Editorial Note: NVIDIA broke below its 200-day in late March, hit RSI 26 on March 30, and has now bounced back to RSI 36. The AI thesis didn't break — the war macro broke on top of it. The real risk isn't demand (still there), it's geopolitics: TSMC fabs in Taiwan, Blackwell supply chain threads running through one of the world's most contested territories. That's the hidden variable that no P/E model captures.

---

The Story Right Now

A year ago NVDA was at $108. Today it's at $172. Even after a -18.5% drawdown from its 52-week high of $212, the stock has gained +60% in 12 months. That's the baseline for understanding what's happening: this isn't a broken thesis, it's a bull market catching its breath while a war macro reshapes everything else.

The March sell-off was brutal and fast. NVDA printed RSI 26 on March 30 — the lowest reading in over a year — on the back of a 5-day straight decline from $178 to $165. Today's bounce to $172.91 (+4.7% from yesterday's $165) brought RSI back to 36.2. Volume on the bounce was 131M shares (vs 181M average) — lighter than the selling volume, which averaged 170-200M+ during the decline. That's classic oversold bounce pattern: sellers exhausted, buyers cautious.

The March 26 full scan put it plainly: NVDA's SMA200 ($178-179) was the critical level. It breached it. The golden cross technically survived by a hair, but with SMA50 at $182.78 and SMA200 at $179.46, the gap is narrowing. A few more bad weeks and the death cross triggers, which would flush trend-following capital that's still holding.

The GTC keynote news (March 20) was unambiguously bullish: Jensen at GTC "going all in on AI vision," Groq 3 AI chip launch, strategic manufacturing collaborations widening NVDA's footprint. The fundamental demand narrative didn't change. What changed was the macro context — war, dollar strength, enterprise freeze — and the supply chain anxiety.

The TSMC Taiwan risk is the one that deserves more space. It's not a new risk, but the Iran war has raised the geopolitical temperature globally. If TSMC fabs are disrupted for any reason — conflict, blockade, earthquake — NVDA's Blackwell supply chain gets hit directly. AMD uses TSMC too, but NVDA's concentration on TSMC for its highest-performance chips (GB200, H100) is extreme. There's no real alternative at that process node. This isn't a reason to sell NVDA now — it's a risk premium the market is actively pricing that didn't exist a year ago.

The March 16 StockTitan note — "NVDA remains the market's focal point with partnership and production updates that widen both software and hardware footprints" — captures the underlying bull story: CUDA moat, AI inference scaling, Blackwell ramp. These are still intact. The Motley Fool piece calling NVDA "even more undervalued than I previously thought" at March 19 prices ($178) looks prescient given the bounce from $165 this week.

---

Quick Snapshot

---

Action Matrix

---

Price Data

30-day price narrative: NVDA ranged $175-187 for the first two weeks of March in a tight consolidation. March 19-20 saw the first major breakdown (from $180 to $172) on above-average volume 170-241M shares. March 23-25 dead-cat bounced to $178. March 26-30 broke down hard to $165 low, breaching the BB lower band, with March 30 hitting RSI 26.1. Today's bounce to $172.91 is the first constructive session in two weeks. The BB lower band at $166.48 held as support.

---

Company Overview

NVIDIA Corporation ($4.2T market cap) — two segments, one story:

The Compute & Networking segment is effectively the data center AI infrastructure business. Revenue grew 122% YoY in the most recent fiscal year. The CUDA software ecosystem is the real moat — 4M+ developers, 3,000+ applications, competitive switching cost is 2-3 year ecosystem migration. AMD is trying (ROCm) but hasn't cracked it at scale.

---

Competitive Analysis

NVIDIA's moat is CUDA. Every ML framework, every research paper, every production model is written CUDA-first. This creates a software lock-in that hardware competitors can't overcome just by making better chips. AMD's ROCm is catching up but still 2-3 years from ecosystem parity.

---

Management

Jensen is the highest-conviction founder-CEO in the market. He saw AI before everyone else (2012 deep learning bet), built CUDA into an unchallengeable moat, and has navigated export controls without losing the overall growth trajectory. No succession risk in sight.

---

Financials

The Blackwell (GB200) architecture ramp is the bull case catalyst for the next 12-18 months. GB200 clusters are 30x more powerful per rack than H100 for inference workloads. Hyperscalers (MSFT, GOOGL, AMZN, META) have all ordered in quantity. The supply constraint is now fab capacity (TSMC), not demand.

