Jensen Huang Nvidia Catalyst

Jensen Huang / NVIDIA Catalyst Lens

Overview

Use this lens to turn public Jensen Huang and NVIDIA dynamics into calibrated short-horizon market context. Treat the output as decision support, not a trade instruction or long-term thesis.

Step 0 - Refresh First

Jensen signals come from many places, not one canonical feed.

1. Pull the latest yan-labs/jensen-huang-nvidia-catalyst repo. 2. Load references/source-map.md and check the relevant official NVIDIA source first: Newsroom, Blog, Investor Relations, events, YouTube/on-demand, or SEC. 3. Search recent web/news for the user's topic and for fresh broad terms: Jensen Huang NVIDIA interview, 黄仁勋 英伟达 最新, Jensen Huang export controls, NVIDIA Blackwell Rubin, NVIDIA sovereign AI, and related ticker/theme terms. 4. Dedupe any new item against data/sightings.json and references/track-record.md. 5. Never call an item "latest" unless the timestamp/source was observed in the current run.

Core Workflow

1. Load references/methodology.md to classify the signal tier and affected market basket. 2. Load references/source-map.md to choose source reliability and follow-up searches. 3. Run the supply-chain overlay for every market-relevant signal: upstream foundry/memory/packaging, midstream networking/servers, downstream cloud or sovereign-AI customers, plus power/cooling constraints. 4. Load references/track-record.md to weight the signal by observed base rate. 5. Output timestamp, source, link, signal type, supply-chain confirmation, affected assets, likely direction, reliability, duration, invalidation trigger, and data caveats. 6. Keep broker/account language separate: public news and public quotes are market context only, never uSMART account truth or fill confirmation.

Navigation

Risk Framing

• State that the analysis is decision support only and not financial advice.
• Separate "can move NVDA today" from "is a durable investment thesis."
• Label sources by reliability; do not treat rumors or unsourced social clips as

confirmed Jensen statements.

• Distinguish product-cycle demand, policy risk, customer capex, and supplier

capacity. They move different baskets.

• Do not analyze Jensen/NVIDIA in isolation when a supply-chain read is

possible; identify the bottleneck and who benefits or gets pressured.

• Never place, amend, cancel, or size trades from this skill.

Maintenance playbook

Purpose

Keep the Jensen Huang / NVIDIA catalyst lens current by tracking public dynamics, deduping sightings, classifying market-relevant signals, and scoring outcomes.

Scheduled run

1. Pull the skill repository first with fast-forward only. 2. Read source-map.md, methodology.md, and track-record.md. 3. Check official NVIDIA sources first: Newsroom, Blog, Investor Relations, Events/GTC/On-Demand, YouTube, and SEC EDGAR. 4. Run recent web/news searches from source-map.md, in both English and Chinese, for the last 24 hours and last 7 days. 5. For every market-relevant Jensen/NVIDIA item, run the supply-chain search pack in source-map.md and record at least one upstream, midstream, or downstream confirmation attempt in the notes, even when the result is "not confirmed." 6. Add new public items to data/sightings.json and data/sightings.csv, deduping by canonical URL first, then by normalized title + date. 7. Update data/source_stats.txt. 8. Classify each market-relevant item with methodology.md, including the mandatory supply-chain overlay. 9. Append confirmed tradable signals above TIMER_LEDGER_START in track-record.md. 10. Score matured rows after one trading day and one week where public quote data is available. 11. Update calibration takeaways only when measured evidence changes the base rate for a pattern.

Sightings fields

Store compact records:

{
  "date_utc": "2026-06-03T00:00:00Z",
  "source": "NVIDIA Newsroom",
  "url": "https://...",
  "title": "...",
  "summary": "...",
  "signal_type": "product|demand|supply|policy|customer|event|other",
  "primary_assets": ["NVDA"],
  "secondary_assets": ["TSM", "MU"],
  "confidence": "high|medium|low",
  "notes": "why it matters / why skipped"
}

Commit hygiene

• Keep .local/ ignored unless the user explicitly asks otherwise.
• Commit tracked updates only after confirming no secrets, account identifiers,

screenshots, broker statements, cookies, or .env data are staged.

• Use public facts only: official NVIDIA pages, public interviews, public

filings, public quotes, public policy documents, public customer/supplier disclosures, and public news.

• Do not commit paywalled article text, private transcripts, private screenshots,

or broker data.

Commit convention

• data: jensen catalyst sightings update (+<n>) <UTC ISO> for sightings-only

updates.

