Thesis · Semiconductors & Compute Silicon

Will a panel-level or substrate-less advanced packaging line (CoPoS, CoWoP, or OSAT 310x310mm panel) reach announced mass production for an AI accelerator product by the end of 2027?

The CoWoS shortfall forces panelization and substrate-less relief paths into real mass production by end-2027, moving part of the AI package bottleneck out of TSMC's wafer-based lines.

6 dated tests · 0 resolved · next on

The tests More theses in Semiconductors & Compute Silicon

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

Advanced packaging capacity limits supply

A CoWoS supply gap above 20% today and above 30% by 2027, plus AI packages exceeding wafer-format reticle and area limits, makes panel formats and CoWoP economically necessary. OSATs and TSMC qualify panel/substrate-less lines, cutting unit-area cost 20-30% and widening effective packaging supply beyond CoWoS allocation.

The case for no

Panel-level lines stay off the mass-production roadmap

Suppliers cut or under-invested after the COVID cycle while AI demand accelerated past forecasts, so capacity remains scarce at critical points and headline capex is partly absorbed by price, duplicated sites and system coordination rather than new panel lines. Packaging allocation tightened early, with pools that xAI, Anthropic, OpenAI and Google all drew from since 2024, and a visible frontier win pressures competitors to spend on existing wafer-based packaging and infrastructure rather than unproven formats. This side would be confirmed if ASE, Powertech and TSMC all push panel-level mass production beyond 2028 in public statements.

2 voices

Advanced packaging capacity limits supply

  1. Daniel Romero

    CoWoS is sold out with NVIDIA locking over half of capacity through 2027 and Google cutting TPU targets. Panels cut unit-area cost 20-30% (CoPoS) and fit oversized packages wafers cannot hold. CoWoP (NVIDIA with SPIL/ASE) drops the substrate…

    Source

  2. Tscsw

    Modern AI accelerators exceed the physical size that EUV lithography can print in a single shot. The solution is not larger chips but better packaging through chiplets. Multiple smaller dies assembled on shared substrates enable higher yields…

    Source

3 voices

Panel-level lines stay off the mass-production roadmap

  1. Longriver

    Suppliers cut or under-invested after the COVID cycle while demand accelerated past forecasts, so capacity is scarce at critical points; headline capex is partly absorbed by price, duplicated sites and system coordination, and the firms that…

    Source

  2. Ruben Dominguez

    In early 2024, TSMC chip packaging capacity was at absolute tightest. That was when xAI began assembling what would become Colossus. Every unit locked up at that point came directly out of a pool that Anthropic, OpenAI and Google were also drawing…

    Source

  3. Citrini

    A visible frontier win pressures competitors to spend, and since they cannot cheaply change accelerator ecosystems the spending lands on Nvidia's next generation. The binding constraint therefore shows up in packaging and infrastructure supply…

    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. Applied Materials, revenue growth from a year earlier, quarter ended October 25, 2026: whether it comes in at or above 45%.

    On the Docket Pendingth012-t04Open question

    Specification pending

    The machinesClaude 6% · DeepSeek 3% · GPT 30%

    Claude: 6%DeepSeek: 3%GPT: 30%

    Your number

  2. Applied Materials, revenue growth from a year earlier, quarter ended October 25, 2026: whether it comes in at or above 52%.

    On the Docket Pendingth012-t05Open question

    Specification pending

    The machinesClaude 3% · DeepSeek 3% · GPT 17%

    Claude: 3%DeepSeek: 3%GPT: 17%

    Your number

  3. Applied Materials, revenue growth from a year earlier, quarter ended October 25, 2026: whether it comes in at or above 58%.

