits 1.07 p.u. Operators pull shunt caps Near miss, then response No blackout NERC incident review LLTF stood up Aug 2024 Level 2 Alert Sept 2024 Moura: grid not designed for 1,500 MW DC loss NERC's John Moura: "The grid is not designed to withstand the loss of 1,500 MW data centers."
The punchline nobody expected in 2024 — the danger wasn't data centers drawing too much power; it was data centers LEAVING too fast. Load loss, not load growth, almost broke the grid. Verified (NERC incident review, Reuters, DCD).
Fig. W2 — Escalation: 2024 → 2026 → nextNERC · Dominion · Jul 2026 event
LOAD DROPPED (MW) 1,500 MW ~60 facilities Jul 2024 NERC forms task force 3,000+ MW voltage ripple DC-to-Chicago Jul 22, 2026 double the scale, two years later Next one? Dominion: 40 GW of contracted data center power as of Dec 2024, up 88% in five months. The exposure keeps compounding — which is the sales pitch for mandatory ride-through.
The say-it-aloud line: "2024 was the warning, 2026 was the escalation, and the regulatory response — NOGRR 282, NERC's alerts, PJM's rulemaking — is the industry deciding not to wait for the third one."
Why this section earns its place on Monday

It converts the Ashburn story from anecdote into pattern. Pattern recognition is what separates "I read the news" from "I understand the market." One suggested sentence: "What struck me is that 2024 and 2026 were the same event at different magnitudes — same county, same failure mode, same protection logic. That's not bad luck, that's a design flaw with a growth rate. And it's why ride-through went from white-paper topic to mandate."

03

Who ON.energy is

The arc: LATAM integrator → US independent power producer → AI infrastructure technology company. Each phase left something in the DNA that shows up in how they do deals.

Fig. 7 — A decade in one picturePress releases · Energy-Storage.news · on.energy
capital & scale 2015–16 Founded Cooper + de Azevedo LATAM BESS · Miami HQ Jun 2023 $20M Series B Ultra Capital · Phalanx 300 MWh in ERCOT Apr 2024 $25M facility Lombard Odier — development capital Jan 2025 $77.6M credit Pathward + BridgePeak $250M+ raised to date Feb–May 2026 AI UPS™ launch Patent 12,614,920 NLR lab validation Jul 2026 CRUSOE · 5 GW multiple hyperscale campuses, 2026–27 Ten years of storage development and $250M+ of financing, then two step-changes in six months. Your role exists because of the last dot.
Identity to remember: unusually, they are developer + system integrator + asset owner in one company — historically retaining ~80% of the US projects they build. That triple identity changes what their contracts must do (Section 06).

The scale numbers they publish

All Company claim — cite as "their materials say."

3 GW+Operating & under construction
10 GW+Production capacity
75+Projects across the Americas
$1BProjects at financial close
$350MGrid-scale assets under management
$120MTax equity & ITC transfer capital
Fig. 8 — Three numbers that are NOT the same numberPrecision here is your job description
DEPLOYED (JD) 2.5 GW hyperscale campuses OP + CONSTR (SITE) 3 GW+ all segments ANNOUNCED (CRUSOE) 5 GW future
Installed ≠ contracted ≠ announced. These figures are different measures, not contradictions. Knowing the difference is exactly the precision a deal-structuring hire is expected to bring. Never blur them in conversation.

Hiring signal

The Greenhouse board carries 56 open roles across Reston/Tysons, Houston, Miami, Atlanta, Amarillo, Mexico City and Argentina. Two adjacent hires tell you the most: a Contracts Counsel (being built in parallel with you — hence "you don't wait for Legal") and a Director of Sales, Data Center & Critical Power (the "counterweight" relationship in your JD). Your role is filed under Finance: your deliverables go to the CFO, and your incentives are margin and risk, not bookings.

"We move fast, debate hard, and learn from each other. If you want predictable, this isn't it. If you want impact, welcome."

on.energy — working at ON.energy

Pair that with the JD's "we cannot teach commercial courage." They are screening for someone who will argue cleanly and stay warm doing it — a muscle you built over nine years in living rooms.

04

AI UPS, explained properly

You don't need to be an engineer. You need the architecture in one picture and its price in another. Here are both — plus the test data that backs the claims.

