From c5a0850ad44c230abbf4ec9eefe9782707dee0c7 Mon Sep 17 00:00:00 2001 From: Billy1900 Date: Thu, 16 Jul 2026 08:32:47 -0400 Subject: [PATCH] feat: add llmquant-ai-supply-chain skill Adds a new category covering the AI industry supply chain from upstream semiconductor materials/equipment through chip design, memory, advanced packaging, networking/interconnect, hyperscaler capex, model/application adoption, and a narrative-vs-earnings quadrant screen. --- README.md | 1 + skills/llmquant-ai-supply-chain/SKILL.md | 56 ++++++++++++++ .../llmquant-ai-supply-chain/assets/.gitkeep | 0 .../llmquant-ai-supply-chain/scripts/.gitkeep | 0 .../workflows/advanced-packaging-tracker.md | 66 +++++++++++++++++ .../workflows/ai-model-application-monitor.md | 67 +++++++++++++++++ .../workflows/ai-network-interconnect-map.md | 66 +++++++++++++++++ .../ai-supply-chain-quadrant-screen.md | 73 +++++++++++++++++++ .../workflows/chip-design-memory-map.md | 67 +++++++++++++++++ .../hyperscaler-capex-datacenter-tracker.md | 67 +++++++++++++++++ .../workflows/semiconductor-upstream-map.md | 66 +++++++++++++++++ 11 files changed, 529 insertions(+) create mode 100644 skills/llmquant-ai-supply-chain/SKILL.md create mode 100644 skills/llmquant-ai-supply-chain/assets/.gitkeep create mode 100644 skills/llmquant-ai-supply-chain/scripts/.gitkeep create mode 100644 skills/llmquant-ai-supply-chain/workflows/advanced-packaging-tracker.md create mode 100644 skills/llmquant-ai-supply-chain/workflows/ai-model-application-monitor.md create mode 100644 skills/llmquant-ai-supply-chain/workflows/ai-network-interconnect-map.md create mode 100644 skills/llmquant-ai-supply-chain/workflows/ai-supply-chain-quadrant-screen.md create mode 100644 skills/llmquant-ai-supply-chain/workflows/chip-design-memory-map.md create mode 100644 skills/llmquant-ai-supply-chain/workflows/hyperscaler-capex-datacenter-tracker.md create mode 100644 skills/llmquant-ai-supply-chain/workflows/semiconductor-upstream-map.md diff --git a/README.md b/README.md index cbdf7cf..9166105 100644 --- a/README.md +++ b/README.md @@ -94,6 +94,7 @@ README.zh-CN.md | [`llmquant-strategies`](skills/llmquant-strategies) | Hedge-fund and PM strategy playbooks. | Equity long/short, long-biased, event-driven, macro, quant, multi-strategy | | [`llmquant-market-intelligence`](skills/llmquant-market-intelligence) | Reusable market utilities and signal views. | Macro view, market sentiment, event probability signals | | [`llmquant-investor-lenses`](skills/llmquant-investor-lenses) | Investor-style reasoning overlays using LLMQuant Data evidence. | Buffett, Graham, Munger, Lynch, Fisher, Burry, Ackman, Damodaran, and more | +| [`llmquant-ai-supply-chain`](skills/llmquant-ai-supply-chain) | AI industry supply-chain research from upstream semiconductors to downstream applications. | Semiconductor upstream map, chip design/memory map, advanced packaging tracker, network/interconnect map, hyperscaler capex tracker, model/application monitor, narrative-vs-earnings quadrant screen | ## Install diff --git a/skills/llmquant-ai-supply-chain/SKILL.md b/skills/llmquant-ai-supply-chain/SKILL.md new file mode 100644 index 0000000..9b3d48b --- /dev/null +++ b/skills/llmquant-ai-supply-chain/SKILL.md @@ -0,0 +1,56 @@ +--- +name: llmquant-ai-supply-chain +description: Router skill for LLMQuant AI supply-chain workflows. Use when the user needs semiconductor upstream, chip design, memory, advanced packaging, networking/interconnect, hyperscaler capex, AI model/application monitoring, or a narrative-vs-earnings quadrant view of AI-linked companies. +input_data_source: LLMQuant Data +category: ai-supply-chain +--- + +# LLMQuant AI Supply Chain + +This category routes AI-industry research workflows that treat the AI buildout as a structural supply chain: upstream materials and equipment, chip design, memory, advanced packaging, networking, hyperscaler data centers, and downstream model/application demand. It is distinct from `llmquant-equities` (single-name memos) and `llmquant-equity-derivatives` (options/hybrids) — this category stitches names together across tiers and tracks whether narrative, capex, and earnings evidence agree. + +## Routing Rules + +1. Identify the supply-chain tier, company set, geography/export-control exposure, horizon, and requested deliverable. +2. Select the closest workflow below. +3. Open only that workflow and any referenced local