Project development & EPC delivery
How a BESS project gets built
Six stages, seven gates, and a construction bill that only arrives once somebody else’s money does. This is the shape of a grid-scale battery project from a piece of land to commercial operation — how long each stage runs, roughly what it costs, and what it costs you when it goes wrong. Underneath it is one move repeated: the developer spends the most expensive money in the project answering the questions nobody else will fund, and each gate is where cheaper money agrees to take an answer off their hands.
The stage machine
HoverTap any stage row in the figure for a summary of that stage
Read it in two halves. Everything left of financial close is development, where the developer spends their own money to answer questions nobody will fund until they are answered. Everything right of it is delivery: notice to proceed (NTP) authorises construction, and the commercial operation date (COD) is when the plant is finally allowed to earn at the contracted rate. The build itself ends at mechanical completion; COD comes after commissioning and the capacity test, and NTP to COD commonly runs 14–18 months — the commissioning spread is mostly retests and the operator’s witness-test slots. Two drawing notes: the dashed early-works bar is construction started before close, at the sponsor’s own risk, and commissioning has its own lane in the figure but belongs to the construction stage’s money and contract.
The bar that sets the schedule is interconnection — permission to connect to the grid, and a study of what the grid needs built before it can accept you. It starts early, runs longest, and it is the one stage whose cost cannot be quoted when it begins: the network-upgrade figure arrives with the study, years after the land is under option, and in congested markets the stage commonly runs three to six years. On the projects I have worked on, this is the stage that ends them — not because the answer is usually bad, but because it is late, and binary. Land to COD typically runs three to five years where the queue moves, and six to eight where it does not — the interconnection bar decides which of those two projects you have.
What each stage costs, and who is paying for it
| Stage | Typical duration | What moves it | If it fails | Share of cost | Who funds it |
|---|---|---|---|---|---|
| Site identification | 2–6 months | Whether the land is already assembled, and how many owners have to agree | You lose the option fee — the payment that held the land — and walk: the cheapest failure available | <1% | Developer equity |
| Permitting | 6–12 months | Whether environmental review is a screening or a full assessment, and whether the fire review goes to a hearing | A refusal, or a condition you cannot live with, ends it after the spend | 1–2% | Developer equity |
| Interconnection | 12–24 months where the queue is clear; 3–6 years where it is congested | Which cluster you land in — the queue studies projects in groups on a fixed cycle — and whether a restudy fires; a restudy is most of the difference between the two | The upgrade bill lands years in and can make the project unfinanceable — too expensive for any lender to carry | 0–20%+, assigned by the study; secured before close, paid as built | Deposits from developer equity, and the security for the assigned upgrades posted before close; the upgrades themselves funded as built, from the construction financing (bottom row) |
| Off-take agreement | 6–12 months | Whether a creditworthy counterparty has a procurement window open, and how much of the revenue you are willing to contract away | Thin contracted revenue means thin debt, so gearing falls until the equity cheque — the cash the developer puts in rather than borrows — grows to fill the gap, and the project stalls rather than dies | <1% | Developer equity |
| Financing | 4–7 months | One lender or a syndicate — a club deal adds a round of everyone reading everything | A condition the lenders set before funding slips, stranding early works already built at the sponsor’s — the developer’s — risk | 1–2% | Developer equity |
| Construction | ≈12 months to mechanical completion; NTP → COD commonly 14–18 | The delivery date on the main power transformer and HV switchgear — usually the long-lead items; the containers usually do not set the schedule | Missing the guaranteed completion date triggers liquidated damages — pre-agreed damages per day of delay. US courts enforce them where the sum was a reasonable forecast of the loss rather than a penalty; England and Australia now ask instead whether it is out of all proportion to a legitimate interest in performance. They are normally capped, and up to that cap they are the owner’s exclusive remedy for delay — reaching it usually opens a termination right instead — under the EPC (engineering, procurement and construction) contract, the off-take, or both | 80%+ | Construction debt and the sponsor’s own equity. US tax equity — investors paid in the project’s tax credits — signs during construction and funds at mechanical completion or COD, taking out the construction lender rather than digging the hole |
How to read this. These are practitioner ranges from projects I have worked on, not figures from a standard. The stages overlap — permitting and the queue start together — so the durations do not add up to the elapsed schedule; the figure above shows the real shape. The cost shares are approximate ranges of total project cost, not a decomposition of it. They do not sum to 100% because each is an independent range observed across different projects rather than a slice of one budget — and because the interconnection assignment is sometimes inside the EPC-financed share and sometimes on top of it, depending on who builds the upgrades. To put a scale on the column at all — a 100 MW / 400 MWh system at roughly $250/kWh is a $100 million build, of which construction alone is $80 million or more. The four developer-equity rows sum to roughly 2–5% of total project cost sunk by financial close — the money lost if the project dies, and the number to hold against everything below. Interconnection’s share is the number the study assigns — commonly 10–20% when upgrades bite, from roughly nothing to well past that on unlucky sites. What the developer spends in cash during the stage is study deposits and readiness postings, typically single-digit millions. What it commits is much larger: the interconnection agreement requires security for the assigned upgrades — normally a letter of credit for the full estimated cost — posted before financial close and drawn as the upgrades are built. On a $100 million project a 10–20% assignment is $10–20 million of credit capacity tied up before a lender has funded anything. Construction carries four-fifths of the money; interconnection carries the mortality — the biggest spend and the likeliest ending are not the same stage.
