Dossier · NVDA · Recently exited
NVDA · NVIDIA Corporation
Last analysed ·
Current thesis
Q1 FY27 binary cleared (~5/28); AI-capex demand printing harder (SpaceX-Google 110k-GPU deal 6/06, Oracle $100B capex 6/05) even as a 6/05 risk-off pulse and clustering AI-bubble warnings (Chamath 6/07) turn the tape two-sided. NVDA is the cluster laggard (+18% YTD vs STM +190%) un-stretched catch-up, not chasing the leader.
Invalidation trigger
Weekly close below the 50-DMA zone (~$186), OR Q2 FY27 Data Center guide <~$40B sequential at the ~August print, OR a named top-4 hyperscaler publicly cuts NVDA order volume for in-house silicon, OR AMD MI400/MI355X verified >70% of GB300 cost/token at a named hyperscaler.
Thesis status
Open commitment scored if the trigger above fires How this is scored →Current Thesis
NVDA is the anchor of the AI-capex buildout, and the Q1 FY27 binary (~2026-05-28) is cleared the drift window is open with no NVDA earnings blackout in front of it until the ~August Q2 print. The supply side reaffirmed in early June (TSMC capacity warning + price-hike hint, 2026-06-04) and the demand side printed harder (SpaceX-Google $920M/month deal locking 110,000 Nvidia GPUs through 2029, 2026-06-06; Oracle capex toward $100B as Stargate expands, 2026-06-05). The complication is new: a 2026-06-05 risk-off pulse (Nasdaq 100's worst day since the Trump-tariff shock) and a clustering bubble-warning tape (Chamath 6/07, "Is the AI bubble about to burst" 6/06, "3 AI stocks to take profits on" 6/05) flipped the theme from a one-way run to a two-sided, late-cycle stance for the first time in weeks. NVDA itself remains the cluster laggard (+18% YTD vs STM +190%, AMD at a 52-week high), so the expression here is un-stretched catch-up rather than a chase of the leader.
Bull Case
- SpaceX-Google deal (2026-06-06): $920M/month, 110,000 Nvidia GPUs locked through 2029 multi-year demand contractually booked rather than Street-modeled.
- Oracle capex toward $100B as Stargate expands (2026-06-05): adds another mega-buyer on top of MSFT/META/GOOGL/AMZN, ~35% Street-modeled to flow to NVDA silicon.
- TSMC capacity warning + price-hike hint (2026-06-04): C.C. Wei says AI capacity stays tight "a very long time"; foundry pricing power flows toward NVDA.
- TSMC "rivals have years to catch up" (2026-06-04): the leading-edge stack is not commoditizing on the timeframe April's custom-silicon bears feared.
- Semiconductor dip bought aggressively (2026-06-05): the chip sell-off found immediate sponsorship a shakeout, not distribution.
- NVDA is the cluster laggard (2026-06-03): +18% YTD vs STM +190% and AMD RSI 75.3 at a 52-week high the lowest-beta, least-stretched way to hold the theme.
- Post-earnings drift live: binary cleared ~2026-05-28, no blackout until ~August; CNBC's Kevin Simpson added (6/05) and BofA framed NVDA as still the answer (6/05).
Bear Case
- Bubble-warning cluster (2026-06-06, 2026-06-07): "Is the AI bubble about to burst" plus Chamath saying companies are overspending as cheaper models close the gap demand durability is now openly questioned at the headline level.
- Risk-off pulse (2026-06-05): Nasdaq 100's worst day since the Trump-tariff shock, with rate-hike bets building (6/03) direct multiple-compression risk for the highest-beta AI names.
- "3 AI stocks to take profits on right now" (2026-06-05): profit-taking lists at the index top are a late-stage sentiment tell.
- NVDA lags its own cluster (2026-06-03): a momentum book buys the leader, and the leading expression is AMD/STM/AVGO/MRVL, not NVDA rotation to the 2nd-derivative names is live.
- Custom-silicon clock still running: Marvell into the S&P 500 on the AI-chip boom (2026-06-06) and Broadcom's strong guide (6/04) partly represent merchant-silicon displacement of NVDA share.
- Cluster overbought: AMD RSI 75.3 (6/03) means mean-reversion risk is complex-wide, and NVDA's tight correlation drags it along on any unwind.