resources. +4. Use LLMQuant Data for filings, price history, revenue/margin trends, capex disclosures, and industry context. +5. Report filing periods, price dates, guidance-vs-actual deltas, stale notices, and missing inputs. + +## Workflow Index + +| User intent | Workflow | +|---|---| +| Map upstream materials and semiconductor equipment exposure (wafers, photoresist, gases, CMP, litho, etch, deposition, test). | [`workflows/semiconductor-upstream-map.md`](workflows/semiconductor-upstream-map.md) | +| Map chip design and memory exposure (CPU/GPU/AI ASIC/FPGA plus HBM/DRAM/NAND) and HBM-attach dynamics. | [`workflows/chip-design-memory-map.md`](workflows/chip-design-memory-map.md) | +| Track advanced packaging capacity and bottlenecks (CoWoS, SoIC, 2.5D/3D, chiplet, OSAT). | [`workflows/advanced-packaging-tracker.md`](workflows/advanced-packaging-tracker.md) | +| Map AI networking and interconnect exposure (CPO, optical modules, switching ASICs, SerDes, InfiniBand/Ethernet). | [`workflows/ai-network-interconnect-map.md`](workflows/ai-network-interconnect-map.md) | +| Track hyperscaler capex guidance vs. actuals, data-center buildout, power, and cooling constraints. | [`workflows/hyperscaler-capex-datacenter-tracker.md`](workflows/hyperscaler-capex-datacenter-tracker.md) | +| Monitor large-model evolution and downstream/enterprise AI application adoption as a recurring watch. | [`workflows/ai-model-application-monitor.md`](workflows/ai-model-application-monitor.md) | +| Classify AI supply-chain companies into a narrative-vs-earnings quadrant (core main line, awaiting verification, potential catch-up, risk zone). | [`workflows/ai-supply-chain-quadrant-screen.md`](workflows/ai-supply-chain-quadrant-screen.md) | + +## LLMQuant Data Contract + +Prefer LLMQuant Data when available. The workflows may need these data capabilities: +- Retrieve SEC filings (10-K/10-Q/8-K) and sections covering capex guidance, segment revenue, and risk factors. +- Retrieve equity price history, market cap, and realized volatility for supply-chain names. +- Retrieve revenue growth, gross margin, and valuation multiples (P/E, P/S) at company and peer-group level. +- Retrieve macro, export-control, and sector news context when it directly affects a supply-chain node. + +Future or optional data capabilities: +- Structured supply-chain node/tier tagging per company (materials, equipment, wafer fab, design, memory, packaging, network, servers, downstream). +- Consensus capex and unit-shipment estimates for cross-company comparison. + +Fallback: +- If filing, price, or fundamental data is unavailable for a name, name the missing input and continue only with retrieved evidence. +- Do not infer capacity, yields, order backlogs, or capex figures from memory; treat unresolvable claims as unverified. +- Do not present narrative or sentiment framing as if it were reported financial data. + +## Output Requirements + +Every workflow response should include: + +1. Answer or recommendation +2. Tier/node evidence table with company-level facts and dates +3. Risks and caveats (cyclicality, concentration, export-control, single-customer exposure) +4. Data used, including filing periods, price dates, and coverage notices diff --git a/skills/llmquant-ai-supply-chain/assets/.gitkeep b/skills/llmquant-ai-supply-chain/assets/.gitkeep new file mode 100644 index 0000000..e69de29 diff --git a/skills/llmquant-ai-supply-chain/scripts/.gitkeep b/skills/llmquant-ai-supply-chain/scripts/.gitkeep new file mode 100644 index 0000000..e69de29 diff --git a/skills/llmquant-ai-supply-chain/workflows/advanced-packaging-tracker.md b/skills/llmquant-ai-supply-chain/workflows/advanced-packaging-tracker.md new file mode 100644 index 0000000..e43a23b --- /dev/null +++ b/skills/llmquant-ai-supply-chain/workflows/advanced-packaging-tracker.md @@ -0,0 +1,66 @@ +--- +name: Advanced Packaging Tracker +description: Track advanced packaging capacity and bottlenecks across CoWoS, SoIC, 2.5D/3D packaging, chiplet integration, and OSAT providers using LLMQuant Data. +input_data_source: LLMQuant Data +pack: workflows +--- + +# Advanced Packaging Tracker + +## Purpose + +Advanced packaging (CoWoS, SoIC, 2.5D/3D integration, chiplet, OSAT) is the most frequently cited bottleneck between chip design and finished AI accelerators. This workflow tracks capacity, expansion disclosures, and which companies are exposed to that bottleneck easing