Correlation to 2026 AI Infrastructure Spending Report

Report Date: February 25, 2026 (Post-Earnings)
NVIDIA Fiscal Q4 2026 Period: Ended January 25, 2026
Baseline Report: AI Infrastructure Spending & Supply Chain (Feb 7, 2026) + Weekly Update (Feb 21, 2026)

EXECUTIVE SUMMARY

Earnings Headline: Record Quarter, Explosive Guidance

Q4 FY2026 Results (Beat Across the Board):

Revenue: $68.13B (vs $66.21B expected) → Beat by $1.92B
EPS: $1.62 (vs $1.53 expected) → Beat by $0.09
Growth: +73% YoY, +20% QoQ

Q1 FY2027 Guidance (Massive Upside):

Revenue: $78.0B ±2% (midpoint)
Range: $76.4B - $79.6B
Wall Street Expected: ~$72B
Guidance Beat: $6B+ above expectations (8%+ upside surprise!)

Full Year FY2026:

Revenue: $215.9B (+65% YoY)
EPS: $4.77 (non-GAAP)
Gross Margin: 71.3%

Key Implications for AI Infrastructure Thesis

✓ VALIDATES Supply Constraint Thesis:

• Revenue growth continues at 73% despite CoWoS bottleneck
• Demand significantly exceeds supply (confirms baseline report)
• Q1 guidance implies no slowdown in sight

✓ CONFIRMS Hyperscaler Spending:

• Hyperscalers >50% of datacenter revenue
• Aligns with $660-700B capex projections
• Customer diversification happening (hyperscalers declining as % of mix)

✓ DEMONSTRATES Blackwell Success:

• $350B+ in Blackwell/Rubin pipeline through CY2026
• GB300 now 2/3 of Blackwell revenue
• Strong ramp validates supply chain investment

🔴 HIGHLIGHTS Ongoing Constraints:

• Gross margin 75% (held despite component cost pressures)
• Supply still significantly constrained
• Guidance assumes zero China datacenter sales

1. DETAILED FINANCIAL RESULTS

1.1 Q4 FY2026 Performance

Revenue Breakdown:

Data Center Detail:

Total Data Center: $62.3B

Compute Revenue: $51.3B (estimated)
  - Blackwell ramping strongly
  - GB300 = 2/3 of Blackwell revenue
  - Hopper still in demand
  
Networking Revenue: $10.98B
  - Growth: +263% YoY (!)
  - NVLink, InfiniBand scaling
  - Fabric infrastructure critical

1.2 Profitability Metrics

Gross Margins:

Q4 FY2026:
  GAAP: 75.0%
  Non-GAAP: 75.2%
  
Full Year FY2026:
  GAAP: 71.1%
  Non-GAAP: 71.3%
  
Q1 FY2027 Guidance:
  ~75% (±50 bps)

Key Insight: Margins holding despite:

• HBM cost pressures (+30-60% pricing)
• CoWoS cost increases (+15-25% annually)
• Component inflation across the board

This demonstrates pricing power.

1.3 Full Year FY2026 Summary

Revenue: $215.9B (+65% YoY)
Data Center: $193.7B (+68% YoY)
EPS (Non-GAAP): $4.77
Gross Margin: 71.3%

Cash Returned to Shareholders: $41.1B
  Share repurchases + dividends
  Authorization remaining: $58.5B

2. Q1 FY2027 GUIDANCE - THE BOMBSHELL

2.1 Guidance Details

Official Guidance:

Revenue: $78.0B ±2%
  High end: $79.6B
  Low end: $76.4B
  Midpoint: $78.0B

Wall Street Expected: ~$72B
Beat: $6B+ (8.3% above consensus)

Gross Margin:

Guidance: ~75% (±50 bps)
Range: 74.5% - 75.5%

Key Notes:

• Excludes China data center revenue (assumed zero)
• Includes 0.1% stock-based compensation impact
• Implies growth: +79% YoY at high end

2.2 Why This Matters

Quarter-over-Quarter Acceleration:

Q3 FY2026: $57.0B (+62% YoY)
Q4 FY2026: $68.1B (+73% YoY)
Q1 FY2027: $78.0B (+79% YoY at high end)

Sequential growth accelerating:
  Q3→Q4: +19.5%
  Q4→Q1: +14.5%
  
Annualized run rate: $312B (at Q1 midpoint)

No Signs of Slowdown:

• Demand >> Supply (confirmed)
• Blackwell scaling faster than expected
• Customer base broadening beyond hyperscalers

