Morgan Stanley's NVDA Top Pick Meets $947M of Insider Selling: The AI Capex Credibility Test

Moore's note made the case in five points, per multiple wire summaries: NVDA is early in a new product cycle, the current supply constraint is a tailwind for pricing power, valuation is attractive relative to growth, NVDA is the 'best beneficiary' of hyperscaler AI spending, and the post-meeting read on management tone was constructive. The $300 target implies ~28% upside from the Oct 2 close, vs. a consensus analyst target of $327.70 per Refinitiv-style aggregation. Moore's historical track record — 61% success rate, 25.9% average return across rated names — is itself a credibility anchor on the call.

Micron is the read-through. Morgan Stanley rotated the top-pick slot from MU into NVDA — yet MU trades at 7.0x forward earnings vs. NVDA's 24.9x. The implication is not that MU is wrong; it is that Moore believes the NVDA product cycle (Vera Rubin, HBM4 ramp, sovereign-AI deals) is worth a ~3.5x forward-earnings premium over the memory incumbent. That is a forecast of growth, not a verification of it.

NVDA's Q2 FY27 results, filed Aug 26, 2026 for the quarter ended Jul 26, delivered $96.22B in revenue (+106% YoY, +18% QoQ) and $2.47 in diluted EPS. That follows Q1 FY27 revenue of $81.6B (+85% YoY, +20% QoQ) with Data Center at $75.2B (+92% YoY). Data Center represented ~92% of total revenue across both quarters. Trailing-twelve-month revenue through Q2 FY27 stands at $302.97B, with TTM net income of $192.88B and TTM EPS of $7.95 — a 63.7% net margin on a TTM basis.

The Oct 5 reference in the broader news cycle on this story is to NVDA insider filings broadly, not to a single Oct 5 transaction by Jensen Huang. [NVDA](nvda]'s IR page lists no Form 4 dated Oct 5, 2026; the most recent Jensen Huang Form 4 the SEC has filed. is the Sept 16, 2026 disclosure covering a 125,000-share sale for ~$23.6M, executed under a pre-adopted 10b5-1 trading plan. The actual heavy lifting on insider distribution in the run-up to Morgan Stanley's call came from Stevens, not Huang.

Burry's Q3 2025 13F, filed Nov 4, 2025, opened $186.6M of notional NVDA puts and $912.1M in Palantir puts. Burry deregistered Scion Asset Management on Nov 10, 2025, and now publishes a newsletter. Per a recent public comment, Burry has called NVDA's $500B AI deal stack 'a sign of desperation' with 'shades of Enron' — a view he has reiterated on his blog. The puts themselves expire, but Burry's published thesis has not.

Two independent supply-chain tracks confirm Moore's demand read. First, memory: SK Hynix signed a $500B+ multi-year partnership with NVDA and SK Group on July 25, 2026, covering HBM co-development for the Vera Rubin platform — SK Hynix holds roughly 70% of NVDA passed NVDA cleared the same bar in the same window. Samsung has guided HBM sales to more than triple in 2026 vs. 2025. HBM4 is sold out into2027.

Second, hyperscaler capex: the four largest buyers of NVDA, Alphabet, Microsoft, and Meta Platforms — have collectively guided to $720–745B in calendar-2026 capital expenditure, up roughly 78% from ~$410B in 2025, per Q2 2026 earnings calls. Amazon is at ~$220B, Alphabet $195–205B, Microsoft ~$175B, and Meta $130–145B. That is the demand side Moore is endorsing.

Short-term (days to quarters), three dates matter. First, NVDA is the most credible single insider read on the stock. Third, the Q3 earnings calls of the four hyperscalers, which historically revise2026 capex guidance upward; a flat print is itself a signal. Each of these is a binary event for the next move in NVDA.

Long-term (one to three years), the structural question is whether the HBM4 supply ramp by SK Hynix, Samsung Electronics and Micron keeps pace with NVDA reaches a level that supports $730B of annualized capex. If HBM4 capacity stays scarce through 2027, Moore's pricing-power thesis stays intact. If hyperscaler AI revenue disappoints, the $730B becomes the level to defend — not the new floor.

Morgan Stanley's Top Pick call is a forecast — Joseph Moore is saying NVDA's next product cycle converts at the multiple the bulls need. The insider record says the people closest to the company's economics are already selling at the multiple [NVDA](nvda] currently commands. Those are not contradictory facts. Stevens' sales are scheduled under 10b5-1 plans with strike prices established well before the current run; Jensen Huang's are too. The question is whether the [NVDA](nvda] earnings print in late November shows the Vera Rubin ramp in the numbers — because if it does, the insider distribution is just noise; if it does not, it is a signal.