Buffett Market Report
Analysis 5 min read

NVDA August 26: The AI Capex Referendum

Four-quarter beat-but-sell pattern; Blackwell ramp, China H200, Rubin as swing factors

The Setup

NVIDIA reports Q2 FY27 earnings after market close on August 26. This is not a normal earnings event — it is the AI capex referendum for the entire hyperscaler complex.

The stock has delivered four consecutive quarters of beat-but-sell: blowout results followed by guidance that fails to match the priced-for-perfection bar. Each time, the market reprices the duration of the AI investment cycle.

Consensus: Revenue ~$91-95B (company guide $91B ±2%), EPS ~$2.08
Whisper: Revenue >$96B, EPS >$2.20
Options-implied move: ~±8-10%


Three Swing Factors

1. Blackwell Ramp (Bull: No delays, Q3 guide >$95B / Bear: Phased ramp, guide <$93B)

Blackwell is the largest silicon project in history. The bull case requires:
- No customer delays (Microsoft, Meta, Google, Amazon all on schedule)
- Yield curves on 4NP/5NP process tracking to plan
- NVLink Switch / NVLink 7.2 / CX9 networking ready for cluster-scale deployment

The bear case: phased ramp (H200 → B100 → B200 → GB200), supply constraints on CoWoS-L, networking bottlenecks pushing meaningful revenue to Q4/FY28.

Watch: Data center revenue decomposition. H100/H200 mix vs. Blackwell pre-orders. Any "supply constrained" language = bearish.

2. China H200 Reopening (Bull: 10+ approved cos, $2-4B incremental Q3 / Bear: Very few chips, Blackwell still barred)

The H200 was the "China-compliant" workaround. If the Bureau of Industry and Security (BIS) has quietly approved shipments to 10+ Chinese cloud vendors (Alibaba, Tencent, ByteDance, Baidu), that's $2-4B of pure incremental revenue with zero R&D cost.

If Blackwell remains barred and H200 shipments are token, the China revenue cliff is real — and the TAM argument for $1T+ market cap takes a hit.

Watch: "China revenue" line item. Any mention of "licensing progress" without volumes = bearish.

3. Rubin Architecture (Bull: $1T TAM articulated / Bear: Vague timeline, no customer commitments)

Rubin (2026) → Rubin Ultra (2027) is the next architecture. The bull needs:
- Named design wins at 2+ hyperscalers
- Clear TAM expansion narrative (inference + training + sovereign AI)
- Roadmap specificity: process node, memory standard (HBM4?), interconnect

Vague "next-gen architecture" slides without customer quotes = sell the news.


The Beat-but-Sell Pattern: Four Quarters of Evidence

Quarter Result Reaction Key Driver
Q4 FY24 (Feb 24) Beat + raise -10% next 5 days "China headwinds" guidance
Q1 FY25 (May 24) Beat + raise -7% next 5 days Gross margin plateau
Q2 FY25 (Aug 24) Beat + in-line -5% next 5 days Blackwell delay fears
Q3 FY25 (Nov 24) Beat + raise -3% next 5 days Capex digestion narrative

Pattern: Market prices perfection. Guidance + qualitative tone > headline beat. The "whisper number" has become the real bar.


Scenario Matrix

Scenario Probability NVDA Move SPX Read-through Semi Leadership (SMH)
Goldilocks 15% +5% to +10% SPX 7,850+ (break 7,800 call wall) SMH breakout, AI capex confidence
Base: Beat + In-line 45% -3% to +2% Chop 7,700-7,820 (managed deceleration) Range-bound, dispersion persists
Bear: Beat + Conservative 30% -8% to -15% SPX breaks 7,650 → 7,450 test Semi leadership breaks, "AI bubble" narrative
Disaster: Miss 10% -20%+ Gamma cascade 7,450 → 7,100 Systemic de-risking, VIX 30+

Positioning Into the Event

Current framework positioning (from Unified Global Macro, Aug 16):
- NVDA puts (2% of portfolio): Core convexity hedge on beat-but-sell
- KOSPI/NKY shorts (3%): Gamma contagion hedge
- Long VIX call spreads (3%): Tail protection
- Long Gold/XLF/XLE (25%): Barbell directional

Tactical adjustment for Aug 26:
- If NVDA >$135 pre-earnings (extended), add to put spreads
- If NVDA <$115 (washed out), reduce hedge, consider call spread for Goldilocks lottery
- Do not chase the print. The tradable window is the transition, not the confirmation (h/t CrossVol).


What I'm Watching Post-Earnings

  1. Hyperscaler capex commentary: Any guide cuts from MSFT/AMZN/META/GOOG = immediate AI capex spiral risk
  2. Broadcom (AVGO) read-through: AI networking revenue, custom silicon ramps
  3. Memory (MU, Samsung, SK Hynix): HBM3e/4 demand signal
  4. China supply chain: YMTC, CXMT, SMIC capacity utilization
  5. Bond market: 30Y yield reaction — if AI capex holds, bear steepening resumes

My Honest Take

I've been cautious on NVDA valuation since it crossed $120 (25x forward). Not because the technology isn't real — it is — but because the funding spiral is the bottleneck, not the silicon.

$720-745B hyperscaler capex must be funded. IG spreads at 130bps. 30Y at 5.19%. The math works until it doesn't. The moment capex/OCF breaches 1.8x or IG spreads widen >250bps, the reflexive loop reverses: higher yields → cheaper options → more autocallables → lower vol → more issuance → higher yields.

NVDA Aug 26 is the canary in that coal mine. A beat-and-raise with clean China commentary buys 3-6 months. A beat-and-conserve accelerates the reckoning.

Position: Maintain put hedge. Size small. Let the market show its hand.