A site-by-site reconciliation of announced pipelines against energized capacity - delays do not destroy demand, they shift it to the right
The real accounting unit of an AI data center is energized GW. The order-of-magnitude gap between announcement (tens of GW) and energization (single-digit GW) is not evidence of a bubble but evidence of physical bottlenecks - transformers, turbines, the grid - and delays do not destroy demand but shift it to the right, prolonging the memory and power shortage.
Crusoe's June 9 suspension of its 1.8GW Wyoming campus (just before groundbreaking, after permitting was complete), Bloomberg's tally of 'nearly half of 2026 US construction delayed or canceled,' and Oracle's FY26 Q4 results scheduled after the market close on the publication date are the first official test of delivery-curve verification.
Beneficiaries: memory (SK hynix, Samsung Electronics, Micron - extended shortage), power equipment (Hyosung Heavy Industries, HD Hyundai Electric, GE Vernova - bottleneck premium, backlog preservation), holders of energized capacity. Largest risk: Oracle-type delivery-dependent balance sheets holding pre-energization commitments as debt ($553 billion RPO, more than half to OpenAI as a single customer) and CoreWeave-type GPU-collateral leverage. The only sell signal: simultaneous occurrence of delay + capacity reservation cancellation + memory contract price break.
On June 9, Crusoe announced it was pausing development its 1.8GW Cheyenne, Wyoming campus (Project Jade, designed for expansion up to 10GW) 'at the customer's request.' In the same announcement it disclosed about 4.9GW of contracted capacity and a pipeline that dwarfs it, along with a new 900MW groundbreaking for Microsoft in Abilene, Texas. For the pause of what could have become the largest single AI campus in the US, the announcement structure itself was a message of 'we have no problem.'
The information value of this case lies in the timeline. July 2025 announcement - January 2026 unanimous local approval - Q1 groundbreaking scheduled - June 9 Crusoe pause. It stopped at the stage just before groundbreaking, after all permits had been obtained. Tallgrass (a Blackstone affiliate) invested $7 billion in infrastructure alone, and total project cost was estimated at up to $50 billion.
But one day after the announcement, the entire picture flipped. On June 10, the Cheyenne power company Black Hills Corp. (NYSE: BKH) nailed it down in an official announcement (GlobeNewswire) - the 1.8GW project has not been halted and proceeds as planned, Crusoe is no longer the development partner of this project, the company is working directly with a potential large-load power customer (a hyperscaler), and it targets initial operation in early 2028. The remarks are from CEO Linn Evans. Per reporting, that customer is understood to be OpenAI, and the direction is to build directly, bypassing the developer. Crusoe's 'pause,' then, was not a halt of the project but Crusoe's own exit.
The background context matters too. The Cheyenne area has up to 70 data center proposals clustered, and local authorities are considering a 12-month moratorium (a temporary halt on development). At the time Crusoe announced the pause the tenant was undisclosed, and waiting for a chip generation, volume relocation, and funding schedule adjustment were floated as hypotheses. The June 10 power-company announcement gave the answer - the project and the demand are intact; what disappeared was a single middle developer, Crusoe.
The essence of this case is therefore not a 'pipeline-to-contract conversion failure.' It is an event where only the middle developer was replaced while the project and demand stayed intact - disintermediation. This actually strengthens this report's thesis that 'demand does not disappear but moves.' It is also the first case in which the risk of financing Type 2 (through a developer + long-term lease) materialized not as 'tenant exit' but as 'the customer excluding the intermediary.' When the customer (a hyperscaler) is large enough, it builds directly with the power company rather than going through the developer.
The criteria for separating individual suspension news into noise and signal are stated explicitly. A first-tier signal is a pipeline-to-contract conversion failure - be wary if frequency rises, but on its own the relocation possibility prevails. A second-tier signal is renegotiation or reduction of firm contracts and - this is a crack in the commitment structure and is a target for immediate response (current specimen: one report in March of OpenAI exiting an Oracle Stargate expansion). A third-tier signal is the breaking of memory contract-price momentum - if capacity is 'sold out' but prices break, it means quiet cancellations are underway, and this is the final warning light that Korean investors can read fastest.
Where does the Crusoe case sit in this frame. As of June 9 it looked like a first-tier-signal candidate, but the June 10 power-company announcement forces a re-judgment - this is not a first-tier signal but a value-chain structural signal. Read as a demand signal it is noise (demand is intact); read as a developer-risk signal it is valid (a developer can be excluded wholesale).
The Crusoe case also exposes the reliability problem of the 'pipeline' figure. For the same announcement, Bloomberg reported the total pipeline as exceeding 40GW, while other coverage put it above 20GW - a figure that doubles depending on how far you sum contracts, negotiations, and advanced development. Pipeline is neither subject to audit nor to disclosure obligation, so the announcing party defines it arbitrarily. This is why this report excludes pipeline from demand indicators and puts only firm contracts and on the ledger, and recommends readers classify 'pipeline XX GW' headlines as marketing figures.
The Crusoe case revealed its identity one day after the announcement - not a demand halt but a developer exclusion. We exclude it from the first-tier signal count, but record it as a value-chain signal that 'when the customer is large enough, the middle developer is skipped.' Demand did not disappear. Only its seat changed.
Sign in with Google — your tier will be checked automatically and access granted if eligible.
Sign in with GoogleThis site runs on ads — the tier system rewards community contributions.
This report is provided for informational purposes only and does not constitute a recommendation to buy or sell any financial instrument. Investment decisions should be made based on your own judgment and responsibility. The analysis and opinions contained herein are based on information available at the time of writing and are subject to change.