Thesis · Hyperscalers, Cloud & Compute Services

Will NVIDIA state by Nov 30, 2027 that inference is over 50% of its data center compute demand?

As AI shifts from building models to running them in everyday products and agents, inference calls could outgrow training runs. The question is whether NVIDIA says so on the record.

6 dated tests · 0 resolved · next on

The tests More theses in Hyperscalers, Cloud & Compute Services

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

NVIDIA will say inference is over half of demand

Rising usage converts directly into recurring compute demand once models are trained, and agentic systems that complete tasks and call tools multiply the number of inference calls per user action. As companies deploy agentic coding assistants and other always-on agents that submit requests far faster than humans, token and compute demand for running models keeps climbing relative to training. This side expects that trajectory to lead NVIDIA itself to confirm, in an earnings release, 10-Q/10-K, or official call transcript on or before November 30, 2027, that inference exceeds half of its data center compute demand.

The case for no

NVIDIA will not make that statement

NVIDIA does not make this disclosure in any dated public filing or call by the deadline. This could happen because the company keeps crediting large frontier training clusters for most of its data center growth, because 100k+ GPU training buildouts keep expanding faster than inference capacity, or because falling inference unit costs keep total inference spending below training spending even as usage grows. Training workloads concentrate in expensive, long-term contracted capacity of the kind hyperscalers and neoclouds build out explicitly for that purpose, a pattern that keeps training's share of disclosed compute demand visible and large.

6 voices

NVIDIA will say inference is over half of demand

  1. Aurelion Research

    Once models are trained they must run on every use, so rising usage converts directly into recurring compute demand; agentic systems that complete tasks and use tools multiply the number of inference calls per user action.

    Source

  2. Hf Best Ideas

    Baron argues inference is now the dominant and fastest-growing part of AI compute, that a fast-inference sub-segment has emerged, and that customer commitments rather than benchmarks validate the architecture's advantage.

    Source

  3. Gaetano

    Training is compute-bound and GPU-focused, but inference is memory-bound and latency-sensitive. As AI moves into production across enterprises, the full system around the GPU matters more: memory capacity, context length, responsiveness, and data…

    Source

  4. Tae Kim

    Companies deploy agentic coding assistants and other AI agents that operate continuously without rest and submit tool requests far faster than humans, creating competitive pressure to adopt more agents for R&D and operations. This raises token…

    Source

  5. Daniel Romero

    Inference will be the largest driver of compute demand in the AI era. AMD's MI300X and upcoming MI355X and MI400 series excel at inference workloads and are gaining design wins with hyperscalers and cloud providers. As inference demand grows much…

    Source

  6. Alex Dc

    Because the vast majority of an AI model's operational life is spent in inference rather than training, and NVIDIA GPUs are irreplaceable for this phase, training cost reductions do not eliminate the need for NVIDIA's hardware, preserving the…

    Source

1 voice

NVIDIA will not make that statement

  1. Kerrisdale Capital Management LLC

    Training workloads require high-end GPUs and can be run remotely, aligning with CoreWeave's model of renting expensive, cutting-edge capacity on long-term contracts. Inference, by contrast, is steady-state, latency-sensitive, and cost-conscious…

    Source

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.

  1. CoreWeave, purchases of property and equipment as a share of revenue, quarter ended September 30, 2026: whether it comes in at or above 220%.

    On the Docket Pendingth004-t01Pre-specified

    Question
    Did the purchases of property and equipment of CoreWeave for the quarter ended September 30, 2026 come to at or above 220.0% of its revenue?
    Source
    SEC EDGAR, XBRL company facts of CoreWeave (Forms 10-Q and 10-K)
    Rule
    Divide the purchases of property and equipment by the revenue that CoreWeave reports for the quarter ended September 30, 2026 in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if the ratio is at or above 220.0%; no if it is below 220.0%.
    If the source is late, revised or silent
    If CoreWeave has not filed by November 16, 2026, or a filing restates a period the rule reads, the test is inconclusive, and the Result says what was read and what was missing.
    Base rate
    CoreWeave's purchases of property and equipment as a share of revenue was 249.4% for the quarter ended June 30, 2026, by its filing on SEC EDGAR.
    Outside view
    A year earlier, for the quarter ended June 30, 2025, the same line read 202.4%, so the rungs sit on the last print.
    What would change the number
    A change in product mix, a one-time charge, or a reclassified line would move the reading.