• data: jensen catalyst ledger update (+<n> signals) <UTC ISO> when market

ledger rows are added or scored.

• skill: jensen catalyst calibration update <UTC ISO> when methodology or

calibration changes.

• Do not create empty commits.

Notification rules

Notify only when:

• new sightings are archived and pushed;
• market ledger rows are appended or scored;
• calibration changes;
• a repeated data-source failure needs user action.

For no-change runs, already-known items, low-quality social leads, or ordinary successful checks with no durable update, return heartbeat DONT_NOTIFY XML only.

Methodology - reading Jensen Huang / NVIDIA public signals

Purpose

Classify public Jensen Huang and NVIDIA dynamics by how directly they can move NVDA, semiconductor suppliers, AI infrastructure stocks, cloud capex names, and related themes. Rank signals by immediate market reliability and separately label durability.

Signal tiers

Tier 1 - Direct Jensen / NVIDIA primary-source signal

Use when Jensen, NVIDIA Newsroom, Investor Relations, an earnings call, official keynote, or official transcript gives a concrete statement.

Common subtypes:

• Product-roadmap timing: Blackwell, Rubin, Vera/Rubin, GB200/GB300, NVL,

Spectrum-X, CUDA, software stack.

• Demand/capacity: supply constrained, sold out, lead times, wafer capacity,

HBM constraints, networking bottlenecks.

• Customer/platform: named cloud, sovereign AI, enterprise/robotics/auto wins.
• Policy: China/export controls, sovereign AI, country-level AI infrastructure.

Reliability: highest when there is a timestamp, exact quote, product/customer detail, and a public source link.

Tier 2 - Confirmed partner/customer/supplier signal

Use when a customer, supplier, regulator, or partner confirms a Jensen/NVIDIA theme but Jensen himself is not the source.

Examples:

• Cloud capex guidance confirming GPU demand.
• TSMC/HBM/networking capacity comments confirming constraints.
• Dell/Supermicro/HPE/NVIDIA server announcements.
• Government sovereign-AI contracts or export-control actions.

Reliability: high for the affected basket when multiple public sources confirm.

Tier 3 - Major-media interview / policy / market narrative

Use for credible media interviews, public event reports, and policy headlines where the quote is visible but not yet in an official transcript.

Reliability: medium. Upgrade if the quote is later confirmed by a transcript or official NVIDIA page. Downgrade if it is paraphrased or ambiguous.

Tier 4 - Social / rumor / clip lead

Use only for lead generation. Do not score as confirmed until date, source, speaker, and context are verified.

Market mapping

Supply-chain overlay - mandatory

For every market-relevant Jensen/NVIDIA item, answer these before producing a final read:

1. What part of the stack changed? Product roadmap, demand, supply, policy, customer capex, server integration, networking, memory, power, or cooling. 2. Where is the bottleneck? Foundry/CoWoS, HBM, networking/optics, server assembly, power/cooling, export licenses, or customer capex. 3. Who confirms it? Official NVIDIA, customer, supplier, integrator, regulator, trade press, or social lead. 4. Who benefits first and second? NVDA first only when economics accrue to NVIDIA; otherwise map the affected supplier/customer basket. 5. What would invalidate it? Supplier denial, capacity overbuild, export restriction, customer capex cut, integration delay, margin compression, or product slip.

Do not raise confidence above medium unless at least one customer, supplier, integrator, policy, or official NVIDIA source corroborates the supply-chain implication.

NVDA primary

• Product roadmap, demand, gross-margin implications, export controls, customer

wins, sovereign AI, software/CUDA moat.

AI server / OEM / integrators

• DELL, SMCI, HPE, Lenovo, WDC/STX secondary when storage is part of AI factory.

Semiconductor suppliers

• TSM for foundry capacity; ASML for advanced lithography; ARM for CPU/IP; AVGO

and MRVL for networking/custom silicon; MU/Samsung/SK Hynix for HBM/memory.

Networking and data-center infrastructure

• ANET, VRT, ETN, CEG, VST, NRG, GE Vernova, Schneider/ABB equivalents where

available, depending on power/cooling/networking language.

Cloud / customer capex

• MSFT, GOOGL, AMZN, META, ORCL, TSLA, CoreWeave when demand or capex is named.

Policy / China

• NVDA first; AMD/INTC second-order; Chinese AI hardware/software names only as

context unless the user's account can trade them.