Fig. 9 — THE distinction: parallel vs inlineIf you internalize one diagram, make it this one
CONVENTIONAL: BESS (PARALLEL) GRID raw grid power DATA CENTER BESS beside the path has to detect → respond → TRANSFER TRANSFER TIME ≈ 4 ms — an eternity voltage already sagged; GPUs already saw the hit ON.ENERGY: AI UPS (INLINE) GRID AI UPS — IS the path AC→DC BATTERY DC→AC DATA CENTER every watt, all the time — load never touches raw grid power TRANSFER TIME = 0 — nothing to transfer always in circuit · medium voltage 13–35 kV · at the site boundary Bonus of medium voltage: equipment moves outdoors (freeing space for compute) and ONE system shields the whole plant — IT load, chillers, pumps — not just the server racks.
The sixty-second version: a BESS sits beside the power path and reacts; even a fast static switch takes ~4 ms, and the GPUs have already seen the disturbance. AI UPS is the power path — AC in, buffered through the battery, clean AC out, continuously. Transfer isn't fast, it's zero. This is why the CTO wrote "Don't Call Me BESS."
Fig. 10 — Ride-through: what the test showedRedrawn from ON.energy NLR test description · Jan 2026
1.0 p.u. 0 V GRID → 0 V (true ZVRT) LOAD HOLDS ≈1.06 p.u. — GPUs never notice ERCOT requires ≥150 ms · ON claims 1 hr+ on battery
Tested at the DOE's National Laboratory of the Rockies (ARIES/Flatirons): 7 MW grid simulator, 20 MW load simulator, full 3.2 MW branch at 13.2 kV. Company claim — but a national-lab venue makes it a strong one. Commercial angle to raise: whether a customer's acceptance test mirrors the NLR protocol or invents a harsher one is worth real money.
Fig. 11 — Transient absorption: the other directionRedrawn from ON.energy NLR test description
GPU LOAD: ±70% swings GRID SEES: a flat line
Integrated storage absorbs the swing with reportedly <1% battery state-of-charge variation. The grid gets a smooth, ramp-rate-compliant profile no matter how chaotically the workload pulls. Company claim
Fig. 12 — Spec sheet → contract clauseEvery spec is a term someone must warrant
THE SPEC THE CLAUSE IT BECOMES Zero transfer · >99.9% availability by design Performance guarantee + availability LDs. Definition of "available," exclusions & measurement window are everything. 3.2 MW modules scaling to multi-GW Milestones & delay LDs per module, not per facility. Modularity is a negotiating lever: phased payment, phased damages. 20-yr Tier-1 warranty 6,000+ cycles Warranty must be back-to-back with the OEM's. Otherwise ON carries an uncovered multi-decade tail. (See Fig. 19.) 1–8 hr configurable backup duration Change-order trigger. Duration drives battery cost — any spec change moves price materially. US-made · FEOC UL 9540 · NFPA 855 etc. Tax-credit eligibility + conditions precedent to payment. FEOC isn't marketing — it protects the ITC. Certification gates cash.
This mapping is the interview. A candidate who can walk from a datasheet line to the clause it creates is demonstrating the exact skill the JD describes: "when you see a contract provision, your first question is 'what does this cost us?'"
05

Competitive map

Know who else is in the room when a hyperscaler runs this procurement — and where ON's leverage actually comes from.

Fig. 13 — Where everyone sitsPositioning per public product announcements, 2025–26
PROTECTION + GRID REVENUE (storage integrated) PROTECTION ONLY PARALLEL / LOW VOLTAGE INLINE / MEDIUM VOLTAGE BESS integrators Tesla · Fluence · Powin grid revenue, no UPS function LV UPS incumbents Vertiv · Schneider · Eaton huge installed base + service; struggle at the hard VRT curve MV UPS challengers ABB HiPerGuard (w/ Applied Digital) Eaton + Resilient Power (SST) closest direct threats ON.energy AI UPS inline + storage = ride-through AND ancillary revenue patent + NLR validation Real incumbent: "do nothing" — diesel + LV UPS. That's what tripped in Ashburn.
The claimed moat is the upper-right quadrant: only an inline system with substantial integrated storage does compliance and earns grid revenue. ABB's HiPerGuard is the nearest rival — ON's counter is that it's a UPS without the multi-hour storage economics. FEOC rules quietly exclude several Chinese vendors from US hyperscale bids — a commercial weapon, not a checkbox.

The honest risks — have a view on these

Concentration

Crusoe is transformative and also one counterparty carrying commissioning risk over years. That's exactly why payment security and change-order mechanics matter — why you exist.

Incumbent response

ABB, Eaton and Schneider have balance sheets, service fleets and existing hyperscaler master agreements. Patents help; a fast follower with an installed base is a durable threat.

Regulatory reversal

TIEC's intervention on ERCOT's authority could reroute NOGRR 282. ON expects the technical requirements to survive — but the tailwind isn't guaranteed.

Supply chain & capital intensity

5 GW of new commitments needs cells, transformers, PCS and working capital. They're hiring a transformer sourcing manager. Back-to-back supply terms are part of your scope.

06

How they make money

This is what separates you from a candidate who only read the homepage. ON.energy earns four different ways — and each way changes what a contract must do.

Fig. 14 — Four revenue postures, four risk shapeson.energy financing page · press history
1 · Equipment supply & integration ON builds Customer owns one-time $$$ + margin Risk lives in: performance guarantees, delivery LDs, 20-yr warranty tail 2 · Energy-as-a-Service / lease / offtake ON owns Customer pays fee annuity, 10–20 yr Converts a price fight into multi-decade risk allocation: availability defs, dispatch rights, termination value 3 · IPP asset ownership Owns & operates grid-scale storage — $350M AUM, ~80% retained stake in US projects historically. ERCOT market revenue arbitrage + ancillary services 4 · Tax structure monetisation ITC optimisation, IRS safe harbouring, ITC transfer, tax equity — $120M of such capital per their site. For storage, often the difference between a viable and a dead project. Novogradac advises. Same hardware, four different contracts. Which posture a deal takes changes every clause you'd negotiate.
Interview question this diagram answers: "What do you understand about our business model?" Most candidates say "they sell a UPS." You say "they're a manufacturer, an integrator and an asset owner at once — and the contract shape follows the posture."
Fig. 15 — Every contract has two readersThe single most useful idea in this dossier
THE CONTRACT your work product READER 1 · CUSTOMER wants: uptime, LDs with teeth, long warranty, low price, broad liability READER 2 · CREDIT COMMITTEE wants: bankability — enforceable guarantees, creditworthy offtaker, step-in rights, assignability, a liability cap that isn't hollow JP Morgan · Lombard Odier · Ultra Capital · Pathward · BridgePeak read what you sign. Terms that fail bankability kill the deal regardless of price — because ON finances what it builds.
The line to land: "In a company that owns assets as well as selling equipment, a contract term isn't just a risk allocation — it's an input to whether the project gets financed. I'd want to know which terms your lenders treat as non-negotiable, because those are the floor of my position, not the ceiling."
Fig. 16 — The grid-services revenue stackON.energy published indicative values
PEAK SHAVING — 15–30% demand-charge reduction DEMAND RESPONSE — $50–200 / kW-yr ANCILLARY SERVICES — $20–100 / kW-yr ARBITRAGE — market-dep. same one asset, stacked: ← the negotiation nobody mentions: WHO captures this? customer-owned → customer EaaS → ON, or shared — needing settlement mechanics, dispatch rights, and a rule for compute-vs-market conflicts
Turns a cost centre into a revenue generator — ON's phrase. The unglamorous, high-value question is who captures the stack and how conflicts resolve. Asking it marks you as someone who thinks past the headline price. Company claim on the figures.
07

The role, decoded

They wrote the JD emotionally — "grey zone," "commercial courage," "when to walk away." Read it as a description of a person, not a task list.

