Phase Space AI

Trade Construction

NVIDIA [NVDA]

NVIDIA [NVDA] — Trade Construction

Spot $190.01 (2026-07-29) · 12-month target $266 (+39.8%) · downside $121 (−36%) · State B · evidence grade A

The memo issues no position verdict. This section specifies how an allocation would be expressed if the book chooses to make one, and what would take it off.

---

1. Vehicle — EQUITY

Equity is the default and needs no argument. Here the alternative was measured rather than dismissed:

A LEAP requires measured IV minus trailing 252-day realised vol, and the number is positive by ~6 volatility points — long-dated vol is being offered at a premium to what the stock has actually delivered. Paying that premium to express a thesis whose central driver is a five-year revenue path is a poor trade construction, so:

Vehicle: common equity. No options overlay is argued for at this price.

For completeness, the LEAP is constructable if the book overrides: NVDA280121C00190000, bid $43.69 / ask $44.04, bid size 155 / ask size 87 contracts, delta 0.6497, IV 42.18%. Quoted size on the offer is 87 contracts ≈ 8,700 shares of exposure at ~0.65 delta, which is real but thin relative to any meaningful allocation. Nothing shorter-dated is admissible: the thesis horizon is 5 years for the valuation criteria and 12 months for the target, and no single catalyst inside a shorter expiry resolves either.

2. Entry

• Entry zone $180–195. Spot sits at 18.16x NTM sales — the 0.6th percentile of NVIDIA's own AI-regime multiple history (796 observations, median 28.73x, minimum 17.50x). The stock has traded lower on this metric on five days out of 796 in three years. There is no valuation argument for waiting.
• Do not chase above $215. At $215 the multiple returns to ~20.6x NTM sales, the AI-regime p10, and the target's upside narrows to +24%.
• Scale, do not lump. Momentum is quintile 3 (12-1 at the 58.2nd percentile, 6-1 at the 61.5th, RSI(14) 44.3, 83.6% of the 52-week high, above the 200-day). The tape is neutral, not confirming. Momentum is MEASURED — entry timing only, never a selection veto. Three tranches: one-third now, one-third on the Q2 FY2027 print, one-third on either a $175 test or confirmation of the Rubin volume ramp.

3. Sizing

Sizing is the book's decision under its own exposure and correlation limits, not the memo's. Three constraints this memo hands the book:

1. Evidence grade A — every load-bearing figure is verified against EDGAR primary filings, to the dollar, on both the 10-K and the 10-Q. Uncertainty here is about the world, not about the data. Per valuation.md Rule 5, uncertainty reduces position size and never the operating assumption; this name's data uncertainty is minimal, so no data-driven haircut applies.
2. The correlation problem is the binding constraint, not single-name risk. NVDA is the central node of the AI complex, and this corpus already covers MU, AVGO, MRVL, AMD, TSM-adjacent names, COHR, FN, LITE, CRDO, KLAC, LRCX, AMAT, STX, WDC, SNDK, DELL, NBIS, CORZ, IREN, WYFI and the hyperscalers. An NVDA position is not a diversifier — it is close to a beta on the sum of them. The book must size it against aggregate AI-complex exposure, not against a single-name limit.
3. Concentration is the FAIL (54% of revenue and 64% of receivables from three customers). Position size is the correct place to carry that, because there is no hedge for it: the three counterparties are not disclosed by name, so a paired short against them is not constructable from NVIDIA's filings.

4. Exit and invalidation — the falsifiable list

Thesis-invalidating (exit, do not average down):

1. Operating margin below 55% in any quarter in which the top-three customers' capex growth is still positive. This is the State B confirmation test in reverse (see Valuation §1) and it would break the 58.0% terminal margin.
2. Total manufacturing/supply/capacity commitments falling two consecutive quarters from $119bn. This is the only quantified forward series NVIDIA publishes; the KLA precedent (RPO $11.40bn → $7.86bn while revenue rose 15.8%) is what a rolling cycle looks like in a commitment series.
3. Any one of the three named direct customers disappearing from the concentration note while total revenue growth decelerates — the customer-deferral downside cause, realised.
4. A fresh excess-inventory / purchase-obligation charge above $5bn not attributable to an export restriction. The $4.5bn H20 charge had a policy cause; an equivalent charge with a demand cause is the commitment asymmetry breaking.
5. Required CAGR from the reverse DCF rising above ~40% on a re-run at a higher price — the point at which the +43.58pp margin closes.

Thesis-confirming (add):

1. Top-three concentration falling below 45% of revenue with total growth intact — the ACIE diversification working.
2. Gross margin holding ≥73% through two further quarters of rack-scale mix shift — the terminal 70.0% proving conservative.
3. H200 or successor China licences converting to disclosed revenue — pure upside to a guide that assumes zero.

Not an exit signal: a gross or operating margin that exceeds anything NVIDIA has printed before. Per MEMO_BRIEF, that is historical implausibility, not arithmetic impossibility, and this memo has already adopted one such figure after verifying it against EDGAR to the dollar.

5. What this memo cannot hedge, and says so

The bear case (custom silicon) plays out inside a mix series NVIDIA stopped publishing this quarter. There is no instrument in NVIDIA's own filings that will show Compute decelerating while Networking accelerates after Q1 FY2027. Monitoring has to be routed through third-party disclosure — Broadcom's and Marvell's reported AI/custom-silicon revenue, and hyperscaler capex splits — as an external proxy. That is a genuine reduction in the monitorability of the position and belongs in the sizing decision.