How DeepStack settles a question · Part 06
Is power scarcity a moat?
Is power scarcity a warning, or a moat?
From Part 06: Power Doesn’t Depreciate. Its Premium Moves.. 7 voices on 2 sides, attributed and dated; the DeepStack Call with its history; 5 dated tests that settle it; the results as they come in.
The mechanism
Electricity is the constraint, but the premium belongs to whoever can turn a contracted megawatt into revenue on time. Compute can be ordered faster than power can be delivered: NVL72 deployment at full use needs about 23 gigawatts by 2027, while grids, substations and transformers take years to expand, and Nebius quotes $40 to 50 million for six months of capacity against $20 to 25 million for one to three years. A megawatt generates rent only when it arrives on schedule; a GPU parked behind a late interconnection generates cost. That is why the Call reads scarcity as a moat for the permit and the interconnect rather than for the hardware, and why it would change if the premium narrowed while power stayed scarce or a hyperscaler canceled because compute was sufficient.
Approved by Marco Cauduro on October 10, 2026.
The DeepStack callConstructive
This is a power-access trade before it is a chip trade. Own the permit and the interconnect, not the loudest fleet.
Our Call takes the side “Scarcity as a moat”.
Call history
- v2Long access, not hardwareConviction High (4/5)Side: Scarcity as a moatSide named by the owner.
- v1Long access, not hardwareConviction High (4/5)First call, as published with the story. The conviction, the actions and the flips were set down for the site on October 7, 2026.
7 voices
Scarcity as a warning
Nebius's own disclosed terms show compute in backwardation: 20 to 25 million dollars per megawatt on one-to-three-year contracts against 40 to 50 million per megawatt for capacity of up to six months; buyers paying twice the rate for near-term use…
GDP growth is heavily reliant on the data center buildout, evidenced by private non-residential construction excluding data centers falling 7.9% year over year in June.
CoreWeave is a contract-financed infrastructure company whose assets become technologically obsolete much faster than the buildings it leases, whose largest customers are competing clouds, and whose only GPU supplier is also an investor, customer…
Profitability requires mature deployments to fund the next build without another financing wave, a condition that has not yet been met and that revenue, backlog, and gigawatts rising alone do not prove common equity earns an attractive return after…
Hyperscalers face hard constraints on grid access, permitting, and environmental impact that are pushing them toward distributed multi-building or multi-campus architectures rather than single massive data centers in one location.
Carriers had already installed enormous amounts of fiber based on expectations that demand would keep outrunning supply, so when DWDM allowed each fiber to carry dramatically more capacity than expected, the remaining fiber pairs sat dark for years…
Data centers are proving more difficult and costly to build than expected, with gigawatt-scale projects running behind schedule, and every delay gives an already-unprofitable customer base more time to encounter financial strain.
6 voices
Scarcity as a moat
AI data center buildout is bottlenecked by power delivery, not compute, since compute can be ordered faster than power can be delivered to a site.
Bloom's real product is time-to-power rather than electrons, because stranding a multi-billion-dollar GPU cluster for 18-24 months costs far more than paying a premium for onsite electricity.
The mismatch between how quickly compute can improve and how slowly physical infrastructure can be built is what drives the investment case for data-center stocks, because interconnection, transformers, turbines, generation, transmission, and…
Power is a key bottleneck in AI because rack power is rising sharply while grids, substations, transformers, and generation take years to expand.
Despite extensive resident opposition on noise, water, and emissions grounds at the planning board hearings, Nebius's 350 MW Vineland, New Jersey data center site plan was still approved, showing there is no real 'war' against data centers.
Because a single data center campus represents a multi-billion dollar capital commitment, multi-year construction payroll, and permanent tax base addition, while incentives amount to only about 2% of construction investment in hyperscale counties…
Where they cross
Who answers whom.
The specific points where one voice meets another: the same evidence read two ways, or the same mechanism with a different sign.