Fig. 17 — Where you sit: the counterweightJob description, translated to a picture
YOU "what does this clause cost us?" own terms, proposal → signature SALES (McGaha) pushes growth & bookings — you are the counterweight, not the department of no FINANCE (Petersen, CFO) your reporting line — risk assessments & deal summaries go here LEGAL (Contracts Counsel) being hired in parallel — you don't wait ENGINEERING / DELIVERY handoff of obligations into execution ACROSS THE TABLE: CustomersCrusoe & hyperscalers EPCRosendin — scope/interface Supplierscells · transformers · PCS Strategic partnersJV · channel · IP terms
Filed under Finance, not Sales or Legal — that's a real signal. Your incentives are margin and risk. And "develop playbooks, templates, frameworks and pricing tools" means the function doesn't exist yet: you'd be building it, which suits a founder.
They wroteThey mean
"You do not wait for Legal to tell you a clause is a problem."Legal is thin and being built now. Self-start on risk identification.
"The commercial counterweight" to SalesReal friction exists between bookings and risk discipline. Hold a line without becoming the department of no.
"Deal summaries and negotiation strategies for CFO and executive review"Direct visibility to Andrea Petersen. Written communication is a core deliverable.
"We can teach modeling; we cannot teach commercial courage"A deliberate opening for a candidate without project-finance modelling. This is your door.
"Not a contracts administration role"They fear paralegal-adjacent applicants. They want an operator.
"7+ years negotiating complex commercial agreements"The hard bar. Address it head-on — Section 10.
08

Clause playbook

The JD names eight contract concepts. Two of them deserve pictures, because pictures are how the stakes actually land. The rest are in the expandable cards below.

Fig. 18 — Why limitation of liability is THE clauseIllustrative magnitudes — the point is the ratio
WHAT'S AT STAKE IF A POWER EVENT KILLS A TRAINING RUN Contract value the money ON gets paid Capped liability e.g. cap at fees paid + consequentials excluded — survivable UNCAPPED: value of lost compute on billions of GPU capex No cap, no waiver One clause is the difference between a bad quarter and an existential event. Your position: mutual consequential-damages waiver + aggregate cap tied to contract value, with a negotiated (not open-ended) carve-out list.
Also watch: indemnities often sit outside the cap — which quietly makes the cap meaningless. Tie indemnity scope to the insurance programme (ON works with Lockton). If a customer insists on uncapped exposure: price it or decline it, out loud.
Fig. 19 — The warranty tail problemBack-to-back, visualized
yr 0 yr 10 yr 20 ON's WARRANTY TO CUSTOMER — 20 years (marketed) OEM WARRANTY TO ON — if only 10 yrs… …ON is NAKED here
Back-to-back means: never warrant more downstream than you're covered for upstream. Also nail exclusions (misuse, unauthorized modification, out-of-spec operation) and a remedy hierarchy — repair, then replace, then refund. A warranty is a repair obligation, not a general indemnity; say so in the text.
Performance guarantees

What it is. A promise the system hits specified numbers — availability, efficiency, capacity retention, ride-through — measured a defined way over a defined period.

What it costs. Definitions are everything. "99.9% availability" with no exclusions means you pay for the customer's own switchgear failure.

Your position. Guarantee what you control; exclude force majeure, customer-caused outages, grid unavailability, scheduled maintenance; measure over an annual window; cure period before remedies trigger.

Liquidated damages (LDs)

What it is. Pre-agreed money for missing a date or a number, instead of litigating actual losses.

What it costs. A delayed campus is delayed revenue on billions of GPU capex — uncapped delay LDs can exceed the whole contract.

Your position. Cap delay LDs (commonly low-to-mid single-digit % of contract value with an aggregate cap), cap performance LDs separately, make LDs the sole and exclusive remedy for that failure. Grace period; relief for delays caused by the customer, the EPC, or permitting. And with a 3.2 MW modular product: LDs per module actually late, not per facility.

Indemnification

What it is. Covering the other side's losses from third-party claims — injury, property damage, IP, environmental.

What it costs. Often sits outside the liability cap; battery systems add fire/thermal exposure that insurers price carefully.

Your position. Mutual, proportional to fault, tied to the insurance programme. IP indemnity cuts both ways given the patent position.

Payment structures & security

What it is. Milestones, advances, retainage; LCs, parent guarantees, bonds, escrow.

What it costs. ON buys cells and transformers months before customer milestones pay — the working-capital gap is real (it's what the Lombard Odier facility was structured for). A customer-demanded LC ties up ON's credit capacity.

Your position. Front-load milestones toward procurement/manufacturing; match payment triggers to cash-out events; negotiate the size, form and step-down of security instruments.

Change orders

What it is. The mechanism for adjusting price and schedule when scope, site conditions or specs change.

What it costs. On multi-year multi-site programmes this is where margin is won or lost. Chip generations, cooling architecture and 800 VDC are all moving — a customer redesign changes your sizing, and someone pays.