Amplifiers

Add conviction when several stack:

1. Exact Jensen quote or official NVIDIA source. 2. Named product generation and launch/timing detail. 3. Named customer, country, supplier, or contract. 4. Verifiable capacity number, revenue guide, order book, or shipment timing. 5. Same theme confirmed by customer/supplier/policy source within the same week. 6. Posted during market hours or just before a major event/earnings. 7. Connects to a known bottleneck: HBM, CoWoS, networking, power, cooling, packaging, export licenses. 8. Identifies a supply-chain beneficiary or pressure point beyond NVDA.

Anti-patterns

Do not over-score:

• Generic "AI is early" vision statements without new details.
• Old clips reposted as current.
• Unsourced Chinese-language summaries without event/date.
• Product hype after the stock has already priced the event.
• Competitor or supply-chain rumors without primary confirmation.
• Paywalled headlines that hide the actual quote or context.
• NVDA-only conclusions that ignore the obvious supply-chain beneficiary,

bottleneck, or customer-capex read.

Output shape

Use this compact format:

Signal: <date/time, source, link>
Summary: <one-line public fact>
Tier: <1/2/3/4> | Type: <product/demand/supply/policy/customer/etc.>
Likely movers: <primary assets> | Secondary: <baskets>
Supply chain: <bottleneck / beneficiary / confirmation source>
Reliability: <high/medium/low> | Duration: <intraday/1-3d/1-2w/structural>
Why it may work: <amplifiers>
Why it may fail: <anti-patterns/invalidation>
Data caveat: <source/quote/price freshness>

Scoring outcomes

• For confirmed market-relevant signals, record initial reaction and T+1w

durability when practical.

• Use public quote sources with provider and timestamp. Do not call data

broker-authoritative.

• A hit means the expected direction showed up in the stated window. Record

magnitude separately.

• Preserve misses. They are the guardrail against overfitting Jensen hype.

Source map - Jensen Huang / NVIDIA public dynamics

Source hierarchy

S-tier: primary sources

• NVIDIA Newsroom: https://nvidianews.nvidia.com/
• Jensen Huang official bio/headshot: https://nvidianews.nvidia.com/bios/jensen-huang
• NVIDIA Blog: https://blogs.nvidia.com/
• NVIDIA Technical Blog: https://developer.nvidia.com/blog/
• NVIDIA Investor Relations: https://investor.nvidia.com/
• NVIDIA Events / GTC / On-Demand:
• https://www.nvidia.com/en-us/events/
• https://www.nvidia.com/gtc/
• https://www.nvidia.com/en-us/on-demand/
• NVIDIA YouTube: https://www.youtube.com/@NVIDIA
• SEC EDGAR for NVIDIA: https://www.sec.gov/edgar/browse/?CIK=1045810

A-tier: major media / wire services

• Reuters, AP, CNBC, Bloomberg, Wall Street Journal, Financial Times, Nikkei,

The Information, Fortune, TIME.

• Use these for interview quotes, policy remarks, supply-chain confirmations,

and customer/capex context. For paywalled sources, capture only public snippets unless accessible through normal browser access.

B-tier: specialist trade press / data vendors

• SemiAnalysis, Tom's Hardware, AnandTech, ServeTheHome, EE Times, TrendForce,

DigiTimes, Wccftech, The Register, TechCrunch, VentureBeat.

• Useful for technical context and early product/channel clues. Verify important

claims through NVIDIA, customers, suppliers, or major media before scoring.

C-tier: social clips / newsletters / forums

• X, LinkedIn, Reddit, YouTube clips, conference attendee posts, newsletters.
• Treat as lead generation unless the clip has a visible source, timestamp, and

enough context to verify the quote.

Recurring search queries

Use recent-date filters where available.

• Jensen Huang NVIDIA interview
• Jensen Huang NVIDIA keynote
• Jensen Huang export controls NVIDIA China
• Jensen Huang Blackwell Rubin
• Jensen Huang AI factories sovereign AI
• Jensen Huang robotics physical AI
• Jensen Huang TSMC HBM supply
• NVIDIA CEO comments data center demand
• 黄仁勋 英伟达 最新
• 黄仁勋 采访 英伟达 Blackwell
• 黄仁勋 中国 出口管制 GPU

Event watchlist

• NVIDIA GTC, CES, Computex, SIGGRAPH, Hot Chips, SC, investor days, earnings

calls, customer cloud events, Taiwan supply-chain visits, policy visits in the US/EU/China/Taiwan/Japan/South Korea.

• Watch for fresh keynote transcripts, product-roadmap slides, customer quotes,

and partner ecosystem announcements.