Edelbridge Alpha qualifies Daniel Romero, on a condition
Steve Eisman qualifies Gaetano, on the evidence
Edelbridge Alpha qualifies Daniel Romero, on the mechanism
Daniel Romero qualifies Ren, on the mechanism
Steve Eisman qualifies Daniel Romero, on a condition
Gaetano qualifies Daniel Romero, on a condition
Edelbridge Alpha qualifies Gaetano, on the mechanism
Edelbridge Alpha qualifies Gaetano, on a condition
Alternative explanations
Three other readings.
What would have to be true for the Call to be wrong, as the other side argues it.
- Permission is the asset politics can revoke: 70 to 75 percent of Americans oppose large data centers near their homes and more than half back a national moratorium (the case Michael Spencer makes).
- Scarcity is the warning fiber gave in 1999: capacity built on the expectation that demand keeps outrunning supply goes dark once a technical step multiplies what each unit can carry (the case Gaetano makes).
- The moat is financed, not owned: the builder's assets go obsolete faster than the buildings it leases, its largest customers are competing clouds and its only chip supplier is an investor (the case Kakashii makes).
What settles it
The dated tests.
The tests the story set, each with its specification where the owner has written one and the piece of the argument it verifies; on the Docket, resolved on the Results page.
ERCOT is due to verify its first large-load batch study; the cleared list shows whether Texas weeds out speculators or rations power.
Part 06: Power Doesn’t Depreciate. Its Premium Moves. Pendingp06-t01Open questionChecks the speculative-load claim
Specification pending
Nebius must turn 800MW–1GW connected against 5GW contracted into Vineland revenue; commissioning and GPU fill test execution.
Part 06: Power Doesn’t Depreciate. Its Premium Moves. Pendingp06-t02 · monthOpen questionChecks the delivery claim
Specification pending
- Next quarterly count
A second quarter near 26 blocked projects would turn a single data point into a permitting trend.
Part 06: Power Doesn’t Depreciate. Its Premium Moves. Pendingp06-t03Open questionChecks the permitting-trend claim
Specification pending
- First cancellation
A hyperscaler cancelling for sufficient compute would turn the fiber analogy into evidence of overbuild.
Part 06: Power Doesn’t Depreciate. Its Premium Moves. Pendingp06-t04Open questionChecks the overbuild claim
Specification pending
Bloom staying sold out while FuelCell Energy keeps signing would show scarcity has moved from the grid to factories.
Part 06: Power Doesn’t Depreciate. Its Premium Moves. Pendingp06-t05Open questionChecks the on-site power claim
Specification pending
What changed
The record of Part 06.
The Part’s publication, every change to the Call, every result that came in and the next dated test, oldest first.
- PublishedPart 06 published; the Call: Long access, not hardware.
- Call updatedThe Call: Long access, not hardware. This is a power-access trade before it is a chip trade. Own the permit and the interconnect, not the loudest fleet. Our Call takes the side “Scarcity as a moat”. by Marco Cauduro
- Next testNext: Dec 10, 2026, ERCOT is due to verify its first large-load batch study; the cleared list shows whether Texas weeds out speculators or rations power. Open question
Sources (11)
- Financial Modeling Prep - Nebius Q2 earnings call
- Financial Modeling Prep - CoreWeave Q2 earnings call
- a16z - The Machine Age Fund
- a16z - Data-center capital charts
- Prof G Markets - Michael Burry and a crash
- Prof G - Why Panic About Data Centers?
- Doomberg - Value Trap
- Hypertech Invest - Data Center Positions
- AI Supremacy - Data-center opposition
- a16z - Head In The Neoclouds
- Prof G - Tyler Cowen interview
Market data are as of the dates cited. Voices are paraphrased from public writing and attributed by name.
Analysis and opinion, not investment advice. Voices are paraphrased from public writing and attributed by name; DeepStack shows no score, ranking or accuracy for any person. Every question, one page each.