Your position. Define the baseline precisely; specify who authorizes; set response clocks; agree pricing bases (unit rates) up front. Include a regulatory-change mechanism: if ERCOT or NERC tightens a requirement mid-build, that's a change order, not your absorption.

Dispute resolution

What it is. Escalation ladder, mediation/arbitration/litigation, seat, governing law.

What it costs. With LATAM operations and international counterparties, enforceability isn't boilerplate. A slow dispute on a live project is itself a cost.

Your position. Tiered escalation with named executives and short clocks; continue-to-perform and pay-undisputed-amounts language; a forum you can actually enforce in.

Do not overclaim

Learning these makes you conversant, not experienced. If asked "have you negotiated an LoL with a hyperscaler," the answer is no, and you say so. What you can honestly claim: you understand what these terms do to economics, and you've negotiated contracts, financing and partner terms in your own business. Overreaching is the fastest way to lose the room — and they'll find out in round two.

08B

The risk lens — how a senior negotiator actually thinks

The realization that upgrades this whole prep: the job isn't negotiating contracts. It's running a repeatable framework for risk-adjusted commercial decisions — so executives get consistent answers to "what is this risk worth and who should own it?" Every clause in Section 08 is just that question wearing different clothes. This section is the lens; your archive in 10B is the proof you already use it.

Fig. R1 — The decision ladderDistilled from your own archive — every rung has a real deal on it
IDENTIFY — what exactly is the risk? (name the failure mode, not the clause) SIZE — probability × dollar impact a $75K LD with 20% trigger likelihood is a $15K expected cost, not a $75K one 1 · ENGINEER IT OUT change the deal so the risk can't occur ON's entire product is this rung: AI UPS engineers grid-fault risk out of the power path. In your archive: restructuring payment timing so cash-flow risk never arises. 2 · CAP IT bound the downside Team Sunshine: 12-month trailing liability cap + 6-year survival sunset. Empower: aggregate liability cap. LDs removed entirely — a cap set to zero. 3 · TRANSFER IT to whoever holds it cheapest Covenant lease: homeowner carries casualty insurance, SPV named loss payee. Warranty terms = manufacturer holds performance risk. Insurance, warranty, indemnity moves it. 4 · PRICE IT if you must hold it, get paid for it Accepted $2.40 TPO redline as the explicit trade for $2.05 commercial. A concession with a price tag is pricing, not losing. 5 · WALK unpriceable or unverifiable → decline Convert Solar: three dealbreakers, walked. Aegis/OWE: diligence failed, pulled mid-negotiation.
The order matters. Elimination beats capping beats transferring beats pricing — because each rung down leaves more residual risk on your book. Walking is the ladder's floor, not its failure mode.
Cap it. Insure it. Engineer it out.

The three-verb answer if anyone asks how you think about contract risk: "First I ask if we can engineer it out — restructure the deal so the risk can't happen. If not, cap it — bound the exposure in time and dollars. If someone else holds it cheaper, transfer it — insurance, warranty, indemnity. Whatever's left, price it into the commercial terms. And if it can't be priced, that's what walking away is for." Then give one real example per verb — you have them all in Section 10B.

How executives hear a deal — four questions above the clause level

What the exec is really askingWhat it means for how you present
"What's the expected return?"Lead with the economics, not the terms. The clauses exist to protect a number — name the number first.
"What's the residual risk after mitigation?"Never present a risk without its rung on the ladder. "Uncapped indemnity" is a problem; "indemnity capped at 12 months trailing, six-year sunset" is a decision.
"Does this fit our risk appetite?"Some risks are cheap for ON to hold (they own the hardware and the O&M team) and expensive for a customer to hold. Allocation should follow who holds it cheapest — that's the whole art.
"Does it fit the strategy?"A below-market deal with Crusoe-scale strategic value may be right; a rich deal that locks in a bad precedent may be wrong. Terms are portfolio decisions, not single-deal decisions.

This is the JD line "advise executives on how commercial terms affect project economics and risk" made operational: recommendation first, residual risk stated, ladder rung named.

Vocabulary discipline — ideas, not acronyms

Formal frameworks exist for all of this (enterprise risk management standards, risk allocation matrices). Know the concepts — probability, impact, mitigation hierarchy, residual risk, risk appetite — but do not name-drop standards you haven't operated under. If an interviewer raises ISO 31000 or COSO, map your ladder to it live: identify → analyze → evaluate → treat is the same spine. Fluency is mapping on demand; costume is citing on spec. Same honesty rule as everywhere else in this dossier: claim only what survives one follow-up question.

09

The people

Monday is a recruiter screen, not a technical panel. Calibrate for that — but know the map behind her.

Fig. 20 — The org, as it touches this roleon.energy/team · Pathward release · Greenhouse
Alan Cooper — CEO & Co-Founder led every major financing personally · Norman Cooper chairs the board Ricardo de Azevedo CTO & Co-Founder AI UPS architect · ex-Goldman, Barclays · "Don't Call Me BESS" Andrea Petersen CFO — YOUR LIKELY LINE treasury · reporting at "public-company standards" Dax Kepshire COO hyperscale delivery · ~20 yrs scaling storage companies David Fernandes CIO & Founding Partner ex-CFO · asset deployment, project financing & capital Kevin Rooney VP Structured Finance ex-Live Oak (ON's own lender) Alberto Avila VP Finance — Data Centers plausible close partner / interviewer ★ THIS ROLE Sr Mgr, Commercial Neg. filed under Finance · Reston ALSO KNOW: Eric McGaha VP Sales (your counterweight) · Lauren Wong VP Procurement · Jorge Bianchi EVP People · José Manuel Díaz Pérez, President LATAM EXTERNAL NAMES: Chris Dolan — Chief DC Officer, Crusoe Clint Summers — PE, Sr Dir Power Eng., Rosendin
Reporting line is unconfirmed — the highlighted path is inferred from the Finance filing and the CFO-review language in the JD. Asking Andrea "who does this report to?" is a natural, smart question.
Andrea Marsiglia — your Monday interviewer