---

Valuation

At $173 and ~$130B revenue, NVDA trades at ~30x forward sales. That sounds expensive until you factor in the 60%+ operating margins and 30-40% expected growth rate. On a PEG basis at current estimates, NVDA is arguably the cheapest Mag-7 stock. The market is applying a war-risk and growth-deceleration discount that creates the entry opportunity.

---

Bull Case

1. AI inference super-cycle: training is done, inference is just beginning. Every deployed AI model runs on NVDA GPUs. The installed base demand is structural
2. Blackwell ramp provides next leg of data center revenue growth — capacity-constrained by supply, not demand
3. CUDA moat: 4M+ developers, 3,000+ apps, 2-3 year switching cost. No competitor cracks this in 12-24 months
4. RSI 36 with golden cross intact; $165 BB lower band held — technically constructive
5. 1Y +60%, 5Y +1,155% — the structural winner of the decade, not a momentum trade
6. GTC 2026 reinforced AI vision + new partnerships widening moat
7. Motley Fool "more undervalued than previously thought" at $178 — now at $173

---

Bear Case

1. TSMC Taiwan concentration risk: if Taiwan conflict escalates (TSMC disruption), NVDA supply chain breaks. No fab alternative at N3/N4 process nodes
2. Death cross risk: SMA50 ($182.78) vs SMA200 ($179.46) gap is only $3.32. Any continued weakness triggers death cross and flushes trend capital
3. Hyperscaler AI capex deceleration: if MSFT, META, GOOGL, AMZN reduce capex, NVDA order book shrinks 6-9 months later
4. China export controls: H20 chip is the only NVDA product legally sold in China. Any further restriction eliminates meaningful revenue
5. RSI 36 is not extreme — still room for further compression to 25-30 before real capitulation
6. $4.2T market cap is the world's highest; any multiple compression at scale is enormous absolute dollar losses
7. AMD ROCm improving + custom silicon from hyperscalers is a 3-5 year threat to market share

---

Historical Price Events

---

Recent News

---

Supply Chain — NVDA Ecosystem

The TSMC dependency is the most underappreciated risk in the NVDA thesis. TSMC Fab 18 in Hsinchu produces virtually all of NVDA's highest-end GPUs. Arizona fabs (TSMC US) are N3/N4 compliant but won't have significant volume until 2026-2027. NVDA's own fabs don't exist — they're fabless. Any Taiwan disruption is a direct supply chain crisis, not an indirect one.

---

Related ETFs

---

Cross-References

• Watchlists: focus, ai-scan, ai-infrastructure, semis, mag7, nvda-ecosystem
• Taiwan/TSMC risk: the Iran War & Oil perspective (geopolitical risk framework)
• Sector peers: the AI-scan summaries — NVDA leads vs AMD/AVGO
• GTC context: the March 26 AI sector scan
• Previous deep dive: the March 24 note

---

Related Tickers

AMD, AVGO, QCOM, INTC (semiconductor peers) | TSM, ASML, AMAT (supply chain) | MU, SKY (memory) | MSFT, GOOGL, META, AMZN (hyperscaler customers) | MRVL, ARM (AI silicon ecosystem)

---

Verdict

NVDA at $173 is a war-macro discount applied to a structurally sound AI infrastructure thesis. The RSI 36 bounce from $165 lows is technically constructive — BB lower band held, MACD histogram is improving, and the golden cross survives by $3. The fundamental case hasn't changed: CUDA moat, Blackwell ramp, inference super-cycle. What has changed is the risk matrix: TSMC Taiwan concentration is a tail risk that deserves explicit premium, and death cross risk is alive if current weakness continues. For long-term holders, $165-175 is an accumulation zone with a 12-month base-case target of $212 (prior highs). For traders, wait for SMA20 ($178) recapture before adding — that confirms the bounce has legs rather than being another dead-cat. The single biggest thing to watch: if SMA50 ($182.78) and SMA200 ($179.46) converge and cross, flush any momentum exposure immediately. That death cross would be the signal that the AI trade needs a deeper reset before resuming.

Updated 2026-03-31. Previous version archived as the March 24 note.

Sources

• Fundamentals figures: company-reported results (quarterly/annual filings) as available at the artifact date; predates the desk's EDGAR reconciliation gate — figures not re-verified after publication.