{
  "schemaVersion": 1,
  "id": "atlas-lesson-production-ramp-allocation",
  "title": "Production ramps, inventory and allocation",
  "category": "learning",
  "format": "Atlas lesson",
  "description": "Tracks time-dependent output and distinguishes qualified supply from a customer’s accessible share.",
  "canonicalUrl": "https://semiaifoundry.com/learn/atlas/production-ramp-allocation/",
  "publishedDate": null,
  "revisionDate": "2026-10-07",
  "author": null,
  "topic": "Data-center physical systems",
  "series": null,
  "license": null,
  "rightsUrl": "https://semiaifoundry.com/ai-atlas/licenses.html",
  "version": "1.3.0",
  "status": "Published educational lesson",
  "evidenceScope": "Allocation fractions and delays are synthetic. TSMC source material supports planning roles, not confidential queues, customer shares or current unit-per-month estimates.",
  "contentCoverage": "published-lesson",
  "artifactUrl": "https://semiaifoundry.com/ai-atlas/lesson-data/production-ramp-allocation-3d4ff12263731960.json",
  "interactiveUrl": "https://semiaifoundry.com/ai-atlas/#production-ramp-allocation",
  "related": [],
  "supersedes": [],
  "supersededBy": [],
  "contentHash": "12330e75e931a266c5a7fc9cef4ed121ccf2617eefe56764091563d94877847f",
  "contentLength": 5304,
  "jsonUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-production-ramp-allocation.json",
  "markdownUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-production-ramp-allocation.md",
  "sections": [
    {
      "id": "section-1",
      "title": "Production ramps, inventory and allocation",
      "start": 45,
      "end": 442
    },
    {
      "id": "section-2",
      "title": "The mechanism",
      "start": 442,
      "end": 1727
    },
    {
      "id": "section-3",
      "title": "Science and assumptions",
      "start": 1727,
      "end": 1755
    },
    {
      "id": "section-4",
      "title": "Inventory conservation",
      "start": 1755,
      "end": 2548
    },
    {
      "id": "section-5",
      "title": "Engineering example",
      "start": 2548,
      "end": 2802
    },
    {
      "id": "section-6",
      "title": "Distinctions and tradeoffs",
      "start": 2802,
      "end": 3237
    },
    {
      "id": "section-7",
      "title": "Specifications and interfaces",
      "start": 3237,
      "end": 3548
    },
    {
      "id": "section-8",
      "title": "Interfaces",
      "start": 3548,
      "end": 3761
    },
    {
      "id": "section-9",
      "title": "Check your understanding",
      "start": 3761,
      "end": 3790
    },
    {
      "id": "section-10",
      "title": "Why can a capacity increase fail to improve next week’s deliveries?",
      "start": 3790,
      "end": 4219
    },
    {
      "id": "section-11",
      "title": "Evidence and scope",
      "start": 4219,
      "end": 4575
    },
    {
      "id": "section-12",
      "title": "Sources",
      "start": 4575,
      "end": 4914
    },
    {
      "id": "section-13",
      "title": "Continue exploring",
      "start": 4914,
      "end": 4937
    },
    {
      "id": "section-14",
      "title": "Foundations",
      "start": 4937,
      "end": 5073
    },
    {
      "id": "section-15",
      "title": "Related lessons",
      "start": 5073,
      "end": 5304
    }
  ],
  "citations": [
    {
      "title": "Explore in Atlas",
      "url": "https://semiaifoundry.com/ai-atlas/#production-ramp-allocation"
    },
    {
      "title": "TSMC 2025 Annual Report",
      "url": "https://investor.tsmc.com/sites/ir/annual-report/2025/2025%20TSMC%20Annual%20Report.E.pdf"
    },
    {
      "title": "Published lesson data",
      "url": "https://semiaifoundry.com/ai-atlas/lesson-data/production-ramp-allocation-3d4ff12263731960.json"
    },
    {
      "title": "Atlas releases and licensing",
      "url": "https://semiaifoundry.com/ai-atlas/licenses.html"
    },
    {
      "title": "Compatible supply and converging bottlenecks",
      "url": "https://semiaifoundry.com/learn/atlas/compatible-supply-capacity/"
    },
    {
      "title": "Facility phases and geographical evidence",
      "url": "https://semiaifoundry.com/learn/atlas/capacity-phase-geography/"
    },
    {
      "title": "Cluster scheduler and resource allocation",
      "url": "https://semiaifoundry.com/learn/atlas/scheduler/"
    }
  ],
  "recordHash": "6397cb8cde5c244bf84cb01e027c22998d56d08e03a83f339a10e7e10467bd84",
  "contentFormat": "text/markdown",
