Thesis · Hyperscalers, Cloud & Compute Services
Will Nvidia report a year-over-year decline in data center revenue in any fiscal quarter reported before December 31, 2027?
The capex wave pauses and Nvidia's data center revenue declines year over year at least once by end-2027 as the overshoot works through.
Who is on each side
The two sides.
Each side in DeepStack’s words, then the voices who argue it, attributed and paraphrased, dated where the date is known. A voice links to its page on DeepStack where one exists.
The case for yes
The buildout overshoots demand
Hyperscaler orders placed out of scarcity fear accumulate as undeployed capacity rather than tracking actual usage, so when the panic buying stops, component prices drop and the excess must be absorbed. Aggregate capital spending projections may not match actual demand growth, and if compute prices fall faster than anticipated, the buildout proves oversized. Nvidia's data center revenue depends on that order flow continuing to grow, so a digestion phase causes it to decline year over year in at least one reported quarter before 2027 ends.
The case for no
Scarcity keeps the buildout from overshooting
Shortages of chips, skilled labor such as plumbers and electricians, and power keep builders from deploying capacity faster than demand can absorb it, so the market cannot reach overcapacity despite high demand. Global internet-era distribution lets new AI capability diffuse to users far faster than in prior technology cycles, so spenders can observe absorption and size commitments accordingly rather than overbuilding blindly. On this view, Nvidia's data center revenue keeps growing year over year in every quarter reported through 2027.
3 voices
The buildout overshoots demand
No hyperscaler can afford to slow down alone: slowing risks losing power, chips, talent, model partners, customers and the next computing platform while AI threatens their core cash-cow businesses; supply scarcity means capacity secured by one is…
Because the binding constraint is deployed datacenter capacity rather than component supply, orders placed out of scarcity fear accumulate as undeployed inventory; when buying panic stops, component prices drop and the stockpiles must be written…
Hyperscalers and specialists are spending massively on data center buildout based on projections of AI demand, but if actual demand growth does not match the buildout pace or if compute prices collapse faster than anticipated, the aggregate capital…
2 voices
Scarcity keeps the buildout from overshooting
Barry Ritholtz
Ankur Crawford argues that if there were an infinite supply of chips, people and power, the market would already be in overcapacity because everyone would build as fast as they wanted; instead, the shortage of chips, workers like plumbers and…
Earlier booms busted because diffusion trailed the mobilization of capital, leaving owners unable to extract expected benefit; with AI adopted in days rather than years, capex spenders can observe how quickly new capacity is absorbed and size…
What would settle it
The dated tests.
Each test names the public source that settles it, the rule and the day. The Call’s number, the readers’ median and the machines’ numbers stand beside it, as on the Docket.
CoreWeave, purchases of property and equipment as a share of revenue, quarter ended September 30, 2026: whether it comes in at or above 190%.
On the Docket Pendingth039-t01Open question
Specification pending
The machinesClaude 55% · DeepSeek 84%
Your numberSign in by email to forecast
CoreWeave, purchases of property and equipment as a share of revenue, quarter ended September 30, 2026: whether it comes in at or above 245%.
On the Docket Pendingth039-t02Open question
Specification pending
The machinesClaude 45% · DeepSeek 57%
Your numberSign in by email to forecast
CoreWeave, purchases of property and equipment as a share of revenue, quarter ended September 30, 2026: whether it comes in at or above 290%.
On the Docket Pendingth039-t03Open question
Specification pending
The machinesClaude 33% · DeepSeek 32%
Your numberSign in by email to forecast
Oracle, operating cash flow as a share of revenue, quarter ended November 30, 2026: whether it comes in at or above 95%.
On the Docket Pendingth039-t04Open question
Specification pending
The machinesClaude 13% · DeepSeek 2%
Your numberSign in by email to forecast
Oracle, operating cash flow as a share of revenue, quarter ended November 30, 2026: whether it comes in at or above 117%.
On the Docket Pendingth039-t05Open question
Specification pending
The machinesClaude 6% · DeepSeek 2%
Your numberSign in by email to forecast
Oracle, operating cash flow as a share of revenue, quarter ended November 30, 2026: whether it comes in at or above 135%.
On the Docket Pendingth039-t06Open question
Specification pending
The machinesClaude 5% · DeepSeek 1%
Your numberSign in by email to forecast
In The Buildout
The Parts on the same question.
The stories that take up this thesis, with their own Calls and tests.
Part 01: Nvidia’s Demand Is Real. Its Financing Is Fragile.
Nvidia’s demand looks real today, but private credit, slow-to-build working compute and fast obsolescence make the revenue recurrence less secure than the backlog suggests.
Part 02: SpaceX Is Funding an AI Bet Before the Proof
SpaceX’s IPO funds a credible AI buildout, but the stock is pricing future capacity before deployment, revenue quality and Starlink coverage are proven.
Part 07: The Fed Hiked. The Long End Still Climbed.
The Fed raised short rates, but long yields climbed. AI financing now sits inside a bond market priced by growth, deficits and a thinner buyer base.
Part 10: The Bubble Is in the Denominator
AI spending made the market look cheaper. The harder test is whether those profits survive depreciation, weak productivity and a bond market financing the buildout.
A thesis page states one question and both sides of it; DeepStack takes no side in words, and its view is the number it gives each test before the date. 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. Pre-specified means the words were approved on or before the test’s date. Specified after the date means they were written afterwards: such a test is counted and shown, and never rated. An open question has no words yet. The Call’s number is the chance, from 0 to 100, that the Call gives the yes answer to a test’s question, recorded before the test’s date; it is DeepStack’s own reading, not a market price and not advice. The readers’ mark on the same line is the readers’ median, the middle of the numbers signed-in readers gave the test, shown from 25 readers. The machines: Each model read this page and nothing else. The number that counts is the latest one each model gave before the test closed.