Why the published interconnection averages are the wrong number
Berkeley Lab’s figures are for the assignment, not the deposits. Total interconnection cost for storage ran about $248/kW in MISO across requests of every status — on a four-hour system at $200–400/kWh, or $800–1,600 per kW, that is roughly a sixth to a third of the build. That MISO number carries the same selection problem as the PJM split below: it averages requests that were never built alongside those that were. On an expensive build — $400/kWh, about $1,600/kW — that is roughly 16%, inside the 10–20% band. On a cheap one — $200/kWh, about $800/kW — it is over 30%, well past it. The share moves on your own $/kWh assumption as much as on the assignment. In PJM the same lab puts storage at about $335/kW across requests of every status, and separates outcomes elsewhere: the projects still active or withdrawn average far more than the handful that completed every study, which come in nearer $4/kW — a sample small enough to be an illustration rather than a statistic, and one whose members happened to draw almost no network upgrades at all. Read those two as a population statistic rather than a planning number: a large assignment is usually not paid but withdrawn from, so built projects average almost nothing and the queue’s average is dominated by its deaths. What you can plan against is the 10–20% band — what an assignment looks like when it lands — and the walk-away test is whether the off-take can carry it, not whether it beats an average computed over projects that died.
Where this applies. The institutions named here are United States ones — cluster studies under FERC’s (the US federal energy regulator’s) queue reform in most regions, though ERCOT, the Texas grid operator, sits outside FERC’s order and runs its own process — plus an authority having jurisdiction, notice to proceed, and a tax-equity market that exists nowhere else. The stage order generalises; the durations do not: Canada, Great Britain and the EU each run their own connection regime, on their own clocks.
The gates, in order
The stages above are where the time and money go. The gates are where the risk changes owner, and that is the engine of the whole lifecycle: the developer’s money, then a buyer’s, then the lenders’ — each cheaper than the last — and then, in the US, the tax-equity investor’s, which is dearer than the debt it takes out but cheaper than the equity it displaces. Each refuses to move until the gate before it has priced what it is taking on. Throughout this section the owner is the project company that signs the EPC contract, which is the developer or sponsor wearing its contracting hat. The figure marks three of the gates — FID, financial close and COD. In full there are seven, and each one ends an argument and starts a clock. Most developers also do not build what they develop: ready to build is a sale point as much as a milestone, and the company that assembled the land, the permits and the queue position often sells there and never sees notice to proceed. The links carry the depth; this table is the sequence.
Two things to hold while reading it. First, none of these is defined by statute, and only commercial operation is defined for you at all — the ISO tariff and the interconnection agreement carry their own COD, which need not be the off-take’s. The wording, the boundaries and even the order are otherwise whatever a contract stack says they are, so this is the usual sequence and not a fixed one. A sponsor holding a transformer slot issues a limited notice to proceed before close — drafted so it does not start the completion clock; some contracts define substantial completion to require COD, and then the two land together. And the vocabulary is the US-style EPC and project-finance stack — mechanical and substantial completion, care, custody and control, punch items (the minor work left unfinished at handover, which the contractor still owes). A FIDIC contract draws the line in one place rather than two: its taking-over certificate is substantial completion — care of the works transfers and delay damages stop — while the performance certificate comes at the end of the defects notification period, a year or more later, and answers to final completion rather than to anything before COD.