  "content": "Data-center physical systems · Atlas v1.3.0\n\n# Production ramps, inventory and allocation\n\nTracks time-dependent output and distinguishes qualified supply from a customer’s accessible share.\n\n[Explore in Atlas](https://semiaifoundry.com/ai-atlas/#production-ramp-allocation)Manufacturing supply to commissioned hall; time-dependent model\n\nLearning objectiveModel ramps and allocation without treating future capacity as immediate inventory.\n\n## The mechanism\n\nA capacity announcement describes a plan or installed capability, while output ramps as equipment, process performance and qualification mature. Inventory can buffer delays, but repeated visits and queued work make factory cycle time longer than processing time. A finished product may then be reserved, contracted or allocated to a different customer. TSMC’s annual report describes customer-linked capacity planning without revealing private allocation shares. Physical output and access require separate accounting: completed equipment can exist without being allocated to the workload in question. Increasing a nominal monthly capacity input must not instantly create finished inventory before its lead time and qualification gates have elapsed.\n\n1.  Represent the announced or installed capability separately from actual starts and completed units.\n2.  Delay output by processing and waiting time, with qualification as its own release gate.\n3.  Update inventory as compatible inputs arrive and finished products consume them.\n4.  Allocate qualified output using explicit scenario rules rather than guessing confidential customer shares.\n5.  Compare the time-dependent delivered result with the intended deployment schedule and identify the current bottleneck.\n\n## Science and assumptions\n\n### Inventory conservation\n\n$$\nI(t+1)=I(t)+A(t)-C(t)\n$$\n\nNext inventory equals current inventory plus arrivals minus consumption during the same interval.\n\nA and C explicitly name the original arrivals and consumption functions; they are counts per interval, not unintegrated rates.\n\nSymbols and units\n\nI(t)\n\nInventory at the beginning of discrete interval t\n\nsame compatible item units\n\nI(t+1)\n\nInventory at the end of that interval\n\nsame compatible item units\n\nA(t)\n\nArrivals during interval t\n\nsame compatible item units\n\nC(t)\n\nConsumption during interval t\n\nsame compatible item units\n\nt\n\nDiscrete interval index\n\ninteger index\n\nEvery term counts the same compatible item over the same time step.\n\nAssumptions No hidden scrap or rework; those flows must be explicitly added when modeled.\n\n## Engineering example\n\nSynthetic three-week lead time means extra starts in week one cannot become completed supply until week four at the earliest. A customer allocated half that output receives only that half, even if total factory output is larger.\n\n## Distinctions and tradeoffs\n\nCapex is not production volume, and supplier capacity does not identify a customer’s allocated share.\n\nInventory improves resilience to some delays but consumes resources and can become mismatched when designs change; conservative ramps reduce overcommitment.\n\nHardware availability for a training program depends on arrival dates and allocation, not merely on a supplier’s eventual nameplate capacity.\n\n## Specifications and interfaces\n\nThree time concepts\n\nProcessing, waiting and qualification delay\n\nAll can affect first usable output without representing the same work.\n\nSupply versus access\n\nQualified production and allocated units\n\nPhysical existence does not establish availability to a particular buyer.\n\n### Interfaces\n\nCapacity plan and incoming inventory Input\n\nTime-varying starts, lead times and screening assumptions\n\nAllocated qualified supply Output\n\nFinished units released to a defined destination and date\n\n## Check your understanding\n\n### Why can a capacity increase fail to improve next week’s deliveries?\n\n1.  Lead time, qualification or another input can delay output\n2.  Capacity and inventory are always identical\n3.  All reservations disappear when a tool arrives\n\nAnswer and reasoning\n\nLead time, qualification or another input can delay output\n\nA change in possible starts takes time to propagate and may encounter another required input or release gate.\n\n## Evidence and scope\n\nAllocation fractions and delays are synthetic. TSMC source material supports planning roles, not confidential queues, customer shares or current unit-per-month estimates.\n\nTrace compatible component rates, a bounded qualification/ramp scenario and package backlog; this is not a detailed fab queue or operator-allocation simulator.\n\n## Sources\n\n-   [TSMC 2025 Annual Report](https://investor.tsmc.com/sites/ir/annual-report/2025/2025%20TSMC%20Annual%20Report.E.pdf)\n\n[Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/production-ramp-allocation-3d4ff12263731960.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html)\n\n## Continue exploring\n\n### Foundations\n\n-   [Compatible supply and converging bottlenecks](https://semiaifoundry.com/learn/atlas/compatible-supply-capacity/)\n\n### Related lessons\n\n-   [Facility phases and geographical evidence](https://semiaifoundry.com/learn/atlas/capacity-phase-geography/)\n-   [Cluster scheduler and resource allocation](https://semiaifoundry.com/learn/atlas/scheduler/)"
}