Setup & Price Structure
NVDA is the un-stretched member of an overbought complex. +18% YTD (6/03) leaves it far below the "50-DMA reclaim while leading" bar that would justify maximum size. The structure that matters is a higher-low reclaim of the 50-DMA the ~$186 zone that has acted as support on rising relative strength versus the cluster. The 2026-06-05 chip sell-off and its immediate dip-buy define a near-term shakeout low; a hold above that level with the broad tape stabilizing is the constructive path. Chasing into the 6/05–6/06 two-sided tape without a reclaim is the lower-quality entry, so a scaled approach beats a lump-sum one here. The name only earns leader status and the case for max sizing on a confirmed 50-DMA reclaim that outpaces STM/AMD on relative strength.
Catalyst Calendar (next 30 days)
- ~2026-06 second half: White House AI-companies meeting "maybe next week" per Trump (2026-06-05) soft, undated policy catalyst.
- Ongoing: hyperscaler capex prints land between earnings Oracle toward $100B (6/05) and the SpaceX-Google GPU deal (6/06) are the recent ones to watch for follow-through.
- No NVDA-specific hard catalyst in the window: Q2 FY27 print is ~August; no earnings blackout until then.
- Macro overhang: the rate-hike-bet build (2026-06-03) plus any hot CPI/jobs print is the near-term swing factor for AI-beta multiples.
What Would Change Our Mind
- A weekly close that loses the 50-DMA zone (~$186) on expanding volume the shakeout converts into a trend break.
- Q2 FY27 Data Center guide below ~$40B sequential when it prints (~August) the demand-acceleration thesis stalls.
- A named top-4 hyperscaler publicly cutting NVDA order volume in favor of in-house silicon moat-compression becomes a booked fact.
- AMD MI400/MI355X verified above ~70% of GB300 cost/token at a named hyperscaler the merchant-silicon displacement clock accelerates.
- The 6/06–6/07 bubble warnings converting into actual capex guide-downs from MSFT/META/GOOGL/AMZN the demand prints stop landing and the multiple deflates.
Correlation Notes
NVDA correlates tightly with TSM, MU, AMD, STM, AVGO, MRVL holding three-plus is effectively one concentrated AI-compute bet. AVGO/MRVL strength is double-edged: it confirms AI-infra spend while partly representing custom-silicon displacement of NVDA share. The cluster is overbought (AMD RSI 75.3, 6/03) while NVDA lags (+18% vs STM +190%), making NVDA the lower-beta way into the theme. Upstream, memory and foundry confirm the ramp SK Hynix 192GB SOCAMM2 mass production (4/20) and TSMC's capacity warning (6/04) both read through NVDA's H2 FY27 Rubin ramp. Macro correlation rose on 2026-06-05: AI names traded as a single risk-on block against the Nasdaq sell-off, so a broad risk-off event hits the whole cluster together regardless of NVDA-specific news.
Notes
- 2026-04-19: AI chip monopoly every narrative cycle anchor; watch post-earnings drift + hyperscaler capex commentary
- Earnings blackout: do not size up 3 trading days before ~2026-05-28 Q1 FY27 print
- Hyperscaler capex block Apr 29–May 1 pre-trades NVDA by ~3 weeks watch GOOGL/MSFT/META/AMZN guide
- Custom-silicon tape is accelerating (GOOG-MRVL, META-AVGO): every new partnership compresses multiple
- Rubin supply chain confirmed via SK Hynix 192GB SOCAMM2 mass production 2026-04-20 bull datapoint
- If entering
- scale in; do not chase the 04-16/17 squeeze leg
- AI chip monopoly every narrative cycle anchor; watch post-earnings drift + hyperscaler capex commentary
- Custom-silicon tape is ACCELERATING (GOOG-MRVL, META-AVGO): every new named merchant-silicon partner compresses moat premium
- Rubin supply chain confirmed via SK Hynix 192GB SOCAMM2 mass production 2026-04-20 bull datapoint
- H2 FY27 ramp validation
- Chamath 'five-alarm fire' on AI compute (2026-04-20) and CoreWeave +50% = demand-side prints are still ripping fade the overheated call with caution
- Computex ~2026-05-19/22 is a mini-catalyst inside the pre-earnings blackout window
- If entering
- scale in on pullback to 50-DMA or higher-low structure; do not chase the 04-16/17 squeeze leg
- 2026-06-04: Q1 FY27 binary CLEARED (~5/28) RECENTLY_EXITED watchlist status is stale; this is now a fresh-entry candidate with NO earnings blackout until ~Aug Q2 FY27 print.
