NVDA

Current Thesis

The April custom-silicon scare never broke the stock, and as of early June the AI-capex tape is RE-ACCELERATING. The Q1 FY27 print (~2026-05-28) cleared the binary that kept us on the bench all of late May, and the post-earnings drift window is now open with no blackout in front of us until the ~August Q2 print. Supply-side confirmation is loud: TSMC CEO C.C. Wei flagged AI capacity constrained "for a very long time" and hinted at price hikes (2026-06-04), and separately said rivals have "years to catch up" (2026-06-04) — both read directly through NVDA's pricing power and moat. The semi complex is ripping (AMD at a 52-week high RSI 75.3 on 2026-06-03, STMicro +190% YTD), and "Nvidia's CEO Accelerates Semi Mania" (2026-06-03) is the narrative engine firing again. The one tell that keeps this MEDIUM-not-SUPREME: NVDA is the YTD LAGGARD of its own cluster (+18% YTD per 2026-06-03 vs STM +190%, AMD at highs). The trade is post-earnings drift in the anchor with a clean scale-in; size up to SUPREME only on a confirmed higher-low reclaim above the 50-DMA while leading the cluster on relative strength.

Bull Case

• Binary cleared (~2026-05-28 Q1 FY27 print): The 5-week dead zone that forced three straight DEFERs (05-21, 05-22, 05-26) is over. No NVDA earnings blackout now until ~August Q2 — clear runway to size into drift. Post-earnings drift has been positive 6 of last 8 quarters.
• TSMC capacity warning (2026-06-04): C.C. Wei says AI capacity stays constrained "a very long time" and hints at price hikes. Supply < demand for the foundry that builds Blackwell/Rubin = pricing power flows straight to NVDA, not away from it.
• TSMC moat datapoint (2026-06-04): "Rivals have years to catch up" — confirms the leading-edge compute stack (TSMC + NVDA) is not getting commoditized on the timeframe the April custom-silicon bears feared.
• Semi mania re-igniting (2026-06-03): "Nvidia's CEO Accelerates Semi Mania" + AMD at 52-week high (RSI 75.3, 2026-06-03) + STMicro +190% YTD = cluster-wide acceleration, not an isolated NVDA move. Strength is the setup.
• Rubin supply chain validated (2026-04-20): SK Hynix mass-producing 192GB SOCAMM2 (double bandwidth) for Vera Rubin — memory partners tool up on committed PO volume, signaling H2 FY27 ramp is real.
• Hyperscaler capex booked: MSFT $80B, META $60–65B, GOOGL $75B (since lifted toward $80B+ per 2026-06-03), AMZN $100B+ from Q4 CY25 calls; ~35% Street-modeled to flow to NVDA silicon. Broadcom CEO Hock Tan (2026-06-04) says AI is BOOSTING software, not denting it — capex appetite intact.
• Anchor is the laggard (2026-06-03): NVDA +18% YTD while STM +190% and AMD at highs. In a still-accelerating theme, the un-stretched anchor has catch-up room rather than blow-off risk.

Bear Case

• NVDA is lagging its own cluster (2026-05-26, 2026-06-03): Rules 1&2 FAILED on 05-26 (-6.3pp vs SPX, rank 44.7%); +18% YTD vs STM +190%. A momentum book buys the leader — the leading expression may be AMD/STM/AVGO, not NVDA. Money is rotating to the 2nd-derivative names.
• Custom-silicon tape still live (2026-06-04): Broadcom's strong guide is partly NVDA displacement (Meta-Broadcom, Google-Marvell TPU talks from 2026-04-20). AVGO/MRVL ripping is a double-edged datapoint — the moat-compression clock is still running.
• Whole complex overbought (2026-06-03): AMD RSI 75.3 at 52-week high = semi complex stretched; mean-reversion risk is cluster-wide, NVDA gets dragged on any unwind.
• Macro repricing risk (2026-06-03): S&P/Nasdaq dropped from records on rate-hike bets, yields surging, oil climbing. High-multiple semis sell off hardest on yield spikes; a mid-June FOMC/CPI hawkish surprise hits NVDA's multiple directly.
• Capex-efficiency pivot (2026-06-03): "AI's Next Valuation Battle Is Being Fought Over Energy Bills" — MSFT/GOOG/OpenAI pivoting to inference efficiency could slow GPU unit-growth narrative even if dollars hold.
• Insider/political selling (2026-06-03): Sen. Whitehouse sold $215K, another senator up to $500K NVDA. Minor in size but a sentiment tell at a stretched-complex moment.
• Catalyst vacuum: No NVDA-specific dated catalyst for ~30+ days until Q2 (~August). Drift can fade in a news vacuum if macro turns risk-off.

