>>> AI INFRASTRUCTURE: WHERE THE MONEY MOVES IN 2026

AI INFRASTRUCTURE: WHERE THE MONEY MOVES IN 2026
WHAT'S HAPPENING
Nvidia-OpenAI deal collapsed. This signals the end of "vendor lock-in" as the dominant narrative. When OpenAI diversifies compute (AMD, Cerebras, Broadcom), the monopoly story dies. But the AI capex cycle continues—the margin just shifts.

THE CORE DYNAMICS
1. Nvidia's Margin Compression Is Structural
Custom silicon (Google TPU, Amazon Trainium) is now operational. When hyperscalers have optionality, they use it. Nvidia pricing power erodes from $35k to $20-25k per GPU within 12-18 months. Consensus doesn't price this because they assume monopoly forever. Wrong.
Signal to watch: Feb 25 earnings. If Huang guides gross margin below 74% for Q1 FY2027, margin compression has started.
2. Microsoft's OpenAI Play Is Binary on Ads
Microsoft owns 27% of OpenAI. Series C will dilute to ~18-20%. Unless OpenAI ads partnership happens (needs Microsoft OR Amazon as partner), equity stake value gets hammered on down-round.
The real play: If OpenAI announces ads partnership with Microsoft by Q2 2026, MSFT at $395 is a buy. If no announcement by Q3 2026, exit.
3. Amazon Is Positioned to Win on Multiple Vectors
  • Trainium/Inferentia: Production ready, cost advantage real
  • Anthropic lock-in: Exclusive AWS partnership, can't defect
  • Margin capture: Training + inference + cloud hosting
  • Equity upside: Anthropic Series C at $200-250B realistic (not 350B hype)
Amazon outperforms not because it beats Nvidia in chips, but because it captures margin from multiple layers (infrastructure + software + equity).

YOUR POSITIONS
LONG: Amazon (AMZN)
  • Trainium traction validates in Q2-Q3 2026
  • Anthropic Series C becomes real (watch for $200-250B close, not 350B)
  • AWS capex guidance (Q1 2026) signals confidence
  • Upside: 15-20% through 2026
HOLD: Google (GOOGL)
  • AI infrastructure value already priced in
  • Keep position for DeepMind/Waymo upside (separate from AI infra)
  • Monitor: If TPU third-party access announced, GCP becomes "just another cloud"
CONDITIONAL: Microsoft (MSFT) at $395
  • BUY IF: OpenAI ads partnership announced (Microsoft or Amazon as partner) by Q2 2026
  • AVOID IF: No ads announcement by Q3 2026 (equity dilution risk outweighs upside)
  • The signal: Any OpenAI ad pilot = entry justified
SHORT: Nvidia (NVDA)
  • Not via outright short (Vera Rubin surprise risk)
  • Via options: Short call spreads ($250-280 Feb 2027)
  • Profit from consolidation, cap your downside
WATCH: Infra Names + PE Vultures
  • Data center REITs (EQIX, DLR): Upside real but wait for Q1 2026 utilization data
  • PE plays: Watch for OpenAI cash pressure or hyperscaler capex slowdown (Q2-Q3 2026)

THE OPTIONS STRATEGIES (February 2026 Earnings Volatility)
Strategy 1: Long AMD Call Spreads
Trade: Buy $120 calls / Sell $135 calls (Feb 2027)
  • Cost: $3.50 per spread
  • Max profit: $11.50 (if AMD > $135)
  • Max loss: $3.50 (if AMD < $120)
Why: AMD Q4 earnings (early Feb) is binary on MI450 credibility. Spread lets you play upside with defined downside.
Listen for red flags:
  • "ROCm ecosystem maturing" (still not ready)
  • MI355 deployment units slowing (customers losing confidence)
  • No MI450 sampling timeline disclosed (too risky)
Green flags:
  • "MI450 customer sampling underway Q1 2026"
  • Gross margin stable or improving
Sizing: 5-10 spreads ($1,750-3,500 capital at risk)

Strategy 2: Short Nvidia Call Spreads
Trade: Sell $280 calls / Buy $250 calls (Feb 2027)
  • Credit: $3-4 per spread
  • Max profit: $3-4 (if NVDA < $250)
  • Max loss: $27-30 (if NVDA > $280)
Why: Nvidia Q4 earnings (Feb 25) signals margin pressure. Spread collects premium while protecting against Vera Rubin surprise. You profit from consolidation.
Listen for red flags:
  • "Customers pre-committing for future quarters" (soft orders)
  • "Demand exceeds supply" with no capacity metrics (backlog uncertainty)
  • Any mention of pricing pressure
Green flags:
  • Gross margin holds 75%+
  • Specific pricing examples (confidence in pricing power)
Sizing: 5-10 spreads ($1,500-3,000 max loss)

Strategy 3: Pair Trade (Self-Financing)
Structure:
  • Long AMD spreads ($120-135) → costs $3.50
  • Short Nvidia spreads ($250-280) → generates $3-4
  • Net cost: ~$0 (nearly self-financing)
What you're betting: AMD gains share from Nvidia over next 12 months. If MI450 validates + Nvidia margins compress, both spreads work. If both disappoint, both lose (limited to spread width).
Capital at risk: ~$5,000-10,000 for 5-10 pair trades
Exit: By Aug 2026, if one spread is clearly winning, close the loser and let winner run. By Sep 2026, close both.

THE TIMELINE
Date Event What To Watch
Feb 25 Nvidia Q4 earnings Gross margin guidance for Q1 FY2027 (74% = compression signal)
Early Feb AMD Q4 earnings MI450 sampling timeline + ROCm maturity claims
Q1 2026 Hyperscaler earnings AI capex %, custom silicon mentions, power constraints
Q2 2026 OpenAI Series C close Valuation (200-250B realistic, 350B is hype) + Microsoft ads announcement
Q2-Q3 2026 Oracle MI450 deployment Real validation if customers actually deploy (not just orders)
Aug 2026 Review options Close winners, exit spreads that are clearly dead
THE BOTTOM LINE
Don't chase consensus. Nvidia's monopoly narrative is breaking. The margin moves to:
  1. Companies with custom silicon optionality (Amazon via Trainium)
  2. Companies with exclusive workload lock-in (Amazon via Anthropic)
  3. Companies with strategic partnerships (Microsoft IF ads work)
The options strategies let you play Feb earnings volatility with defined risk. AMD/Nvidia pair trade is self-financing and captures relative performance (the real thesis).
The key question: When does OpenAI hit a cash wall? If burn rate exceeds $2B/month by Q2 2026, PE vultures enter. If muddles through, Amazon/Microsoft benefit from continued capex.
Position accordingly.

Graham Advisors | January 31, 2026