- AI chip monopoly every narrative cycle anchor; watch post-earnings drift + hyperscaler capex commentary (the operator's standing add-reason).
- Cluster-dup risk: NVDA correlates tightly with held/watched TSM/MU/AMD/STM/AVGO/MRVL do NOT double-book the AI-compute bet; owning 3+ = one concentrated position.
- NVDA is the YTD cluster LAGGARD (+18% vs STM +190%, AMD 52wk-high RSI 75.3 on 6/03) buy as the un-stretched catch-up expression, NOT as the leader; upgrade to SUPREME only on a confirmed 50-DMA reclaim while LEADING on relative strength.
- Scale in; do not chase a vertical extension. Stop inverting conviction vs outcome on the AI anchor.
- TSMC supply-constraint commentary (multi-year shortage + price hikes, 6/04) is a structural bull read-through for NVDA pricing power; custom-silicon (AVGO/MRVL) headlines remain the moat-compression clock any 'NVDA volume CUT' framing is a sell trigger.
- Macro watch: mid-June FOMC/CPI (~6/17 est.) + rising yields (6/03) = size-down window for the whole semi complex, not an entry trigger.
- Earnings blackout: do not size up within 3 trading days of the ~August 2026 Q2 FY27 print; no blackout until then.
- AI chip monopoly every narrative-cycle anchor; watch post-earnings drift + hyperscaler capex commentary (standing add-reason).
- Cluster-dup risk: NVDA correlates tightly with TSM/MU/AMD/STM/AVGO/MRVL owning 3+ is effectively one concentrated AI-compute bet.
- NVDA is the YTD cluster laggard (+18% vs STM +190%, AMD RSI 75.3 at 52-wk high 6/03) the un-stretched catch-up expression; only upgrade to SUPREME on a confirmed 50-DMA reclaim while LEADING on relative strength.
- New 6/05–6/07 tells: risk-off pulse (Nasdaq worst day since tariff shock), AI-bubble-warning cluster, take-profit lists theme shifted from one-way ACCELERATING to two-sided late-cycle, while demand prints (SpaceX-Google, Oracle) stayed loud.
- Demand anchor: SpaceX-Google $920M/month, 110,000 Nvidia GPUs locked through 2029 (6/06) multi-year contractual demand, not modeled.
- Scale in; do not chase the two-sided 6/05–6/06 tape without a higher-low 50-DMA reclaim.
Related · shared themes
TSM
Taiwan Semiconductor Manufacturing Company Ltd.
ASML
ASML Holding N.V.
ASML is the cleanest second-derivative on the AI-capex super-cycle: EUV monopoly, CEO's "supply-limited market for quite a while" framing (5/20), High-NA first product data on logic AND DRAM later in 2026, JPM Overweight PT $2,200 (6/3). Theme is classifier-ACCELERATING with cluster confirmation and 7 rules fired (momentum top-25%, +19pp vs SPX 20d, near-52w-high structure, RSI 67.6, StockTwits +87%). ASML is the quality expression, not diversification vs TSM. Entry is a vertical near-52w-high, not the pullback-with-volume gate. Avoid: add AI-infra via a non-correlated name, or take ASML only on a 20/50-DMA pullback once the TSM cluster slot frees.
AVGO
Broadcom Inc.
Q2 FY26 (2026-06-03) printed a monster $30B AI bookings vs $10.8B shipped (2.8x book-to-bill), $100B+ FY27 AI guide but AVGO sold the news (-12% on 2026-06-04) and kept falling into a rate-driven risk-off. Fundamentals ACCELERATING, tape CORRECTING; the buyable structure is the higher-low that holds, not the knife.
AMD
Advanced Micro Devices, Inc.
AI-chip ACCELERATING but (rank 34%), parabolic ATH chase, ARK persistent distribution dossier wants the ~$420 shelf retest.