    The Call, before the dateExpects yes. Gives the yes answer 72%.

    72%

    0 · no100 · yes

    The machinesClaude 55% · DeepSeek 86%

    Claude: 55%DeepSeek: 86%
    The ladder 3 rungs on one reading, Nov 9, 2026: each asks whether it lands above its threshold, so the Call’s numbers down the rungs sketch where it expects the reading to land.

    1. Above 220%Call 72% · Claude 55% · DeepSeek 86%

      72%

    2. Above 245%Call 50% · Claude 45% · DeepSeek 55% · GPT 42%

      50%

    3. Above 265%Call 30% · Claude 30% · DeepSeek 64% · GPT 32%

      30%

    0 · no50100 · yes

    Your number

  2. 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 Pendingth004-t02Pre-specified

    Question
    Did the purchases of property and equipment of CoreWeave for the quarter ended September 30, 2026 come to at or above 245.0% of its revenue?
    Source
    SEC EDGAR, XBRL company facts of CoreWeave (Forms 10-Q and 10-K)
    Rule
    Divide the purchases of property and equipment by the revenue that CoreWeave reports for the quarter ended September 30, 2026 in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if the ratio is at or above 245.0%; no if it is below 245.0%.
    If the source is late, revised or silent
    If CoreWeave has not filed by November 16, 2026, or a filing restates a period the rule reads, the test is inconclusive, and the Result says what was read and what was missing.
    Base rate
    CoreWeave's purchases of property and equipment as a share of revenue was 249.4% for the quarter ended June 30, 2026, by its filing on SEC EDGAR.
    Outside view
    A year earlier, for the quarter ended June 30, 2025, the same line read 202.4%, so the rungs sit on the last print.
    What would change the number
    A change in product mix, a one-time charge, or a reclassified line would move the reading.

    The Call, before the dateEither answer fits the Call. Gives the yes answer 50%.

    50%

    0 · no100 · yes

    The machinesClaude 45% · DeepSeek 55% · GPT 42%

    Claude: 45%DeepSeek: 55%GPT: 42%

    Your number

  3. CoreWeave, purchases of property and equipment as a share of revenue, quarter ended September 30, 2026: whether it comes in at or above 265%.

    On the Docket Pendingth004-t03Pre-specified

    Question
    Did the purchases of property and equipment of CoreWeave for the quarter ended September 30, 2026 come to at or above 265.0% of its revenue?
    Source
    SEC EDGAR, XBRL company facts of CoreWeave (Forms 10-Q and 10-K)
    Rule
    Divide the purchases of property and equipment by the revenue that CoreWeave reports for the quarter ended September 30, 2026 in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if the ratio is at or above 265.0%; no if it is below 265.0%.
    If the source is late, revised or silent
    If CoreWeave has not filed by November 16, 2026, or a filing restates a period the rule reads, the test is inconclusive, and the Result says what was read and what was missing.
    Base rate
    CoreWeave's purchases of property and equipment as a share of revenue was 249.4% for the quarter ended June 30, 2026, by its filing on SEC EDGAR.
    Outside view
    A year earlier, for the quarter ended June 30, 2025, the same line read 202.4%, so the rungs sit on the last print.
    What would change the number
    A change in product mix, a one-time charge, or a reclassified line would move the reading.

    The Call, before the dateExpects no. Gives the yes answer 30%.

    30%

    0 · no100 · yes

    The machinesClaude 30% · DeepSeek 64% · GPT 32%

    Claude: 30%DeepSeek: 64%GPT: 32%

    Your number

  4. Oracle, purchases of property and equipment as a share of revenue, quarter ended November 30, 2026: whether it comes in at or above 128%.

