NVIDIA (NVDA)

Executive Summary & Action Plan

Verdict

HOLD — Into the 8/26 print: whisper $93–95B vs $91B guide, and the stock faded after each of the last four beats. At $217 vs $247 path, the add gate fails.

Price $216.85 | Market cap $5.25T | Target $247 | Upside +13.9% 12M market-path target · intrinsic $164–$268 (blended $242) — see Fair Value · Street $304 (n=58, reference)

The Take

The tape is cautious: four straight post-print declines despite four straight beats, and options price ±6–7% against a 2.8% average realized move. Consensus revenue of $91.9B sits just 0.9% above guide while the whisper runs $93–95B — the October-quarter guide, bar ~$103B, is the real event. What consensus misses: China is pure option (the guide embeds zero), and FCF conversion has compounded four quarters to $48.6B. We hold the core, add only at $205 and $190, trim above $265.

Action Plan

Hold the core through the print; add at $205 and $190, trim above $265, stop at $168.

• R/R 0.7:1 at $216.85 (target $247 vs bear $175 — below the 1.5 gate, so no add at market; 2.1:1 restored at the $198 average add zone) | Prob-weighted 12M +13.9% | Confidence: Medium | p_thesis_wrong 35%
• Ladder basis: 50d SMA $207, 50d low $190, 200d SMA $195; stop $168 kept from the 7/1 desk level, converging with the $164–170 DCF/bear floor.

Forecast Path

The cone is event-shaped: the 8/26 print and October-quarter guide set the 3M branch, the Rubin/HBM4 ramp and hyperscaler CY27 capex guides set the rest. Probabilities update the 7/1 vector (22/48/30) to bull 27% / base 48% / bear 25% — EPS revisions +1.64%/30d at 1.0 breadth, 2026 hyperscaler capex raised to ~$725B, and a now-live China H200 license track, against four sell-the-news prints and visible ASIC deployment; p_thesis_wrong 35% ≥ bear 25%, the 10-point gap absorbing the lower half of the base band.

By the Charts

Margins fully rebuilt from the H20 charge: gross 60.5% → 74.9%, operating 49.1% → 65.6%.

Every line grew on both axes: revenue +85% YoY, net income +211% — but $4.8B of Q1 net income is non-operating.

Data Center at $75B is 92% of revenue and effectively all of the growth.

Methods span $164–$268: the DCF is a structural floor; the forward multiples carry the case.

Fair Value — Two Layers

Layer 1 intrinsic: log-median of DCF $164 / EV/EBITDA $268 / EV/FCF $242 = $242 — down from the 7/1 desk value of $275 on DCF discipline (the prior $285 leg is not reproducible under the house 5-year/3%-terminal cap), not on deteriorating fundamentals. Layer 2 12M path: 27% × $300 + 48% × $255 + 25% × $175 = $247. The two layers sit 2% apart — no premium or compression assumption is needed. The 12M bear ($175) is consistent with the risk desk's miss-scenario landing zone ($175–195 at 3M) and the valuation bear anchor ($170). Decomposition (arithmetic identity): EBITDA growth ×1.625 ($165.5B TTM → $269B NTM (E)) × re-rate ×0.70 (31.5x → 22.1x EV/EBITDA) × payout ×1.00 = +14.1%, vs +13.9% target gap (0.2%p).

촉매 캘린더

단기 (0~6개월)

Five days out, the print is the spine of the calendar and everything after it is conditional: Jackson Hole lands the next morning, the Rubin ramp and China licenses decide whether an in-line guide gets forgiven, and the late-October hyperscaler capex prints are the first CY27 tell.

Risks & Thesis Breakers

• Priced for beat+raise (55% × Med-High): May precedent — beat plus $91B guide still fell 5.5% over two sessions; an in-line guide near $97–99B repeats it. / Hyperscaler CY27 capex plateau (30% × High): equity tolerance, not demand, is the crack — capex raises now get sold, and external financing is funding the gap. / ASIC share shift (35% × Med, slow-burn): >1M TPU v7 serving Anthropic, >1M Trainium at AWS, OpenAI–Broadcom 10GW from 2H26; share math ~85% → ~75%.
• Breakers: ① Oct-qtr guide < $97B; ② first hyperscaler CY27 capex guide-down or "flat" (Oct'26 calls); ③ GM guided <72% two consecutive quarters; ④ CoWoS/HBM lead times compress below ~6 months (demand no longer supply-gated); ⑤ Vera Rubin volume slip past Q1 CY27 with order deferrals.

Appendix