Public info on her is thin. The invite itself tells you more: it was issued in GMT−05:00 Bogotá, and ON runs real LATAM operations — she's most likely on Colombia time. Inference

Her job is signal and fit, not clause-drafting quizzes: can you explain your story clearly, do you get what ON does, are comp and location workable, will the hiring manager thank her for you. Her prep email told you the test: STAR-structured answers, and possible probing on BESS. Have three STAR stories ready and the Fig. 9 explanation fluent. Be warm and brief — recruiters remember candidates who make their job easy.

10

Your fit & your gaps

Straight assessment. You will interview better having already decided how you talk about the gap, rather than discovering it live.

Fig. 21 — You vs the JD, honestlyTheir own words define both columns
WHAT THE JD ASKS YOUR EVIDENCE Commercial courage · push back, walk away 1000s of one-call closes · founder carrying own P&L Energy sector fluency 9 yrs US solar · financing structures · Circle of Excellence '20 Financial literacy (modeling teachable) BS Corporate Finance, CU Leeds · ran a business Cross-functional communication, exec-ready writing strong verbally · exec writing unproven — say so if asked 7+ yrs complex B2B agreements (EPC, supply, LDs, LoL…) THE GAP IS SCALE, NOT KIND — 10 dealer-scale negotiations, zero at infrastructure scale. Own it. Location / on-site Reston Springfield, VA — plus the Ashburn story makes geography an asset
The strategy the picture implies: they wrote, unprompted, "we can teach modeling; we cannot teach commercial courage, negotiation skills, or sound judgment." That sentence is an invitation to a candidate shaped exactly like you. Make the green rows undeniable; be honest and unbothered about the red one.
The backup move that costs nothing

If Andrea signals the years-of-B2B bar is firm, ask directly whether there's a level below — a Manager or Commercial Analyst seat on the same team with the same trajectory. That converts a rejection into a redirect, and demonstrates the exact self-awareness the role demands. With 56 roles open, including several in Reston, odds are decent something fits.

10B

Your STAR bank

Andrea's prep email asked for STAR-structured answers explicitly. These five are built from your actual history — pulled from our past sessions — and each is mapped to the JD line it proves. Two need you to fill in the numbers; the scaffolding is done.

Fig. 22 — Her prep email vs this dossierAndrea Marsiglia's four tips, checked off
1 · Research company, role & energy storage industry → Sections 01–08: industry forces, ON's history, AI UPS, competitors, business model, JD decoded line-by-line 2 · STAR stories + technical spot-checks (BESS · ISO · QMS) → This section (five stories below) + BESS answer in §11 + ISO/QMS one-liners below and in the glossary 3 · Questions for us → Section 12: five for Andrea, seven held for the hiring manager — team, culture, day-to-day, future challenges all covered 4 · Logistics check → Fig. 0 (the time-zone trap: 5:00 PM ET, not 4:00) + the −60 and −15 minute items in Section 13's run-of-show
On BESS / ISO standards / QMS: we established last session this trio is boilerplate from their technical-role email template — don't over-rotate. But have one clean sentence for each: BESS is the Fig. 9 answer; ISO standards in a manufacturing context means the ISO 9001 family of quality-management standards (careful — in this industry "ISO" also means grid operators like CAISO/ERCOT, so let context tell you which); QMS is a Quality Management System, the documented processes a manufacturer runs to make quality repeatable and auditable — relevant because ON builds hardware in US factories.
Fig. 23 — The negotiation vault: ten counterparties, ninety daysMay–June 2026 · your own archive
TEAM SUNSHINE signed — live nego w/ president GREEN BRILLIANCE active advisory relationship POWUR v25 + term sheet, in play EMPOWER / BOLD 23-term guide delivered YELLOWLITE drafted to v4 CONVERT SOLAR WALKED — 3 dealbreakers AEGIS / OWE PULLED — failed diligence BOUNDLESS/FREEDOM origin of the playbook JAHNUR JV 60/40 operating agreement OWN PARTNER AGMT 24 sections — you wrote the paper others negotiate against Two signed. Two walked. The rest negotiated to a decision. That ratio — not the win count — is what "commercial judgment" means.
The reframe this proves: the JD's "7+ years negotiating complex commercial agreements" gap is about scale, not kind. Every clause category in their posting appears in this archive — at dealer scale, five-figure deal economics, not infrastructure scale. Say that distinction yourself before they do.
JD clauseWhat you actually didWhere
Payment structures30/40/30 milestone splits; 45-day post-PTO backstop; 14-day payment deadlines; redline locked at submission; proportional tranches within 7 business days for cash & commercialTeam Sunshine, Yellowlite, Empower
Liquidated damagesIdentified and removed two $75,000 LD clausesEmpower / Bold Energy
Limitation of liabilityAdded trailing 12-month liability cap + 6-year survival sunset to their indemnification; separate aggregate cap negotiatedTeam Sunshine, Empower
IndemnificationCaught asymmetric indemnity (you: negligence; them: gross negligence only) with their investors inside your obligation; pushed mutual + investors removedConvert Solar, Empower
Warranties & performance guaranteesProduction guarantee designed as one of nine mandatory lessor obligations in your own lease productCovenant Solar Lease℠
Security requirements$7,000 forgivable signing advance with partial-recovery floor; marketing advance converted to capped $500 deductionsAegis, Team Sunshine
Change order mechanismsTiered-notice ladder on pricing changes: 7 days documented pass-throughs / 14 days <10% reductions / 30 days material changes; one-increase-per-quarter cap; pipeline rate lockEmpower, Yellowlite
Dispute resolutionArbitration seat negotiated to responding party's home state; continuous portal access + payments continue during disputes; injunctive-relief carve-outConvert term sheet, Aegis, Yellowlite
Order of precedence / unilateral amendment§1 order-of-precedence placing your agreement above a 46-page platform rulebook; switched off their unilateral-change and 30-day-acceptance provisionsPowur

Every row is from your own documents, May–June 2026. If asked for detail on any of them, you have it.