| Gate | What it unlocks | What is now owed | Who stands behind it |
|---|---|---|---|
| Ready to build where the queue and permits close, just left of FID | The first gate a buyer will pay a firm price for — earlier stages trade, but at a discount for what is still unproven — before it the project is a position, after it an asset with a price | Keeping the checklist simultaneously valid: permits, options and queue milestones each age on their own clock | No one issues it — it is a checklist, not a certificate; a buyer’s technical due diligence is the only real test |
| Final investment decision the FID mark — inside the financing bar | The first spend the sponsor cannot stop cheaply: the vote authorises signature or unconditionality of the EPC and supply contracts, and releases early works. Walking away now costs cancellation charges, not a forfeited option fee | Everything the vote was priced against is dated — EPC validity, held equipment slots, queue security postings. A package aged six months is a decision about a different project | No certificate and no counterparty: a board or investment committee — each sponsor’s, where the project has more than one |
| Financial close the financial-close mark | The right to borrow: the construction facility becomes drawable, though the first draw still has its own conditions to satisfy. US tax equity is bound into the close package but funds later, at mechanical completion or COD | Every later release has its own conditions to satisfy, each signed off by the lenders’ independent engineer — close is a gated series, not one payment | The lenders, on their independent engineer’s report |
| Notice to proceed days to weeks after close | The first date the contractor owes rather than the owner — the guaranteed completion date is computed from this notice | The contractor owes that date against liquidated damages, and the owner owes what it promised to hand over — access, permits, outage windows, free-issue equipment. A notice issued before those are real buys an extension-of-time claim from week one, moving the date the damages were protecting | Nobody — the owner issues it, which is why the owner has to check itself |
| Mechanical completion end of the construction bar | The first completion milestone in the payment schedule — the delivery milestones are larger and were paid long before. This is the point the plant stops being built and starts being tested — normally an EPC payment milestone, and in US tax-equity structures the funding test under the equity capital contribution agreement | Delay damages keep running past this certificate until the milestone the EPC guarantees is certified, so mechanical completion is not schedule relief — and commissioning runs months, not weeks | The contractor gives notice; the owner issues or withholds the certificate, and where debt or tax equity funds against the milestone the independent engineer certifies too |
| Substantial completion inside commissioning | The handover itself: care, custody, control and risk of loss stop being the contractor’s and become the owner’s — on a plant that can export, but not yet at the off-take price | Delay damages, running since the guaranteed completion date, stop when the milestone the EPC guarantees — usually this one — is certified. The workmanship warranty starts here, so an early certificate spends it on a plant not yet in service | The contractor gives notice; the owner issues or withholds it, through its engineer only where the contract appoints it to that role |
| Commercial operation date the COD mark | Permission to sell at the contracted rate. Energy exported during commissioning settles under the off-take’s test-energy provisions, which do not pay it. The construction facility — a short-term loan a longer-term one replaces — is repaid or converts to term debt here | Everything long-dated now measures against the capacity result — at the boundary, ambient and state-of-charge window the EPC names, so this test is won in the contract months earlier, not on the day | Declared, not discovered: the owner serves notice with the evidence attached, and the off-taker has a window to accept or dispute. The contract decides what a dispute does to the date — and revenue follows that effective date, not the first electron |
The gap between the two completions. Delay damages under the EPC stop at substantial completion, but the off-take’s clock runs on to COD. Where the two are guaranteed independently, the owner pays that gap at the off-take’s per-MW-per-day rate — and what runs in it is permission to operate and ISO market registration, commonly one to three months and longer where test slots are scarce. That is why sponsors try to guarantee substantial completion and COD as one date. Signing is also the moment the owner stops holding a clock over the contractor: the delay damages stop, and what is left is retainage — money withheld until the snags are finished — punch-list rights, and whatever the performance guarantees carry. An April export against a June COD, meanwhile, is two months of market exposure and no contracted revenue, and in a tax-equity structure the construction lender is usually taken out at COD rather than converted.
Three of those checks have their own entries: the technical due diligence a buyer runs at ready-to-build, the independent engineer whose reports the lenders rely on and who certifies completion where debt funds against it, and the owner’s engineer, which argues the owner’s side of every one of them — certifying substantial completion only where the contract appoints it into the Engineer role, and witnessing the capacity test at COD. And the baseline set at COD belongs to the capacity warranty — which is why a rushed COD test quietly shifts a decade of degradation risk onto the owner.