    On the Docket Pendingth012-t06Open question

    Specification pending

    The machinesClaude 2% · DeepSeek 3% · GPT 6%

    Claude: 2%DeepSeek: 3%GPT: 6%

    Your number

  4. NVIDIA, revenue growth from a year earlier, quarter ended October 25, 2026: whether it comes in at or above 85%.

    On the Docket Pendingth012-t01Pre-specified

    Question
    Did the revenue of Nvidia for the quarter ended October 25, 2026 grow at or above 85.0% from the quarter ended October 26, 2025?
    Source
    SEC EDGAR, XBRL company facts of Nvidia (Forms 10-Q and 10-K)
    Rule
    Compare the revenue that Nvidia reports for the quarter ended October 25, 2026 with the quarter ended October 26, 2025, in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if growth is at or above 85.0%; no if it is below 85.0%.
    If the source is late, revised or silent
    If Nvidia has not filed by December 11, 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
    NVIDIA's revenue growth from a year earlier was 105.9% for the quarter ended July 26, 2026, by its filing on SEC EDGAR.
    Outside view
    The year-earlier quarter is not on the same concept in the filings, so the rungs sit on the last print alone.
    What would change the number
    The results-calendar consensus for the quarter ended October 25, 2026 implies 90.9%, and the middle rung is centred there.

    The Call, before the dateExpects yes. Gives the yes answer 65%.

    65%

    0 · no100 · yes

    The machinesClaude 15% · DeepSeek 40%

    Claude: 15%DeepSeek: 40%
    The ladder 3 rungs on one reading, Nov 18, 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 85%Call 65% · Claude 15% · DeepSeek 40%

      65%

    2. Above 92%Call 48% · Claude 7% · DeepSeek 5% · GPT 22%

      48%

    3. Above 100%Call 30% · Claude 4% · DeepSeek 7% · GPT 25%

      30%

    0 · no50100 · yes

    Your number

  5. NVIDIA, revenue growth from a year earlier, quarter ended October 25, 2026: whether it comes in at or above 92%.

    On the Docket Pendingth012-t02Pre-specified

    Question
    Did the revenue of Nvidia for the quarter ended October 25, 2026 grow at or above 92.0% from the quarter ended October 26, 2025?
    Source
    SEC EDGAR, XBRL company facts of Nvidia (Forms 10-Q and 10-K)
    Rule
    Compare the revenue that Nvidia reports for the quarter ended October 25, 2026 with the quarter ended October 26, 2025, in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if growth is at or above 92.0%; no if it is below 92.0%.
    If the source is late, revised or silent
    If Nvidia has not filed by December 11, 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
    NVIDIA's revenue growth from a year earlier was 105.9% for the quarter ended July 26, 2026, by its filing on SEC EDGAR.
    Outside view
    The year-earlier quarter is not on the same concept in the filings, so the rungs sit on the last print alone.
    What would change the number
    The results-calendar consensus for the quarter ended October 25, 2026 implies 90.9%, and the middle rung is centred there.

    The Call, before the dateEither answer fits the Call. Gives the yes answer 48%.

    48%

    0 · no100 · yes

    The machinesClaude 7% · DeepSeek 5% · GPT 22%

    Claude: 7%DeepSeek: 5%GPT: 22%

    Your number

  6. NVIDIA, revenue growth from a year earlier, quarter ended October 25, 2026: whether it comes in at or above 100%.

    On the Docket Pendingth012-t03Pre-specified

    Question
    Did the revenue of Nvidia for the quarter ended October 25, 2026 grow at or above 100.0% from the quarter ended October 26, 2025?
    Source
    SEC EDGAR, XBRL company facts of Nvidia (Forms 10-Q and 10-K)
    Rule
    Compare the revenue that Nvidia reports for the quarter ended October 25, 2026 with the quarter ended October 26, 2025, in its 10-Q or 10-K filing on SEC EDGAR. The answer is yes if growth is at or above 100.0%; no if it is below 100.0%.
    If the source is late, revised or silent
    If Nvidia has not filed by December 11, 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
    NVIDIA's revenue growth from a year earlier was 105.9% for the quarter ended July 26, 2026, by its filing on SEC EDGAR.
    Outside view
    The year-earlier quarter is not on the same concept in the filings, so the rungs sit on the last print alone.
    What would change the number
    The results-calendar consensus for the quarter ended October 25, 2026 implies 90.9%, and the middle rung is centred there.

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

    30%

    0 · no100 · yes

    The machinesClaude 4% · DeepSeek 7% · GPT 25%

    Claude: 4%DeepSeek: 7%GPT: 25%

    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.