№1 — Team Sunshine — the anchor, now with its real numbers

S: "I was building Raynora's installer channel and needed a Massachusetts installer relationship. Team Sunshine's president, Ahmad Hodroj, skipped my term sheet entirely and sent back a pre-executed 17-page channel agreement on their paper — classic anchor move."

T: "Sign their paper as-is, or negotiate a live counterparty who'd already shown he negotiates hard — without losing the relationship."

A three-part: (1) built a one-page locked Addendum through ten versions plus a four-column discussion guide through eleven, sent the PDF the day before with a warm note and held the guide for the live call; (2) traded, didn't demand — deleted chargebacks entirely, framed as a balanced trade against their cancellation rights and non-exclusivity; split payment timing: install-completion for TPO and lease deals, proportional payment within seven business days for cash and commercial; countered commercial pricing at $2.05/W as the explicit trade for accepting their $2.40 TPO; added a 12-month trailing liability cap and a six-year sunset to their indemnity instead of fighting the clause head-on; (3) conceded live — when Ahmad pushed back on the "culpable act or omission" trigger and the personal-guaranty exemption, he was right that both were overreach, and I dropped them on the spot.

R: "Signed agreement, working relationship, and a repeatable two-document playbook I then reused with four other counterparties. And a lesson your JD describes exactly: knowing when to push and when to concede is the same skill."

The 30-second version for the screen

"An installer's president answered my term sheet by sending back a pre-signed contract on his paper. Over ten drafts and a live text negotiation I traded rather than demanded — gave up chargebacks to win payment timing and a liability cap, took $2.40 on TPO to get $2.05 on commercial — and dropped two of my own asks the moment he correctly called them overreach. We signed. The playbook from that deal became my template for every negotiation since."

№2 — The Covenant Solar Lease℠ — structuring, with the numbers attached

Designed a Sharia-compliant lease product (Ijara Muntahia Bittamleek) satisfying the IRS true-lease test (Rev. Rul. 55-540 eight factors) and Islamic finance constraints simultaneously; locked terms include 20-year term, 8% nominal target IRR on the lease payment stream, ~11–12% all-in SPV yield including ITC and MACRS bonus depreciation, Year-6 purchase-option gate deliberately set one year past the §48E five-year recapture window, FMV methodology at 13% of cost declining 1%/yr to a 5% floor, net buyout = gross FMV minus offsetting obligations floored at zero; benchmarked the design against four national contracts (Sunrun PPA, Participate Energy, Skylight, EnFin) clause by clause, including hell-or-high-water payment provisions; separately produced a §48E Beginning-of-Construction reference covering the 5% Safe Harbor vs the four-year continuity safe harbor and §6418 transferability at 88–94 cents on the dollar.

Caution: keep counterparty-confidential specifics out of it (you've restricted Covenant details in other applications for competitive reasons — same discipline applies here; the structure and the model are yours to discuss, the deal specifics may not be).

When your JD says modeling is teachable — I taught myself the version that matters: what a term does to the economics.

№3 — Convert Solar — the walk-away

S: signed originator agreement in hand, then ran a clause-by-clause plain-English breakdown before scaling the relationship.

T: decide whether the paper was fixable.

A: flagged eight watch-outs and isolated three dealbreakers — zero-day at-will termination for them while I owed 60 days' notice; undefined territorial exclusivity with commission forfeiture as the penalty; and compensation forfeiture if a closed sale never reached completed-install status even with no fault on either side. Drafted the replacement addendum, made the asks.

R: "The economics of their position didn't work at any price I could charge, so I walked and kept my Virginia and Maryland optionality. Your JD says know when to walk away — this is what mine looked like: named dealbreakers, a written record, and no burned bridge."

№4 — Aegis / One World Energy — the diligence pivot

S: mid-negotiation on a channel agreement, terms substantially settled including a disclosed 10% margin cap and a four-step commission formula.

T: before sending for signature, verify the upstream installer network actually existed where they claimed.

A: real-time diligence found no East Coast operating presence — registered-agent addresses instead of warehouses, no Google Business footprint, no response by phone, email, or text. Pulled the agreement and sent a due-diligence email instead, asking specifically who installs in the DMV, Maryland, New Jersey, and Massachusetts.

R: "The deal never closed, and that was the win. Your JD's first behavioral bullet is 'identify commercial risks early' — the cheapest place to catch counterparty risk is before your signature, and I've built that reflex the expensive way, by watching an installer I depended on [Boundless/Freedom Forever] fail."

№5 — The playbook — proof you can build the function

The repeating pattern developed across Boundless, Team Sunshine, Empower, Convert, and Yellowlite: one clean locked executable addendum as a PDF, paired with an editable four-column discussion guide (Section / Proposed language / Rationale / Their notes) for the live call; doctrine of aggressive-but-reasonable v1 baselines, never self-negotiate, let the other side redline; document hygiene rules enforced (column says "Rationale" not "Why," counterparty column says "agreed" not "counter"). Plus the other side of the table: authored Raynora's own 24-section Sales Partner Agreement.

