Thesis · Memory & Storage
Will Nvidia's data center revenue growth rate for any quarter reported between 2026-11-15 and 2027-12-31 be explicitly attributed in its earnings call to memory (HBM) supply constraints?
Memory, not GPU die or foundry capacity, is the binding constraint, and Nvidia management will publicly cite HBM/memory supply as a limiter on shipments or product configuration.
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
HBM is the binding constraint
HBM stacking yields and a three-supplier oligopoly cap bits available per accelerator while TSMC advanced packaging and die supply expand faster. Accelerator vendors trim memory configurations or guide below demand, attributing the gap to memory supply rather than wafer or CoWoS capacity. Customers reportedly request far more HBM capacity than current expansion plans can supply, giving memory suppliers pricing leverage, and this shows up as Nvidia naming memory supply on an earnings call.
The case for no
Nvidia will not name memory supply as the binding constraint
Yield and stacking economics keep DRAM a commodity-process constraint rather than a named earnings-call limiter, and contracts that trade peak pricing for duration lengthen visibility and lower the cyclicality that would force a public attribution. Even accounts of tight HBM server availability treat it as a near-term revenue drag addressed by a roadmap toward hybrid servers with alternative memory, not as something Nvidia singles out by name. On this side, Nvidia instead cites advanced packaging, foundry wafers, power, or no single named constraint when explaining any quarter's growth rate.
5 voices
HBM is the binding constraint
Because memory supply cannot keep pace with AI demand, companies like Nvidia must scale back planned specifications (e.g., Vera Rubin with less memory) and guide lower growth than they otherwise could, while memory suppliers like Micron benefit…
The demand-supply imbalance in HBM memory creates a classical Economics 101 scenario where constrained supply meets accelerating demand. Customers requesting 5x–6x more capacity than current expansion plans can supply creates a pricing lever that…
Sands Capital
Growing AI demand, particularly for high-bandwidth memory in AI servers and for conventional memory in AI inference workloads, strains current supply. Memory industry consolidation into an oligopoly limits capacity expansion speed, allowing…
HBM requires Advanced MR-MUF molding that keeps 30-micron wafers from warping, a base die compatible with TSMC process, and sixteen-stack assembly where losing one ruins the whole part. Yield is earned through process learning, not specified…
Leading AI labs generated a revenue run rate over $80B, four times the prior year, creating enormous compute demand. This led to specific demand for high bandwidth memory, which consumes three times the capacity of lagging edge memory and created…
3 voices
Nvidia will not name memory supply as the binding constraint
Ray Wang
Yield and stacking economics keep wafers in commodity DRAM
Early quarter weakness from limited HBM server availability constrains revenue realization, but management's roadmap toward hybrid servers with alternative memory reduces future dependency on the scarce input, allowing revenue to scale less…
Contracts trade peak pricing for duration: the premium per bit compresses but earnings visibility lengthens, which lowers cyclicality risk and justifies a higher multiple on memory names like Micron, further helped by Micron regaining ~10% share in…
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.
Sandisk, operating cash flow as a share of revenue, quarter ended October 2, 2026: whether it comes in at or above 74%.
On the Docket Pendingth059-t01Open question
Specification pending
The machinesClaude 48% · DeepSeek 38%
Your numberSign in by email to forecast
Sandisk, operating cash flow as a share of revenue, quarter ended October 2, 2026: whether it comes in at or above 78%.
On the Docket Pendingth059-t02Open question
Specification pending
The machinesClaude 22% · DeepSeek 7%
Your numberSign in by email to forecast
Sandisk, operating cash flow as a share of revenue, quarter ended October 2, 2026: whether it comes in at or above 82%.
On the Docket Pendingth059-t03Open question
Specification pending
The machinesClaude 22% · DeepSeek 4%
Your numberSign in by email to forecast
Sandisk, gross margin, quarter ended October 2, 2026: whether it comes in at or above 80%.
On the Docket Pendingth059-t04Open question
Specification pending
The machinesClaude 18% · DeepSeek 4%
Your numberSign in by email to forecast
Sandisk, gross margin, quarter ended October 2, 2026: whether it comes in at or above 83%.
On the Docket Pendingth059-t05Open question
Specification pending
The machinesClaude 8% · DeepSeek 3%
Your numberSign in by email to forecast
Sandisk, gross margin, quarter ended October 2, 2026: whether it comes in at or above 86%.
On the Docket Pendingth059-t06Open question
Specification pending
The machinesClaude 4% · DeepSeek 2%
Your numberSign in by email to forecast
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.