    On the Docket Pendingth004-t04Pre-specified

    Question
    Did the purchases of property and equipment of Oracle for the quarter ended November 30, 2026 come to at or above 128.0% of its revenue?
    Source
    SEC EDGAR, XBRL company facts of Oracle (Forms 10-Q and 10-K)
    Rule
    Divide the purchases of property and equipment by the revenue that Oracle reports for the quarter ended November 30, 2026 in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if the ratio is at or above 128.0%; no if it is below 128.0%.
    If the source is late, revised or silent
    If Oracle has not filed by January 18, 2027, or a filing restates a period the rule reads, the test is inconclusive, and the Result says what was read and what was missing.
    Base rate
    Oracle's purchases of property and equipment as a share of revenue was 147.3% for the quarter ended August 31, 2026, by its filing on SEC EDGAR.
    Outside view
    A year earlier, for the quarter ended August 31, 2025, the same line read 57.0%, so the rungs sit on the last print.
    What would change the number
    A change in product mix, a one-time charge, or a reclassified line would move the reading.

    The Call, before the dateExpects yes. Gives the yes answer 68%.

    68%

    0 · no100 · yes

    The machinesClaude 28% · DeepSeek 60% · GPT 2%

    Claude: 28%DeepSeek: 60%GPT: 2%
    The ladder 3 rungs on one reading, Dec 14, 2026: each asks whether it lands above its threshold, so the Call’s numbers down the rungs sketch where it expects the reading to land.

    1. Above 128%Call 68% · Claude 28% · DeepSeek 60% · GPT 2%

      68%

    2. Above 143%Call 50% · Claude 18% · DeepSeek 20% · GPT 1%

      50%

    3. Above 160%Call 32% · Claude 11% · DeepSeek 6% · GPT 1%

      32%

    0 · no50100 · yes

    Your number

  5. Oracle, purchases of property and equipment as a share of revenue, quarter ended November 30, 2026: whether it comes in at or above 143%.

    On the Docket Pendingth004-t05Pre-specified

    Question
    Did the purchases of property and equipment of Oracle for the quarter ended November 30, 2026 come to at or above 143.0% of its revenue?
    Source
    SEC EDGAR, XBRL company facts of Oracle (Forms 10-Q and 10-K)
    Rule
    Divide the purchases of property and equipment by the revenue that Oracle reports for the quarter ended November 30, 2026 in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if the ratio is at or above 143.0%; no if it is below 143.0%.
    If the source is late, revised or silent
    If Oracle has not filed by January 18, 2027, or a filing restates a period the rule reads, the test is inconclusive, and the Result says what was read and what was missing.
    Base rate
    Oracle's purchases of property and equipment as a share of revenue was 147.3% for the quarter ended August 31, 2026, by its filing on SEC EDGAR.
    Outside view
    A year earlier, for the quarter ended August 31, 2025, the same line read 57.0%, so the rungs sit on the last print.
    What would change the number
    A change in product mix, a one-time charge, or a reclassified line would move the reading.

    The Call, before the dateEither answer fits the Call. Gives the yes answer 50%.

    50%

    0 · no100 · yes

    The machinesClaude 18% · DeepSeek 20% · GPT 1%

    Claude: 18%DeepSeek: 20%GPT: 1%

    Your number

  6. Oracle, purchases of property and equipment as a share of revenue, quarter ended November 30, 2026: whether it comes in at or above 160%.

    On the Docket Pendingth004-t06Pre-specified

    Question
    Did the purchases of property and equipment of Oracle for the quarter ended November 30, 2026 come to at or above 160.0% of its revenue?
    Source
    SEC EDGAR, XBRL company facts of Oracle (Forms 10-Q and 10-K)
    Rule
    Divide the purchases of property and equipment by the revenue that Oracle reports for the quarter ended November 30, 2026 in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if the ratio is at or above 160.0%; no if it is below 160.0%.
    If the source is late, revised or silent
    If Oracle has not filed by January 18, 2027, or a filing restates a period the rule reads, the test is inconclusive, and the Result says what was read and what was missing.
    Base rate
    Oracle's purchases of property and equipment as a share of revenue was 147.3% for the quarter ended August 31, 2026, by its filing on SEC EDGAR.
    Outside view
    A year earlier, for the quarter ended August 31, 2025, the same line read 57.0%, so the rungs sit on the last print.
    What would change the number
    A change in product mix, a one-time charge, or a reclassified line would move the reading.

    The Call, before the dateExpects no. Gives the yes answer 32%.

    32%

    0 · no100 · yes

    The machinesClaude 11% · DeepSeek 6% · GPT 1%

    Claude: 11%DeepSeek: 6%GPT: 1%

    Your number

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.