Setup & Price Structure

• No live price feed supplied this cycle; anchoring to known structure. Prior stop was 186.66 (2026-05-26); Jan-2026 ATH reclaim zone ~$153; NVDA +18% YTD (2026-06-03). The 5/26 stop implies price was in the high-$190s/low-$200s pre-print.
• Binary cleared: the ~2026-05-28 print removed the single biggest blocker. The setup is now post-earnings drift, not a pre-print gamble.
• What I want before SUPREME sizing: a confirmed weekly higher-low reclaim above the 50-DMA (~$186–190 zone, near the prior stop) on volume, AND NVDA leading — not lagging — the cluster on 1–2 week relative strength.
• Cluster context: AMD 52-week high RSI 75.3 (stretched), STM +190% YTD, TSM bullish but -2% on profit-taking (2026-06-04). The complex is hot; NVDA is the un-stretched member. The reclaim tells whether NVDA is coiled to catch up or being left behind.
• Posture: scale-in on confirmation, not a chase. Do not buy a vertical extension; buy the higher-low that holds the post-earnings base.

Catalyst Calendar (next 30 days)

• 2026-06-04: NVDA stock-moving headline session ("What's Going On With NVIDIA Stock Thursday?") + TSMC price-hike/capacity commentary read-through — momentum confirmation tape.
• ~2026-06-17 (est.): June FOMC / CPI prints — macro risk, NOT a bull catalyst. Rate-hike-bet repricing (per 2026-06-03) is the live threat to the multiple; treat as a sizing-down window, not an entry trigger.
• Ongoing (no hard date): Hyperscaler capex follow-through commentary; TSMC price-hike confirmation; custom-silicon partnership headlines (GOOG-MRVL, META-AVGO) — any "NVDA volume CUT" framing is a sell trigger.
• ~2026-08 (est.): Q2 FY27 print — OUTSIDE the 30-day window; the next NVDA binary. No blackout before then → clear to size now.
• catalyst_date: null — no NVDA-specific dated binary inside 30 days. This is a feature: no earnings blackout means the engine is free to size the drift.

What Would Change Our Mind

• Upgrade to SUPREME: confirmed weekly close reclaiming the 50-DMA on volume + NVDA leading the cluster on relative strength + a hyperscaler ADDING NVDA capacity on the Q2 commentary trail.
• Exit / SKIP (thesis broken): weekly close below ~[entry redacted] (prior stop / 50-DMA zone); a named top-4 hyperscaler announces a CUT to NVDA order volume in favor of in-house silicon; Q2 FY27 Data Center guide < ~$40B sequential; AMD MI400/MI355X benchmark verified >70% of GB300 cost/token at a named hyperscaler.
• Theme flip to SATURATED: CNBC wall-to-wall + fresh retail euphoria while NVDA fails to make new highs as AMD/STM keep ripping (relative-strength breakdown = the anchor topping while the cluster runs).
• Macro override: hawkish June FOMC/CPI surprise that breaks the semi complex together — stand down regardless of NVDA-specific narrative.

Correlation Notes

• Tight cluster: AMD, TSM, MU, STM, AVGO, MRVL all move with NVDA. Prior decisions explicitly flagged NVDA as a "correlated dup of held TSM/MU" (2026-05-22). Owning NVDA + AMD + TSM is ONE concentrated AI-compute bet, not diversification.
• NVDA = anchor but YTD laggard: +18% vs STM +190%, AMD at highs (2026-06-03). If we already hold a faster cluster name, NVDA adds correlation without adding leadership — only justified as the un-stretched catch-up expression.
• AVGO/MRVL are partial hedges: long custom-silicon names offsets the NVDA-vs-merchant-silicon axis. Long both = long the theme, neutral on who wins the moat fight.
• Macro beta: the whole complex sells off together on yield spikes (2026-06-03). Position size must account for cluster-wide drawdown, not just NVDA-specific risk.

Bottom Line

Binary cleared, theme re-accelerating, cluster confirmed, no blackout for ~10 weeks — this is the post-earnings drift setup the playbook exists to catch. The only governors are NVDA's laggard relative strength and an unconfirmed 50-DMA reclaim, which cap it at MEDIUM/HIGH and dictate a scale-in rather than a chase. Don't repeat the W21 mistake of under-sizing the anchor on "already ran" caution — but don't go max until NVDA leads the cluster, not trails it.