{
  "docId": "019dd923-622c-750b-8b98-491cf8c853e8",
  "docSlug": "3b36a8e60d92",
  "documentTitle": "Pure Storage, Inc. (PSTG)",
  "authorId": "53_Kerrisdale_Capital",
  "authorName": "Kerrisdale Capital",
  "documentKindSlug": "research-note",
  "documentKindLabel": "Research note",
  "sourceTypeSlug": "short_seller",
  "sourceTypeLabel": "Short seller",
  "presentationDate": "2025-09-01 00:00:00",
  "orientation": "portrait",
  "aspectRatio": 0.77272725,
  "pageNumber": 19,
  "pageCount": 35,
  "prevPage": 18,
  "nextPage": 20,
  "slideType": "preempt_rebuttal",
  "function": "preempt_rebuttal",
  "density": "dense",
  "nDataPoints": 3,
  "notes": "Contains multiple expert testimonials and references to specific industry studies (Microsoft, Alibaba, Meta).",
  "elementsJson": [
    "paragraph",
    "quote_block"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-622c-750b-8b98-491cf8c853e8/19",
  "deckHref": "/decks/019dd923-622c-750b-8b98-491cf8c853e8",
  "deckJsonHref": "/decks/019dd923-622c-750b-8b98-491cf8c853e8.json",
  "deckAnchorHref": "/decks/019dd923-622c-750b-8b98-491cf8c853e8#slide-19",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "The GPU utilization argument understandably resonates with GPU-obsessed investors, but the picture painted by the pro-flash camp is distorted.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-fb90-708a-8691-3c92ab9bfd8b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.8,
        "x": 0.1,
        "y": 0.14
      },
      "kind": "list",
      "text": "1) more workloads are migrating to the hyperscalers... 2) they still represent a small minority of total workloads... 3) while flash may be required to maximize GPU utilization... 4) the true arbiter of GPU performance is not the media holding the data",
      "attrs": null,
      "subkind": "numbered",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "07c7e078-0671-4466-9edc-e68cd645ae4a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.8,
        "x": 0.1,
        "y": 0.09
      },
      "kind": "paragraph",
      "text": "As a corollary to the access density argument, there is a tendency to frame the flash versus HDD debate purely through the lens of AI or, even more narrowly, LLM training. We believe this is a mistake.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "22ff52c1-e431-451e-9279-0cc491cfcfc7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.8,
        "x": 0.1,
        "y": 0.71
      },
      "kind": "paragraph",
      "text": "Some investors will point to xAI's Colossus as indicative of the predominance of AI workloads and the need for massive amounts of flash. Interestingly, it has been disclosed that DDN is supplying storage for Colossus...",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "76807726-0f3d-4dc9-a3c9-8f7a3d4d9d1b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.7,
        "x": 0.15,
        "y": 0.54
      },
      "kind": "quote",
      "text": "Spinning hard drives are very efficient way to do this and they provide adequate bandwidth when you gang them together. Over time in training environments, it went from all flash to mostly hard drives with some flash. That trend seems to be continuing.",
      "attrs": null,
      "subkind": "testimonial",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "408ba644-5a09-4ca1-abc3-71b4bba8665e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.7,
        "x": 0.15,
        "y": 0.64
      },
      "kind": "quote",
      "text": "Inferencing does require low latency storage, but it's not a done deal that that storage will have to sit on SSDs.",
      "attrs": null,
      "subkind": "testimonial",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "41f7be37-f4de-4100-9d26-d61dccbb6461",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "quote",
      "text": "\"Spinning hard drives are very efficient way to do this and they provide adequate bandwidth when you gang them together. Over time in training environments, it went from all flash to mostly hard drives with some flash. That trend seems to be continuing.\" — Director of Hardware Engineering at Meta (Tegus); \"Inferencing does require low latency storage, but it's not a done deal that that storage will have to sit on SSDs.\" — Data storage product manager at leading hyperscaler; \"They are very much putting hard drives in the network attached dedicated storage servers where they need lots and lots of cheap capacity...there's a reasonable case to\" — Unattributed source.",
      "attrs": null,
      "subkind": null,
      "toolName": "Authority citation",
      "toolSlug": "authority-citation",
      "confidence": null,
      "componentId": "019dd952-fb90-708a-8691-40224e673c25",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "frameworks": [
    {
      "name": "preempt_rebuttal",
      "slug": null,
      "matchId": "da456d58-7b78-482c-8bb6-1080479bf92d",
      "evidence": "The slide systematically addresses and refutes four common arguments for all-flash storage in AI environments.",
      "confidence": 0.9
    }
  ],
  "arcBeats": [
    {
      "to": 23,
      "from": 11,
      "beatId": "043f4323-8300-47cd-bd67-d0d7b1e3780a",
      "arcName": "Problem-Agitate-Solution",
      "arcSlug": "problem-agitate-solution",
      "beatName": "Preempt Rebuttal",
      "beatSlug": null,
      "evidence": "The document preempts potential rebuttals to its arguments",
      "position": 2,
      "confidence": 0.8,
      "parentBeatName": null,
      "parentBeatSlug": null
    }
  ],
  "loops": [
    {
      "to": 32,
      "from": 5,
      "name": "Cost Of Inaction",
      "slug": "27-cost-of-inaction",
      "bestFor": "Urgent budget requests, compliance, risk mitigation",
      "matchId": "f175e9df-40bd-4f68-b964-eed15e12ff5a",
      "evidence": "The document highlights the risks of investing in Pure Storage",
      "position": 0,
      "objective": "To highlight the risks of investing in Pure Storage",
      "structure": "The Status Quo -> The Hidden Costs Accumulating -> The Future State of Inaction -> The Tipping Point",
      "confidence": 0.7,
      "description": "Quantify what happens if the audience does nothing"
    },
    {
      "to": 23,
      "from": 10,
      "name": "Quick Win Big Bet",
      "slug": "47-quick-win-big-bet",
      "bestFor": "Transformation planning, 100-day plans, resource allocation",
      "matchId": "7a3fb01b-6d25-4fa6-b338-aeb407c58cc3",
      "evidence": "The document presents an alternative view to Pure Storage's TCO analysis",
      "position": 1,
      "objective": "To present an alternative view to Pure Storage's TCO analysis",
      "structure": "The Full List -> Quick Wins (Low effort, High impact) -> Big Bets (High effort, High impact) -> Sequenced Roadmap",
      "confidence": 0.6,
      "description": "Separate initiatives into immediate wins and longer-term strategic bets"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}