Your JD's last bullet — develop playbooks, templates, negotiation frameworks — isn't aspirational for me. I have one. It's just priced in kilowatts instead of gigawatts.

№6 — Reserve stories — deploy only if the question fits

The five-offer week: May 2026 comp negotiation with layered asks and a pre-decided walk-away — use only if asked about negotiating under personal stakes.

Founding the school: $6M budget, 1,000+ students, operations from zero — use for building-from-scratch questions if the playbook story is already spent.

The 2020 Circle of Excellence year: being the COVID year (adaptability).

Scale honesty — say it before they do

These are dealer and channel agreements with five-figure per-deal economics — not EPC or nine-figure supply contracts. Volunteer that distinction: "Same clause surface as your JD, smaller number of zeros — and the zeros are the teachable part." The Green Brilliance / PJM advisory work is usable only at the level of detail you can personally defend when a CFO asks "which substation, which tariff, what did you conclude?" If you can't answer the follow-up, don't make the claim.

11

Hard questions, with answers

Rehearse out loud. The shape and the honesty matter; the exact words don't.

"Walk me through your background."
Shape — 90 seconds, arc not chronology

"I've spent nine years in US solar, mostly on the front end — Sunrun, where I was a Circle of Excellence honoree in 2020, then running my own advisory business, Raynora, in the Mid-Atlantic. What kept pulling me deeper was the part of the deal after the handshake: how it's financed, what the terms actually obligate you to, what happens when something underperforms. I have a corporate finance degree and I've been running my own P&L, so I've lived on that side by necessity. This role is that work at the scale where it really matters, in the market on my doorstep."

"You haven't negotiated EPC or supply agreements at this scale. Why you?"

The one you must not fumble. Concede fast → redirect to their own stated criteria → offer proof → name reality.

Shape

"Fair — I haven't negotiated a nine-figure supply agreement with a hyperscaler's counsel, and I won't pretend otherwise. What I can tell you is that in the last ninety days alone I've negotiated ten entity-to-entity commercial agreements — and the clause surface was your job description: milestone payment structures, liquidated damages I got removed, liability caps I added, asymmetric indemnification I caught and reversed, tiered change-notice mechanisms, payment security, arbitration terms. Two I signed, two I walked away from on named dealbreakers, one I pulled mid-negotiation when due diligence failed. The gap between my record and this seat is the number of zeros on the contracts — and your posting says the zeros are the part you can teach. What you say you can't teach — the judgment about when to push, trade, and walk — is documented in my own archive. I'd also genuinely want your read: is that a bar you can flex on? If not, I'd rather discuss whether there's a level below this seat with the same trajectory."

That last move is the trait they're screening for — knowing when to push and when to name reality — while keeping the door open.

"What do you know about ON.energy?"
Shape — a point of view, not the homepage

"You started around 2015 as a storage developer and integrator in Latin America, became an IPP in ERCOT, and in February launched AI UPS — which is the interesting move, because it isn't a BESS. It's inline at medium voltage: no transfer time, and the whole plant sits behind it, not just the racks. Then in July, 5 GW with Crusoe. What strikes me commercially is you're three businesses at once — manufacturer, integrator, asset owner — so contract terms don't just allocate risk, they decide whether a project gets financed. That's a more interesting problem than vendor contracting."

"Tell me about a time you pushed back and held your position."

STAR, as she requested — and you have two ready in Section 10B: Team Sunshine (№1) if the question is about a commercial deal, or the five-offer week (№3) if it's about holding a position and walking away. Be specific about the number at stake and the outcome. Losing a deal on principle is a better answer than winning one, given the JD's "know when to walk away."

"Why are you leaving your current role?"
Shape — direction, not escape

"Nothing's wrong where I am — I'm producing. But I've spent nine years paid per transaction, and the work I find most interesting has moved upstream into structure and terms. I want a seat where the output is the quality of the deal, not the volume of them, and where I'm building something durable — playbooks, frameworks — rather than starting from zero every month."

Honest, forward-facing, and it maps directly onto a JD that asks you to build playbooks and templates.

"What are your compensation expectations?"

You already did this work: last session, against your fixed monthly obligations, you landed on $140–160K base as the target, with any equity conversation deferred to later rounds. Walk in with that number owned.

Shape

"I'm targeting a base in the $140–160K range with bonus on top, based on senior commercial roles in this market — but I'd rather calibrate to your band. What range is the role approved at?"

"What's a BESS?" — the spot-check she warned you about
Shape — answer, then take the extra step

"A battery energy storage system — batteries plus power conversion plus controls, sitting beside the power path, storing and dispatching energy over minutes and hours: charge cheap, discharge expensive, sell ancillary services. ON has built and owned those for years. But your CTO's point is that a BESS is parallel — it has to detect a fault and transfer, and even a fast static switch is a few milliseconds, so the GPUs already saw the hit. AI UPS is inline; there's nothing to transfer. I gather you'd rather people didn't conflate the two."

12

What you ask

Pick three or four. Ask the ones only someone who understood the business would ask.

For Andrea, on Monday

  • "Who does this role report to, and who would I meet in the next round?" — confirms the Fig. 20 inference.
  • "Is this a new seat or a replacement — and what triggered opening it now?" — the honest answer is almost certainly Crusoe.
  • "What does the process and timeline look like from here?"
  • "With 56 roles open, how is Reston specifically growing?"
  • "What separates candidates who make it through from those who don't?"

