Shovel-Ready Is Not Bankable

A project can have land, power, permits, a design and a contractor, and still be impossible to finance.

I have watched it happen for twenty years, across LNG, power and infrastructure, in the United States, Europe, the Gulf and Asia. It is happening again now, at speed, in the AI infrastructure buildout. The pattern has not changed. What has changed is the cost of getting it wrong, because in this cycle the things a delayed project loses are the things that made it valuable.

The industry has become very good at measuring how hard it is to build. Data Center Watch counted roughly $130 billion of US data center investment blocked or delayed in the first quarter of 2026, driven mainly by local opposition. Grid capacity, interconnection queues, permitting exposure and community risk are now screened, scored and tracked by people who do it well.

That work is real and necessary. The conversation around it stops one gate too early.

Those numbers count projects that never cleared development. They do not count the projects that did, and then stopped anyway.

Many of those projects are not formally cancelled. They simply go quiet. A data room stops being updated. A term sheet never arrives. A credit committee never convenes. Nine months pass and nothing has formally gone wrong.

Nobody publishes that number. In my experience it is the most expensive number in this industry.

Terms, before we go further

The industry uses shovel-ready, development-ready, finance-ready, bankable and investment-ready more or less interchangeably, which is part of the problem. These are the working definitions I use.

Development-ready means the project can be built. Site control, power, permitting path, design, contractor.

Institutionally ready means the transaction can be approved. The evidence, structure, risk allocation and governance are sufficiently developed that a credit or investment committee can reach a decision on them.

A project can be the first without being the second. Most stalled projects I see are exactly that.

Two different questions

Development asks: can this be built?

Financing asks: can this be approved, and on what terms?

They are not the same question. They are not answered by the same people, with the same evidence, at the same time. And the second is often the harder of the two, because technical problems usually have engineering answers while structural ones require several counterparties to agree.

A committee is not only assessing whether steel can go in the ground. It is assessing whether the revenue is contracted, whether the counterparty can pay, whether risks sit with parties who can carry them, whether the assumptions hold, and whether every material claim has a document behind it.

One distinction worth making early. Most of what follows is written from the lender’s side, because debt is usually the binding constraint and the discipline is sharpest there. Equity asks different questions and will accept risk a lender will not, at a price. But an equity story that cannot eventually support debt is a smaller story than the sponsor thinks it is.

Where projects actually stop

Nine recurring failure patterns. These are not survey findings. They are what I have seen repeatedly across two decades of transactions, and what I now look for first.

1. The offtake is not really an offtake. An LOI is not revenue. Neither is an MOU, or a contract with conditions nobody has satisfied. Lenders size debt against contracted, creditworthy revenue with a tenor that supports the amortization. A famous name plus a non-binding term sheet is not an offtake. It is a hope with a logo on it. And a ten-year contract against twenty-year debt is not a name problem. It is a math problem.
2. Nobody has checked who is actually signing. The rated parent in the press release is very often not the entity on the contract. Frequently it is a subsidiary with no balance sheet and no rating. Where the obligation sits, and what credit support stands behind it, decides what the offtake is worth. This gets discovered late, and late is expensive.
3. Construction risk has no home. Fixed price or open book. Who absorbs delay. What the liquidated damages actually cap at, and whether the contractor could survive paying them. And whether anyone has ordered the long-lead equipment. Turbines, transformers and switchgear have been the schedule constraint for two years. No lender will underwrite a timeline that depends on equipment nobody has bought.
4. The capital structure is a picture, not a plan. Three colored bars on a slide is not a capital stack. A capital stack is a set of terms that named institutions have indicated they will provide, against coverage ratios someone has tested under downside cases. Until sponsor equity is committed and the debt has been sized by a party willing to lend it, there is nothing to approve.
5. The governance will not survive diligence. No project company. No shareholder agreement. Unclear decision rights. Undocumented related-party arrangements. One person holding everything, with no plan if they stop. Governance is almost never the stated reason a financing dies. In my experience it is very often the real one.
6. Nobody has decided who carries what. Construction all-risk and delay in start-up cover, and whether it is even available in the jurisdiction on acceptable terms. Then the larger question: who bears interconnection delay, curtailment, change in law, force majeure. Every risk in a project ends up somewhere. Unallocated, it does not disappear. It gets priced into the debt, pushed back to equity, or converted into a condition precedent that has to be closed before funding. All three cost the sponsor money or time, and the sponsor rarely sees the bill coming.
7. The model is an argument, and the argument has no support. Power prices, uptime, capex contingency, escalation, operating costs. A model is an analytical instrument, and a good one is essential. But it inherits the credibility of its inputs and nothing more. The question is whether those inputs are sourced, benchmarked, stress-tested and reviewed by someone with nothing to gain from the answer.
8. The evidence is not in the form the lender expects. Every material claim needs a document behind it. A capacity study, an independent engineer’s report, a legal opinion, a title report, a permit, an executed contract. Sustainability and regulatory evidence belongs inside this list rather than above it, and the standard shifts with the source of capital. A development bank, a European lender and a US private credit fund want materially different things. Know which one you are actually approaching, and bring what they require rather than what you happen to have.
9. The closing conditions have no owners and no order. Most stalled transactions I have seen do have a conditions list. Far fewer have that list with a named owner, an authority entitled to accept that each item is resolved, a date and a dependency mapped against every line. Without those four things, conditions get worked in the order somebody notices them instead of the order in which they gate funding. The schedule slips by months and no single decision caused it. And even where the list exists, evidence submitted against a line is routinely treated as the line closed. It is not the same thing. Closure requires evidence that meets a stated standard, accepted by whoever has the authority to accept it. Confusing the two is how a condition marked closed reopens three weeks before signing.