or persisting. + +--- + +## Input Data Source + +Use **LLMQuant Data** for filings, capex disclosures, and price history. State which capabilities were used and cite returned dates or periods. + +--- + +## LLMQuant Data Contract + +Required data capabilities: +- SEC filing discovery and section retrieval (capex guidance, capacity-expansion commentary) for named packaging/OSAT companies. +- Equity price history for named companies. + +Optional data capabilities: +- Industry/trade-press context on packaging capacity additions and yield commentary. +- Consensus capex estimates for packaging-exposed names. + +Freshness: +- Use the latest available 10-K/10-Q/8-K unless the user specifies a period. +- State filing dates, periods of report, and price date ranges. + +Fallback: +- If capacity or yield figures are not disclosed in retrieved filings, name the gap and state that only qualitative or price-based evidence is available. +- Do not estimate CoWoS/SoIC capacity in units without a sourced figure. + +--- + +## Workflow + +1. Confirm scope: a specific packaging technology (CoWoS, SoIC, chiplet) or the full packaging/OSAT set. +2. Retrieve filings and price history for named companies. +3. Extract capex, capacity-expansion, and yield commentary as reported; separate from narrative or rumor. +4. Identify which downstream chip-design names are most exposed to packaging capacity as a gating factor. +5. Note geographic concentration (packaging capacity is concentrated in a small number of regions). + +--- + +## Output Format + +1. **Answer**: current bottleneck read (easing, stable, tightening) with evidence basis. +2. **Capacity/Capex Evidence Table** +3. **Downstream Exposure Map** (which design/hyperscaler names are gated by this bottleneck) +4. **Risks / Caveats** +5. **Data Used** + +--- + +## Guardrails + +- Do not present capacity-expansion announcements as completed, qualified capacity until filings confirm ramp. +- Do not state precise yield percentages without a sourced figure. +- Do not make personalized financial advice claims. diff --git a/skills/llmquant-ai-supply-chain/workflows/ai-model-application-monitor.md b/skills/llmquant-ai-supply-chain/workflows/ai-model-application-monitor.md new file mode 100644 index 0000000..27fdfbe --- /dev/null +++ b/skills/llmquant-ai-supply-chain/workflows/ai-model-application-monitor.md @@ -0,0 +1,67 @@ +--- +name: AI Model and Application Monitor +description: Monitor large-model evolution and downstream enterprise/consumer AI application adoption as a recurring watch, and connect adoption signals to which sectors and names see earnings impact first, using LLMQuant Data. +input_data_source: LLMQuant Data +pack: workflows +--- + +# AI Model and Application Monitor + +## Purpose + +Track two linked, fast-moving layers as a recurring watch rather than a one-off memo: (1) large-model evolution — release cadence, open- vs. closed-model economics, inference cost trends — and (2) downstream application adoption across enterprise and consumer sectors. The goal is to move past news aggregation into which non-tech sectors and names see earnings impact first from AI adoption. + +--- + +## Input Data Source + +Use **LLMQuant Data** for filings, segment commentary, and price history. State which capabilities were used and cite returned dates or periods. + +--- + +## LLMQuant Data Contract + +Required data capabilities: +- SEC filing discovery and section retrieval (AI-related segment commentary, product disclosures) for named model labs, platform companies, and adopting enterprises. +- Equity price history for named companies. + +Optional data capabilities: +- Wiki or paper research search to define model architectures or benchmark terminology. +- Industry/trade-press context on model releases and enterprise adoption case studies. + +Freshness: +- Use the latest available filings and earnings-call disclosures unless the user specifies a period. +- State filing/report dates and the observation window for any adoption claim. + +Fallback: +- If adoption metrics are not disclosed by name, state that only qualitative or anecdotal evidence is available and flag it as such. +- Do not present model benchmark scores or inference-cost figures from memory as current; require a retrieved or user-provided source. + +--- + +## Workflow + +1. Confirm scope: model-layer focus, application/adoption focus, or both; and which sectors or