Held for the hiring manager

  • "With Crusoe phased over 2026–27, how are you handling change-order mechanics as chip and cooling architectures move? An 800 VDC transition changes the load profile you sized to."
  • "Because you own assets as well as selling equipment, your lenders effectively read every contract. Which terms do your credit partners treat as non-negotiable?"
  • "In an EaaS structure, who captures the grid-services stack — and how do you resolve a conflict between a market dispatch opportunity and the customer's compute demand?"
  • "NOGRR 282 is a real tailwind, but TIEC has challenged ERCOT's authority. How do you price regulatory-change risk into a multi-year agreement?"
  • "How back-to-back is the 20-year warranty with your Tier-1 suppliers today — and where does the uncovered tail sit?"
  • "Is there an existing playbook and template set, or is building that the first six months?"
  • "What should this person have delivered by day ninety?"
13

The 30-minute plan

Thirty minutes with talent acquisition is roughly 12 minutes of you talking. Budget it.

  • −60Test the Teams link. Confirm the calendar reads 5:00 PM ET. Quiet background, solid connection, water, resume + this page on a second screen.
  • −15Search "ON.energy" and "PJM data center" for weekend news. Thirty seconds of fresh news is disproportionately valuable.
  • 0:00Warmth first. Thank her for the confirmation and mention you followed her prep note. Recruiters notice.
  • 0:02Your background — 90 seconds, arc not chronology, land on why this role. Then stop talking.
  • 0:05Her overview. Listen properly. Ask one clarifying question about something she actually said.
  • 0:12Her questions. STAR. If BESS comes up, run the Fig. 9 answer. Keep each answer under two minutes.
  • 0:20Bring Ashburn in yourself if she hasn't opened the door. You cannot leave without using it.
  • 0:23Your questions — three, not seven. Reporting line, trigger for the role, process.
  • 0:27Address the experience gap yourself if it hasn't come up — better from you than discovered later. Ask about adjacent levels if the signal is negative.
  • 0:29Close explicitly: "I want this role. What's the best next step from your side?" Confirm timeline.
  • +2hThank-you email, four sentences: thanks, one specific from the conversation, one line tying your interest to Ashburn, restated availability.
One behaviour to watch

In-home closing rewards filling silence and steering. A screen is a listening exercise: answer, stop, let her lead. If you're talking more than half the time, you're losing.

14

Glossary

Skim the night before. Recognize all of these instantly; be able to define about half.

TermMeaning
AI UPS™ON's patented inline medium-voltage UPS with integrated storage.
BESSBattery Energy Storage System. Parallel to the power path; minutes-and-hours energy management.
Double conversionAC→DC→AC continuously, so the load never sees raw grid power.
ZVRT / LVRTZero / Low Voltage Ride-Through — staying connected through a voltage collapse.
NOGRR 282ERCOT ride-through mandate for computational loads ≥75 MW. Effective 1 Aug 2026.
LCLLarge Computational Load — ERCOT's term for such a facility.
Ramp rateHow fast load may rise/fall from the grid's view; regulated because instant swings destabilize.
Residual riskWhat remains after mitigation — the thing executives actually decide on. Every negotiated clause is a residual-risk statement.
Risk allocation matrixA table mapping each project risk to its owner (developer, EPC, supplier, insurer, customer). The formal version of what a term sheet does implicitly.
Risk appetiteHow much residual risk an organization will knowingly hold. Allocation should send each risk to whoever holds it cheapest.
p.u.Per unit — voltage as a fraction of nominal. 1.0 is normal; 0 is total collapse.
Medium voltage~13–35 kV; where most large facilities interconnect.
PCSPower Conversion System — the inverter/rectifier stack.
STSStatic Transfer Switch — fast, but still ~4 ms. ON's argument against parallel designs.
N+1 / 2N+1Redundancy: one spare unit / full duplication plus a spare.
Grid-formingInverter establishes voltage & frequency itself — enables islanded operation.
PJM / ERCOTMid-Atlantic RTO (largest US operator, largest DC concentration) / Texas operator (first to regulate large loads).
FERC / NERCFederal economic regulator / reliability-standards body.
Co-locationSiting large load at a generator — the subject of FERC's PJM docket EL25-49.
Ancillary servicesPaid grid support: frequency regulation, reserves, voltage support.
ITC / safe harbour / tax equityInvestment Tax Credit (now transferable) / locking a credit vintage by early spend / investor capital exchanged for tax benefits.
FEOCForeign Entity of Concern — sourcing restrictions affecting tax-credit eligibility.
EaaS / offtakeProvider owns the asset, customer pays a service fee / long-term purchase contract that makes a project financeable.
EPCEngineering, Procurement & Construction contractor. Rosendin is a named ON partner.
LD / LoLLiquidated Damages / Limitation of Liability (cap + consequential-damages exclusion).
Back-to-backPassing an obligation upstream on identical terms — never warrant more than you're covered for.
BankabilityWhether a lender will finance against the contract. Failing terms kill deals regardless of price.
IPPIndependent Power Producer — non-utility owner-operator. ON is one.
ISO 31000 / COSO ERMThe two big enterprise risk-management frameworks. Know the spine (identify → analyze → evaluate → treat); don't cite them unprompted.
ISO standardsIn a manufacturing context (as in the recruiter's email): the ISO 9001 family of quality-management standards. Watch the double meaning — in grid conversation, "ISO" means an Independent System Operator (CAISO, NYISO, ERCOT). Let context tell you which.
QMSQuality Management System — the documented processes a manufacturer runs so quality is repeatable and auditable. Relevant because ON builds hardware in US factories; certification often gates customer payment.
ARIES / NLRDOE research platform and national lab where AI UPS was tested at 13.2 kV.
15

Sources

Everything traces to one of these. Company materials are marketing, not audit — treat accordingly.