At Millentra, we maintain the governed record. The relevant lender, investor, regulator or other authorized institution retains authority to accept or reject closure.

What makes AI infrastructure different

Everything above applies to any project-financed asset. Five things are specific to this sector, and they are where I see the most avoidable damage right now.

The counterparty is often new. The fastest-growing buyers of AI capacity are compute providers with short operating histories, venture balance sheets and no rating. That is a legitimate business and a very different credit from an investment-grade hyperscaler. Financing structured as though the two are interchangeable does not survive committee.

The contract is short and the asset is long. Compute agreements are frequently three to seven years. Infrastructure debt wants fifteen to twenty. Something has to bridge that, whether it is residual value, re-contracting assumptions, sponsor support or a shorter tenor with refinancing risk priced in. Leaving the gap unaddressed is the single most common structural weakness I see in this sector.

The equipment ages faster than the building. A powered shell has a multi-decade life. The compute inside it does not. If your revenue case depends on a specific generation of hardware, you are asking a lender to finance long infrastructure against an economics that turns over every few years. That is solvable, but only if it is confronted in the structure rather than in the model.

Concentration is the norm, not the exception. Single-tenant campuses mean one counterparty, one contract and one point of failure. Diversification is the usual answer in infrastructure and it is frequently unavailable here. The alternative is credit support, and credit support has to be negotiated early, not discovered late.

And the one that gets missed most often: this is a project on a project. An AI campus with dedicated generation is not one financing. It is two developments bolted together. The data center depends on a power project with its own permitting, its own EPC, its own interconnection, its own schedule and its own lenders. Each is contingent on the other. Miss the sequencing and you have two projects that are each individually viable and jointly unfinanceable.

I spent years on exactly this structure in LNG-to-power, where a plant depends on a terminal, and the terminal depends on supply, and each financing waits on the one behind it. The lesson transfers directly. Interface risk between two linked projects is not a detail to resolve after both are developed. It is the structure. Decide it first, document it in both directions, and align the schedules and conditions precedent across them. Sponsors who leave this to the end lose a year and usually do not know why.

What good looks like

The teams I have seen close fastest are not doing more diligence. They are doing it earlier and in a different order.

• They treat institutional readiness as a design discipline rather than a late compliance exercise. Offtake structure, counterparty credit support, risk allocation and governance are decisions, not discoveries, and they are made while development is still running.
• They commission evidence in the sequence in which it gates funding, not in the sequence in which somebody thinks of it.
• They maintain a live conditions map with an owner, a date and a dependency against every item, and they know at any moment which three things are actually on the critical path.
• They pick the capital source before they build the package, and they build to that source’s standard rather than to a generic one.
• And when the honest answer is that the project is not ready, they say so early and fix it, rather than testing the market and burning the relationships they will need in six months.

None of that requires a finished project. All of it requires deciding to look at the second question while you are still working on the first.

The second gate

Finding sites, checking grid capacity and screening permitting used to take months. Now it takes days. That is real progress and the companies that delivered it deserve the credit.

The layer above it remains fragmented and largely manual. Establishing whether a developed project can actually be approved and financed is still often done manually, by advisers working in sequence, producing outputs in whatever format each prefers, against standards that change depending on who is writing the check. It is now the slowest step in the chain I work in, and increasingly the one that decides outcomes.

That is the layer we are building Millentra to address. Through the Project Readiness & Execution Diagnostic, we establish a Baseline Execution State: a governed record built from our Institutional Readiness & Bankability Assessment methodology, assessed across offtake and revenue, technical and delivery, market and commercial, permitting and regulatory, governance and sponsor readiness, and ESG and institutional alignment. Every material finding links to the evidence behind it. Every condition blocking approval is named, owned and sequenced in an Institutional Conditions Register, with a record of what closes it and who has authority to accept that it has. We deliver it today as a methodology-led engagement with human review, while the software to scale it remains in development.

The Diagnostic establishes that baseline. The projects in this article’s opening paragraph did not fail a diagnostic — they went quiet afterward, the way a data room stops being updated once nobody is required to keep it current. Institutional readiness does not stay static once established: evidence changes, dependencies move, and conditions can reopen. Where continuing governance is warranted, we keep the record current through the decision: evidence reviewed, conditions accepted, rejected or reopened, and the state versioned as it moves.

We do not replace the advisers you already trust. We establish what has to be resolved, in what order, and to what standard.

An illustrative application of this framework, built against a sample project, is available on request.