names to check for earnings impact. +2. Retrieve filings and disclosures for named model labs, platform providers, and adopting companies. +3. Separate model-layer evidence (release cadence, pricing changes, open vs. closed positioning) from application-layer evidence (disclosed adoption, revenue attribution, headcount or cost commentary tied to AI tools). +4. Map adoption evidence to the sector or name most likely to show earnings impact first, and state the mechanism (cost reduction, revenue line, productivity claim). +5. Flag where adoption claims are promotional/anecdotal versus filing-disclosed and quantified. + +--- + +## Output Format + +1. **Answer**: current state of model-layer competition and where adoption evidence is strongest. +2. **Model-Layer Evidence Table** +3. **Application-Layer Evidence Table** +4. **Sector/Earnings-Impact Map** +5. **Risks / Caveats** +6. **Data Used** + +--- + +## Guardrails + +- Do not present vendor marketing claims about adoption or ROI as verified filing disclosure. +- Do not predict which model will "win" without evidence; describe competitive dynamics, not forecasts. +- Do not make personalized financial advice claims. diff --git a/skills/llmquant-ai-supply-chain/workflows/ai-network-interconnect-map.md b/skills/llmquant-ai-supply-chain/workflows/ai-network-interconnect-map.md new file mode 100644 index 0000000..0361bde --- /dev/null +++ b/skills/llmquant-ai-supply-chain/workflows/ai-network-interconnect-map.md @@ -0,0 +1,66 @@ +--- +name: AI Network and Interconnect Map +description: Map AI networking and interconnect exposure across co-packaged optics, optical modules, switching ASICs, SerDes, and InfiniBand/Ethernet fabric using LLMQuant Data. +input_data_source: LLMQuant Data +pack: workflows +--- + +# AI Network and Interconnect Map + +## Purpose + +Map the networking layer that connects AI accelerators inside and across data centers: co-packaged optics (CPO), optical modules/transceivers, switching ASICs, SerDes IP, and InfiniBand/Ethernet fabric providers. This layer scales with GPU cluster size and is a distinct cycle from chip design or packaging. + +--- + +## Input Data Source + +Use **LLMQuant Data** for filings, segment revenue, and price history. State which capabilities were used and cite returned dates or periods. + +--- + +## LLMQuant Data Contract + +Required data capabilities: +- SEC filing discovery and section retrieval (segment revenue, product-mix, capacity commentary) for named networking/optics companies. +- Equity price history and realized volatility for named companies. + +Optional data capabilities: +- Consensus revenue and unit-shipment estimates for optical modules and switching silicon. +- Industry context on transceiver speed transitions (e.g. 800G to 1.6T) and CPO adoption timing. + +Freshness: +- Use the latest available 10-K/10-Q unless the user specifies a period. +- State filing dates, periods of report, and price date ranges. + +Fallback: +- If shipment or speed-transition data is unavailable in retrieved filings, name the gap and rely on reported segment revenue trends instead. +- Do not estimate module ASPs, port counts, or transition timing from memory. + +--- + +## Workflow + +1. Confirm scope: optics/transceivers, switching ASICs, SerDes/fabric, or the full networking layer. +2. Retrieve filings and price history for named companies. +3. Extract segment revenue, product-mix, and speed-transition commentary as reported. +4. Note which companies are exposed to InfiniBand vs. Ethernet fabric choices, since customer architecture decisions shift demand between them. +5. Flag customer concentration (networking suppliers are often concentrated on a small number of hyperscaler or accelerator customers). + +--- + +## Output Format + +1. **Answer**: current read on networking-layer demand and technology-transition positioning. +2. **Networking Evidence Table** +3. **Fabric/Technology Transition Notes** (InfiniBand vs. Ethernet, CPO adoption stage) +4. **Concentration Risks** +5. **Data Used** + +--- + +## Guardrails + +- Do not present a technology-transition timeline as certain; describe it as company-disclosed guidance or industry expectation. +- Do not conflate design-win announcements with recognized revenue. +- Do not make personalized financial advice claims. diff --git a/skills/llmquant-ai-supply-chain/workflows/ai-supply-chain-quadrant-screen.md