The unpriced deal is now infrastructure's default language

Zankore's $3.1 billion loan to an Indonesian AI cloud venture went out against a shareholder list, while Invenergy and HASI put 2.7 gigawatts of American renewables on the tape with no published capital stack behind them and Alcazar closed a 131-megawatt wind financing whose coverage carries no tariff, no offtake counterparty and no lender. Three announcements from three corners of the same business, each missing the terms that would let a lender check the arithmetic.

Announcing before pricing is an old habit in renewables, which is why a withheld spread on a wind close usually draws a shrug. The habit has moved, and the same missing terms now turn up in GPU-backed lending, options over land and power, hyperscaler pre-sales, and refinancings, the one transaction type where a price should be public because the price is the reason the transaction exists at all. GPU debt has spent its short life borrowing the vocabulary of infrastructure finance; a refinancing is where that vocabulary is supposed to turn into a number.

A refinancing owes a reader the spread, because a refinancing reprices debt that already exists. Umbrella published the size of a €1.9 million small-solar refinancing in Spain and left the counterparty out, which is a small loss on a deal that size and a telling habit at any size, while Sonnedix's $1.3 billion Chilean refinance arrived with a size and no terms. When the spread stays out of the record, the only fact left standing is that a lender showed up. The market reads that as validation; it is better read as a data point about demand rather than about price.

A pre-announcement is allowed to be thin: nothing is financed yet, and a developer holding a site option and a grid application has little else to publish. The informative cases this week are the ones where a gate has already been passed: Alcazar's wind financing is described as closed, Ohio's regulators cleared 149 megawatts of solar-plus-storage at Hamden Energy without a dissent, and Sonnedix refinanced $1.3 billion in Chile. In each, the market gets the milestone and not the math.

The energy-transition sequence in PWD's deal log shows how routine the omission has become: Invenergy and HASI announced 2.7 gigawatts with a count and nothing an underwriter could use, Vattenfall committed to build in Germany without a site, a megawatt rating, a connection point or a buyer, and Sungrow says it will assemble 10 gigawatt-hours of batteries in Egypt by mid-2027 without the coverage saying what the plant costs. A Namibian fund backed green fertiliser without naming the fund, the figure or the instrument, leaving the project upstream of the money it needs.

Germany's backup-capacity tender is the cleanest case because it is an auction: oversubscription proves there are bidders, but a clearing price is what proves there is a deal. Germany has published the first and not the second, leaving a queue of firm-capacity projects whose developers can point to demand and cannot yet point to a revenue line.

Ohio runs the same experiment at smaller scale, where a unanimous approval retires the consent risk on 149 megawatts and leaves the offtake question that decides whether the panels are ever ordered, a reminder that a permission and a power contract are granted by different people and need not arrive together.

The smaller notices carry the same blank: Hyperion's Romanian solar milestone arrives with a capacity and a country and no price, Plenitude brought 19 megawatts of Italian solar online with no partner, offtaker or price in the record, and Fortis moved 375 megawatts-peak of Balkan solar and storage without naming a buyer. Feldberg's 6 gigawatt-hours of solar offtake, aimed at property-owner demand, was reported without a generator, a term or a price, which makes it a reading on demand rather than an underwriting event.

It is a disclosure habit: firms publish the facts that establish position and withhold the facts that would let a rival or a lender build a comparable. What changes is which facts fall on which side of that line, and this week the line runs between a consent and a cash flow. The likely consequence is thinner comparables in exactly the places capital is being raised fastest: if a 2.7-gigawatt announcement, a $1.3 billion refinancing and a $1 billion pre-sale all land without a spread, the next deal in each category has less to be priced against.

The announcements do keep two things, consistently enough that the pattern deserves a name: consent arrives, and queue position gets claimed. Buckinghamshire's Wycombe approval covers six data halls with no stated power, granted on recycled employment land where planning is cheap and the record never has to say what will run inside the buildings.

Bucharest makes the point through a grid request: S+B's 12-megawatt plan turns on access to power rather than floorplate, the right order of operations in a city already hosting at least 31 data centers against a Romanian pipeline announced at 80 megawatts and 70 megawatts. In Lower Saxony, a zoning exemption at a hydrogen park lets Friesen Elektra monetise power rights instead of building data centers, making the wires, rather than the buildings, the gated item in northern Germany.

Read the consent stories together and the sequencing is what stands out: a permit for six data halls can be secured before the power contract, and a zoning exemption can be worth more than the building it was granted for. Britain spent the week pushing the same decision down to mayors, ruling out a national ban and leaving consent to be bargained site by site, where the currency is likely to be local spend, a formula that preserves a national permission while dropping national certainty and produces a slower and more expensive permitting regime inside the same framework.

The sharpest exhibit is an option: BTC Digital's two-year, non-binding memorandum with two Vietnamese partners buys it a place in the Hai Phong power queue for no capacity figure and no stated dollar commitment. Read that as a purchase of position rather than of assets, a cheap claim on time inside a queue that is filling in a market where the queue itself is becoming the scarce item.

The deal that matters most to the financing market is Zankore's $3.1 billion sponsor-underwritten loan, secured against its shareholder list and framed in the coverage as a test of 2026's GPU-debt trade. Size is the least informative number in any loan: it says how much paper exists, not what the paper was underwritten against, who carries the risk in year five, or what the security is worth if the rental market moves. A shareholder list as collateral shifts the question to the sponsors and their willingness to keep funding, which is a different credit from the one a signed data-center lease would support.

Nvidia's Australian target is the same wager written in gigawatts: eight partners, 2 gigawatts by 2027, no named tenant. Building merchant capacity ahead of the customer inverts the sequence of a conventional data-center deal, and only a sponsor that can carry a site through however long a connection takes can afford the inversion, so a grid queue, not a hyperscaler's signature, sets the timetable and the first revenue depends on a tenant the announcement does not name.

Bain-backed Hscale's $1 billion contract for Spain was signed with a hyperscaler whose name is not in the record, so the pre-sale trades on the category rather than on the counterparty, and 2027 becomes the year a contract has to become an operating asset. Ericsson's private-5G deployment at an airport works operationally and arrives with no disclosed contract value, term or ownership model, which leaves it a reference site for a next sale rather than an asset anyone can underwrite.

The one large number published in full this week was Kevala's finding, prepared for the Coalition for Community Solar Access, that California's grid can host 15,560 megawatts of front-of-meter solar and storage, a technical ceiling rather than a transaction, which is precisely why it could be printed: nobody's negotiating position moves when the number belongs to a grid.

The first spread will have to land somewhere. Germany's tender has to clear at a price, Alcazar's next financing has to name a lender, and the holders of Zankore's paper will eventually have to mark it against something other than the sponsors' capacity to raise. Three of this week's blanks carry the same deadline: Nvidia's 2 gigawatts are targeted for 2027, Sungrow's Egyptian lines are due to assemble 10 gigawatt-hours by mid-2027, and Hscale's Spanish contract has to be an operating asset that year. Until one of those dates produces a number, the market will keep bidding for the two things the announcements actually delivered: a consent and a place in line.