IC Memo — NVIDIA (Core Equity Seed)

Condensed from the full IC memo. Full memo: deliverables/2026-07-11-seed/IC_Memo_NVDA.docx

Action: Seed a 6.0% Core Equity position in NVDA (≈$60,000 on the fund's $1.0mm NAV; ≈284 shares at the live price). Benchmark SPY; IPS max single position 10%.

Business quality & moat. NVIDIA is the leading merchant supplier of accelerated-computing hardware and full-stack systems for AI training and inference. The durable moat is CUDA — a decade-deep parallel-computing software layer and toolchain (cuDNN, TensorRT, NCCL) with a large registered developer base (low-single-digit millions, approximate) that creates real switching costs, not just brand loyalty. NVIDIA has extended this from "best chip" to "best system" via DGX/HGX platforms and NVLink-scale racks, raising the bar for single-point competitors.

Unit economics (approximate, FY2025 actual — not live-verified). Revenue ~$130.5B, Data Center ~$115.2B (~88% of total) — the segment this thesis is levered to. Non-GAAP gross margin ~75%, though quarterly compression appeared during the Blackwell ramp. R&D ~$12.9B, reflecting an annual architecture-refresh cadence that is itself part of the moat. Non-GAAP diluted EPS ~$3.13 (split-adjusted).

Valuation anchor. Live price $210.99 (platform quote, 2026-07-10 close, +4.09% on the day). Against the ~$3.13 approximate FY2025 non-GAAP EPS, that implies a trailing P/E of ~67x. This is a trailing, not forward, multiple — no live consensus feed was available this session, so no forward P/E is asserted. A ~67x trailing multiple only holds up if data-center growth continues at a substantial (if decelerating) pace; re-run against live consensus before any add beyond the seed weight. (Cross-check: a companion 5-year DCF model, built independently with moderate/declining-growth assumptions and a 10.5% WACC / 3.5% terminal growth, lands at ~$68/share base case — well below the live price, which is itself informative about how much growth the market is pricing in. See "NVDA DCF — Assumptions & Output.")

Three named risks.

1. Competitive / custom silicon — hyperscaler in-house accelerators and merchant alternatives target the same AI-compute budget, especially in inference where CUDA's moat is thinner.
2. AI capex cyclicality — Data Center's ~88% revenue share means a digestion phase or ROI scrutiny at a top hyperscaler disproportionately hits growth.
3. Customer concentration — a handful of hyperscale/enterprise/sovereign buyers represent an outsized share of Data Center revenue.

Exit plan. Per Core Equity IPS: weekly review vs. SPY, drift rebalance at a ±5% band around the 6% target. Thesis-break triggers: (a) a named hyperscaler shifting a majority of a new large training cluster to non-NVIDIA silicon, (b) two consecutive quarters of Data Center growth decelerating materially below the underwritten path, or (c) non-GAAP gross margin falling and staying meaningfully below ~75% without a ramp-related explanation.