b/skills/llmquant-ai-supply-chain/workflows/ai-supply-chain-quadrant-screen.md new file mode 100644 index 0000000..a9349b0 --- /dev/null +++ b/skills/llmquant-ai-supply-chain/workflows/ai-supply-chain-quadrant-screen.md @@ -0,0 +1,73 @@ +--- +name: AI Supply Chain Quadrant Screen +description: Classify AI supply-chain companies into a narrative-vs-earnings quadrant (core main line, awaiting verification, potential catch-up, risk zone) using price performance, revenue growth, margin trend, and valuation multiples from LLMQuant Data. +input_data_source: LLMQuant Data +pack: workflows +--- + +# AI Supply Chain Quadrant Screen + +## Purpose + +Classify a set of AI supply-chain companies (spanning upstream materials, equipment, design, memory, packaging, networking, servers, and downstream demand) into four quadrants by crossing narrative strength against earnings validation: + +- **Core Main Line**: strong narrative + strong earnings evidence. +- **Awaiting Verification**: strong narrative, earnings evidence not yet conclusive. +- **Potential Catch-Up**: strong earnings evidence, narrative/valuation lagging. +- **Risk Zone**: high valuation, weak or decelerating growth evidence. + +This is the cross-tier synthesis workflow — use the tier-specific workflows first to gather node-level evidence, then use this workflow to classify. + +--- + +## Input Data Source + +Use **LLMQuant Data** for filings, price history, and fundamentals. State which capabilities were used and cite returned dates or periods. + +--- + +## LLMQuant Data Contract + +Required data capabilities: +- Equity price history (1-month and 6-month performance) for named companies. +- Revenue growth rate, gross margin, and valuation multiples (P/E, P/S) from filings or fundamental data. + +Optional data capabilities: +- Peer-group or supply-chain-node median metrics for relative comparison. +- Consensus estimate revisions as a narrative-strength proxy. + +Freshness: +- Use the latest available filing period and most recent price data unless the user specifies otherwise. +- State filing dates, periods of report, and price date ranges for every company classified. + +Fallback: +- If a required metric (growth, margin, or multiple) is unavailable for a company, state that the classification is incomplete for that name and classify only on available metrics, flagged as partial. +- Do not assign a quadrant based on narrative alone without at least one quantitative input. + +--- + +## Workflow + +1. Confirm the company set and supply-chain tier(s) in scope; if the user gives a theme instead of tickers, ask for or infer a representative list from prior workflow output. +2. Retrieve price performance, revenue growth, gross margin, and valuation multiples for each company. +3. Score narrative strength using price momentum and estimate-revision direction (if available) as a proxy; score earnings validation using revenue growth and margin trend. +4. Place each company into one of the four quadrants and state the deciding evidence. +5. Aggregate a node-level read (e.g. median quadrant placement per tier) if multiple companies per tier are scored. + +--- + +## Output Format + +1. **Answer**: quadrant summary across the company set. +2. **Quadrant Table**: company, tier, price performance, growth, margin, valuation, assigned quadrant. +3. **Node-Level Aggregation** (if applicable). +4. **Risks / Caveats** (data gaps, single-metric classifications, cyclicality). +5. **Data Used**. + +--- + +## Guardrails + +- Do not treat quadrant placement as a price target or trade recommendation. +- Do not classify a company using narrative alone; require at least one quantitative input per axis. +- Do not make personalized financial advice claims. diff --git a/skills/llmquant-ai-supply-chain/workflows/chip-design-memory-map.md b/skills/llmquant-ai-supply-chain/workflows/chip-design-memory-map.md new file mode 100644 index 0000000..46cdf7d --- /dev/null +++ b/skills/llmquant-ai-supply-chain/workflows/chip-design-memory-map.md @@ -0,0 +1,67 @@ +--- +name: Chip Design and Memory Map +description: Map chip design exposure (CPU/GPU/AI ASIC/FPGA/DPU) alongside memory exposure (HBM/DRAM/NAND) and the HBM-attach dynamics linking them, using LLMQuant Data. +input_data_source: LLMQuant Data +pack: workflows +--- + +# Chip Design and Memory Map + +## Purpose + +Map the chip-design layer (CPU, GPU, AI ASIC, FPGA, DPU/NIC, power-management chips) together