Five questions to ask of any schedule
Everything above compresses into five questions. Someone showing you a development schedule has answered them, whether or not the schedule says so.
- Does the elapsed total assume the stages run one after another? They overlap — permitting and the queue start together — so a schedule that adds the stage durations up has built in slack it will spend somewhere else.
- Is a restudy priced into the interconnection line? A restudy is most of the distance between twelve months and six years, and it fires when projects ahead of you withdraw and the cluster’s assumptions change — and again if you materially modify your own request.
- What is the transformer lead time, and when was it ordered? The build is governed by the main power transformer and HV switchgear, not by crew size, so the construction bar was really set at the purchase order.
- Which completion is guaranteed — substantial completion or COD? If the EPC guarantees one and the off-take the other, the owner carries the gap between them at the off-take’s per-MW-per-day rate — reconciled cheaply before signature, expensively after.
- How much is sunk before financial close, and whose money is it? Roughly 2–5% of total project cost in cash, plus the interconnection deposits, all of it at the developer’s risk — no project debt arrives before close, and tax equity later still. Ask what is committed as well as spent: the interconnection security is posted before close and never shows up as development spend.
Read a development schedule and know where it is lying
The table gives you the shape. The Complete Guide gives you the reasoning underneath it: one of its 11 sections is project development and finance, another is how battery storage makes money, and most of the rest is the engineering those stages are deciding about — so you can tell a bankable project from one that is quietly finished.
See the Complete Guide curriculum Or start with the learning path
Where each stage is explained in full
This page is the map. Each stage below is owned by a page that goes properly into it — or take the whole lifecycle in order in the Complete Guide.
-
Site identification
Finished when the land option is executed with the lease attached as a negotiated exhibit. An option, not a lease — it keeps walking away cheap while permitting and the queue do their work — and it is the first line of the ready-to-build checklist; the other lines arrive over the next three stages, and each one ages.
-
Permitting
Finished when every discretionary approval is issued and its conditions are accepted — read the conditions as design scope, not paperwork. Engage the fire authority before the layout freezes: a UL 9540A report answers what they will ask anyway, and a spacing objection after the civil design is fixed is a redesign, not a comment.
-
Interconnection
You cannot move the queue, but you can hedge it: the developers who survive this stage carry more than one position, look hard at surplus interconnection and retiring-plant sites, and check whether accepting a lower export limit drops them out of the upgrade that is killing the project. Finished when the interconnection agreement is executed and the network-upgrade cost is allocated. Set the walk-away number before the study answers — assignments commonly land in the 10–20% band when they bite, so a bill far past that band is where a disciplined developer stops paying to hold the position. And watch the queue above you, not just your own application: a restudy fires when projects ahead withdraw and the cluster’s assumptions change.
-
Off-take agreement
An off-take is the contract that turns uncertain revenue into contracted cash flow — for a battery usually a tolling structure, where the buyer pays a fixed fee for the right to dispatch the plant and keeps what the market pays, or a capacity-only contract, where the buyer pays for availability and the seller keeps both dispatch and the energy revenue — rather than a sale of energy at an agreed price. Finished when one a lender will bank is signed. Contract away enough revenue to carry the debt your model assumes and no more: over-contract and the merchant upside that justified the project — the revenue you leave uncontracted and exposed to market prices — is gone; under-contract and the debt shrinks until the equity cheque grows to fill the gap.
-
Financing
Finished at financial close, with every condition precedent satisfied — what the lenders require to be true before they are bound to lend: executed land, interconnection and off-take agreements, permits in force, the independent engineer signed off. The sponsor commits months earlier, at the final investment decision (FID), and bridges the gap with its own money — size that bridge deliberately, because a condition that slips finds the early works already built.
-
Construction
Finished when COD is declared and the off-taker’s window to dispute it has run. Notice to proceed (NTP) starts the clock, and the guaranteed completion date is computed from it — delay damages then run from that date, not from the notice. The schedule was set at the purchase order: the long-lead items are usually the main power transformer and HV switchgear, not the battery containers — and foundations not accepted when a container lands turn a delivery into a laydown problem, a container with nowhere to go.