Customer / supplier cross-checks

Every Jensen/NVIDIA investigation should carry a supply-chain read. Do not stop at "NVDA up/down" when the signal implies upstream capacity, midstream integration, or downstream customer capex.

Downstream customers / demand confirmation

• Hyperscalers and labs: MSFT, GOOGL, AMZN, META, ORCL, TSLA, xAI,

OpenAI-related partners, CoreWeave.

• Enterprise / OEM / sovereign-AI buyers: Dell, HPE, Lenovo, government AI

factories, telecom/enterprise data-center operators.

System integrators / AI-server layer

• DELL, SMCI, HPE, Lenovo, Foxconn/Hon Hai, Quanta, Wiwynn, Wistron, Inventec,

Celestica, Jabil.

• Watch order books, rack-scale shipments, liquid-cooling readiness, quality

issues, and margin pressure.

Foundry / packaging / equipment

• TSM, ASML, AMAT, LRCX, KLAC, Tokyo Electron equivalents when visible.
• Watch CoWoS/advanced packaging, 3nm/2nm capacity, reticle/wafer constraints,

and Taiwan geopolitical risk.

Memory / HBM

• MU, Samsung, SK Hynix.
• Watch HBM3E/HBM4 qualification, capacity expansion, pricing, yield, and

allocation to Blackwell/Rubin.

Networking / optical / custom silicon

• AVGO, MRVL, ANET, COHR, LITE, AAOI, ARM.
• Watch Spectrum-X, InfiniBand/Ethernet share, optical transceivers, CPO,

switching silicon, and custom accelerator spillover.

Power / cooling / data-center infrastructure

• VRT, ETN, CEG, VST, NRG, GEV, Schneider/ABB equivalents where available.
• Watch AI factory power contracts, grid bottlenecks, liquid cooling, UPS,

switchgear, transformers, and nuclear/gas power tie-ins.

Use these as confirmation and mapping. A customer capex statement confirms demand; a supplier capacity comment confirms bottlenecks; an integrator shipment comment confirms conversion from roadmap to revenue.

Supply-chain search pack

Pair Jensen/NVIDIA searches with these supply-chain searches:

• Jensen Huang TSMC CoWoS Blackwell capacity
• NVIDIA Blackwell HBM3E HBM4 SK Hynix Samsung Micron
• NVIDIA GB200 GB300 liquid cooling Vertiv Eaton
• NVIDIA Spectrum-X Broadcom Marvell Arista optical
• NVIDIA AI factory Dell Supermicro HPE Foxconn Quanta
• NVIDIA data center power demand CEG VST ETN VRT
• 黄仁勋 台积电 CoWoS Blackwell 产能
• 英伟达 HBM SK海力士 三星 美光 Blackwell
• 英伟达 AI服务器 鸿海 广达 纬创 纬颖 散热 电力

Source caveats

• Jensen often uses broad strategic language. Do not over-score visionary

comments unless paired with product timing, customer names, supply capacity, or policy detail.

• Media summaries can compress technical terms. Preserve the original quote or

official transcript where possible.

• Conference clips can be miscaptioned or old. Verify date and event.
• Do not scrape around paywalls or private events. Use public sources.

Track record - live catalyst ledger

Purpose

Measure whether Jensen Huang / NVIDIA public signals actually move NVDA and the AI-infrastructure basket, instead of assuming every quote has edge. Keep this file append-only once the timer starts.

Calibration takeaways

• Most reliable immediate signals: exact Jensen/NVIDIA primary-source

comments about product timing, demand, capacity constraints, and named customers.

• Most durable signals: customer capex + supplier capacity + NVIDIA product

roadmap all confirming the same theme.

• Most fade-prone signals: generic AI-optimism quotes, social clips without

date/context, and trade-press product hype after a known keynote.

• Highest policy risk: China/export-control comments. These can move NVDA

sharply but need regulatory confirmation.

Live sightings archive

data/sightings.json and data/sightings.csv store public items found by the timer, including items that are not worth adding to the market ledger.

Live market ledger

Append new confirmed market-relevant signals above this marker after each refresh.

Rolling hit-rate

Scoring rules

• Score only confirmed public items with source, timestamp, URL, and affected

asset list.

• Use public quotes only as market context and label provider/timestamp.
• Define a hit as the predicted direction showing up over the expected window.
• Record durability separately. A keynote spike that fully fades is not a

durable hit.

• Do not rewrite history. Corrections should be appended in Notes.