with the memory layer (HBM, DRAM, NAND, enterprise SSD), since AI accelerator demand and HBM attach rates are now tightly coupled. Treat these as one workflow because a design-side read (accelerator mix shift) is usually incomplete without the memory-side read (HBM supply, DRAM pricing). + +--- + +## Input Data Source + +Use **LLMQuant Data** for filings, segment revenue, price history, and margin trends. State which capabilities were used and cite returned dates or periods. + +--- + +## LLMQuant Data Contract + +Required data capabilities: +- SEC filing discovery and section retrieval (segment revenue, gross margin, product-mix commentary) for named design and memory companies. +- Equity price history and realized volatility for named companies. + +Optional data capabilities: +- Memory spot/contract pricing context (HBM, DRAM, NAND) when available. +- Consensus revenue and gross-margin estimates for cross-company comparison. + +Freshness: +- Use the latest available 10-K/10-Q unless the user specifies a period. +- State filing dates, periods of report, and price date ranges. + +Fallback: +- If HBM/DRAM/NAND pricing data is unavailable, name the gap and rely on filing-disclosed gross margin trends instead. +- Do not estimate chip yields, HBM attach rates, or wafer allocation from memory. + +--- + +## Workflow + +1. Confirm scope: design-only, memory-only, or the combined design-memory read. +2. Retrieve filings and price history for named companies across both layers. +3. Separate design evidence (unit mix, ASIC vs. merchant GPU share, architecture roadmap disclosures) from memory evidence (HBM/DRAM/NAND revenue mix, gross margin trend). +4. Where both are in scope, explicitly connect them: rising AI-accelerator shipments implies HBM-attach demand; check whether memory-company filings corroborate that with revenue or margin evidence. +5. Flag where design-side narrative outruns memory-side confirmation, or vice versa. + +--- + +## Output Format + +1. **Answer**: design/memory read and whether the two layers corroborate each other. +2. **Design Evidence Table** +3. **Memory Evidence Table** +4. **Design-Memory Coupling Check** +5. **Risks / Caveats** +6. **Data Used** + +--- + +## Guardrails + +- Do not present product roadmap rumors as filed disclosure. +- Do not infer HBM attach rate as a precise number without a sourced figure; describe direction and qualitative confidence instead. +- Do not make personalized financial advice claims. diff --git a/skills/llmquant-ai-supply-chain/workflows/hyperscaler-capex-datacenter-tracker.md b/skills/llmquant-ai-supply-chain/workflows/hyperscaler-capex-datacenter-tracker.md new file mode 100644 index 0000000..7944fa3 --- /dev/null +++ b/skills/llmquant-ai-supply-chain/workflows/hyperscaler-capex-datacenter-tracker.md @@ -0,0 +1,67 @@ +--- +name: Hyperscaler Capex and Data Center Tracker +description: Track hyperscaler capex guidance versus actuals, data-center buildout pace, and power/cooling constraints as a leading indicator of AI compute demand using LLMQuant Data. +input_data_source: LLMQuant Data +pack: workflows +--- + +# Hyperscaler Capex and Data Center Tracker + +## Purpose + +Track capex guidance and actuals across major hyperscalers, data-center buildout pace, and power/liquid-cooling constraints. Hyperscaler capex is the top-of-funnel demand signal for the entire upstream AI supply chain, so guidance revisions here are a leading indicator for chip design, memory, packaging, and networking names. + +--- + +## Input Data Source + +Use **LLMQuant Data** for filings, capex disclosures, and price history. State which capabilities were used and cite returned dates or periods. + +--- + +## LLMQuant Data Contract + +Required data capabilities: +- SEC filing discovery and section retrieval (capex guidance, capex actuals, data-center/infrastructure commentary) for named hyperscalers. +- Equity price history for named companies. + +Optional data capabilities: +- Power-grid, utility, or permitting news context relevant to data-center buildout. +- Consensus capex estimates for cross-quarter comparison. + +Freshness: +- Use the latest available 10-K/10-Q/8-K and most recent earnings-call disclosures unless the user specifies a period. +- State filing/report dates and the specific quarter being compared. + +Fallback: +- If a hyperscaler's capex breakdown (AI-specific vs. total) is not disclosed, name the gap and use total capex as the reported proxy. +- Do not estimate power availability or interconnection queue timelines from memory. + +--- + +## Workflow + +1. Confirm scope: which hyperscalers, and whether the ask is guidance-vs-actual, quarter-over-quarter trend, or power/cooling constraint focus. +2. Retrieve filings and price history for named companies. +3. Compare capex guidance given last quarter against actual reported capex this quarter; compute the delta and direction. +4. Extract any data-center buildout, power, or liquid-cooling commentary as reported. +5. Translate the capex trend into a top-of-funnel demand read for upstream names (chip design, memory, packaging, networking). + +--- + +## Output Format + +1. **Answer**: capex trend read (accelerating, decelerating, in line with guidance) and top-of-funnel implication. +2. **Capex Guidance-vs-Actual Table** +3. **Buildout / Power Constraint Notes** +4. **Upstream Demand Implication** +5. **Risks / Caveats** +6. **Data Used** + +--- + +## Guardrails + +- Do not present forward capex guidance as a guaranteed outcome; label it as management guidance subject to revision. +- Do not infer power-grid capacity or permitting timelines without a sourced disclosure. +- Do not make personalized financial advice claims. diff --git a/skills/llmquant-ai-supply-chain/workflows/semiconductor-upstream-map.md b/skills/llmquant-ai-supply-chain/workflows/semiconductor-upstream-map.md new file mode 100644 index 0000000..199fcd1 --- /dev/null +++ b/skills/llmquant-ai-supply-chain/workflows/semiconductor-upstream-map.md @@ -0,0 +1,66 @@ +--- +name: Semiconductor Upstream Map +description: Map upstream materials and semiconductor equipment exposure across wafers, photoresist, specialty gases, CMP consumables, and fab tool categories using LLMQuant Data. +input_data_source: LLMQuant Data +pack: workflows +--- + +# Semiconductor Upstream Map + +## Purpose + +Build an evidence-first map of the upstream semiconductor layer: materials (silicon wafers, photoresist, electronic specialty gases, CMP slurries/pads, wet chemicals, sputtering targets) and equipment (lithography, etch, deposition, ion implantation, cleaning, metrology/inspection, test). This layer sets the capacity ceiling for everything downstream. + +--- + +## Input Data Source + +Use **LLMQuant Data** for filings, price history, segment revenue, and industry context. State which capabilities were used and cite returned dates or periods. + +--- + +## LLMQuant Data Contract + +Required data capabilities: +- SEC filing discovery and section retrieval (segment revenue, capex, backlog commentary) for named companies. +- Equity price history for the named companies and a peer index. + +Optional data capabilities: +- Consensus revenue/backlog estimates for equipment and materials names. +- Export-control or trade-policy news context relevant to named companies or geographies. + +Freshness: +- Use the latest available 10-K/10-Q unless the user specifies a period. +- State filing dates, periods of report, and price date ranges. + +Fallback: +- If a company or sub-category is not covered by retrieved data, name the gap and continue only with retrieved evidence. +- Do not estimate fab utilization, tool lead times, or order backlogs from memory. + +--- + +## Workflow + +1. Confirm the sub-scope: materials, equipment, or both; and whether the user wants a specific company set or the full node map. +2. Retrieve filings and price history for the named companies. +3. Group evidence by node: wafers, photoresist, specialty gases, CMP/wet chemicals, targets (materials); litho, etch, deposition, ion implant, cleaning, metrology, test (equipment). +4. Note capacity, capex, and backlog commentary from filings; separate reported facts from narrative. +5. Flag single-customer or single-geography concentration risk where filings disclose it. + +--- + +## Output Format + +1. **Answer**: which nodes/companies are in scope and the key read. +2. **Node Evidence Table**: company, node, filing period, capex/backlog facts, price trend. +3. **Concentration / Cyclicality Risks** +4. **Risks / Caveats** +5. **Data Used** + +--- + +## Guardrails + +- Do not present tool lead-time or capacity-utilization claims as fact unless sourced from a retrieved filing or disclosure. +- Do not conflate equipment-order announcements with recognized revenue. +- Do not make personalized financial advice claims.