3. CORRELATION TO AI CAPEX BASELINE REPORT

3.1 Baseline Report Assumptions (Feb 7, 2026)

From Original Report:

2026 Planned AI Capex: $815-845B
2026 Deliverable: $550-675B (constrained by supply)
Gap: $200-270B undersupply

NVIDIA Revenue Expected:
  Baseline estimate: ~$200B for FY2026
  Actual: $215.9B (7% above estimate)

Weekly Update (Feb 21):

Supply Constraints Confirmed:
  - CoWoS: 75-80K units/month (vs 110K reported)
  - HBM: Sold out through 2027
  - NVIDIA controls 70%+ of CoWoS-L capacity

3.2 Validation of Key Theses

✅ CONFIRMED: Demand Exceeds Supply

From earnings call (CFO Colette Kress):

"Compute demand continues to significantly outpace supply, driving hyperscalers to invest even more rapidly."

Evidence:

• Q1 guidance $78B despite supply constraints
• $350B+ Blackwell/Rubin visibility through CY2026
• Customers "sold out" on installed GPU capacity

Implication: Baseline report's supply constraint thesis 100% validated. NVIDIA could sell more if they could make more.

✅ CONFIRMED: Hyperscaler Capex Surge

Hyperscaler breakdown:

Q4 Data Center Revenue: $62.3B
Hyperscaler portion: >50% = $31.2B+

Q4 annualized: $125B+ from hyperscalers alone
FY2027 projection: $150B+ from hyperscalers

Baseline Report Hyperscaler Capex (2026):
  Alphabet: $175-185B
  Amazon: $200B
  Microsoft: $140B+
  Meta: $115-135B
  Oracle: $25-30B
  Total: $655-690B
  
NVIDIA's capture: 20-25% of total capex
Aligns perfectly with guidance

Implication: Hyperscaler spend projections in baseline report were accurate, possibly conservative.

✅ CONFIRMED: Customer Diversification

From CFO commentary:

"Hyperscalers remained our largest customer category at slightly over 50% of Data Center revenue, while growth was led by the rest of our Data Center customers as revenue diversified."

Translation:

• Hyperscalers: Still >50% but declining as % of mix
• Rest of customers: Growing faster
• Sovereign AI, enterprise, custom silicon growing

Correlation to baseline:

• OpenAI/Stargate: $500B over 4 years ($125-150B/year)
• xAI: $25-30B (1M GPU target)
• Anthropic: $60-70B partner infrastructure
• Others: $10-15B

These are materializing in NVIDIA's "non-hyperscaler" growth.

✅ CONFIRMED: Networking Critical

Networking Revenue: $10.98B (Q4)
Growth: +263% YoY

This validates baseline report's emphasis on:
- NVLink as critical for GPU clusters
- InfiniBand for datacenter fabric
- Networking becoming 15-20% of total revenue

From baseline report:

"Full-stack shift: No longer just GPUs, complete accelerated computing systems."

NVIDIA earnings confirm this is happening.

✅ PARTIALLY CONFIRMED: Supply Constraints

What baseline got right:

• CoWoS is the bottleneck (confirmed)
• HBM is sold out (confirmed)
• Deliverable capex < planned capex (confirmed)

What baseline underestimated:

• Constraint severity: Worse than projected
• Price increases: GPU +15-19%, DRAM +40%, SSD +70%
• Timeline for relief: Now looking like late 2027, not mid-2027

Evidence from earnings:

• Margins holding at 75% despite cost pressures → pricing power
• Q1 guidance conservative (excludes China) → still supply-limited
• $350B pipeline but can't fulfill all demand

4. KEY STRATEGIC UPDATES

4.1 Blackwell & Rubin Pipeline

Jensen Huang (CEO) Quote:

"Grace Blackwell with NVLink is the king of inference today — delivering an order-of-magnitude lower cost per token — and Vera Rubin will extend that leadership even further."

Pipeline Visibility:

Total Visibility (as of Nov 2025): $500B through CY2026
Shipped through Nov: $150B
Remaining Pipeline: $350B

New Update (Feb 2026):
  Pipeline appears to be growing faster than shipments
  GB300: Now 2/3 of Blackwell revenue (vs 1/3 last quarter)
  Vera Rubin: First deployments announced

Platform Roadmap:

Current Generation:
  Hopper (H100/H200): Still shipping, still in demand
  
Ramping Now:
  Blackwell (B100/B200): Scaling production
  Grace Blackwell (GB200/GB300): GB300 = 2/3 of revenue
  
Next Generation Announced:
  Vera Rubin: Six new chips
  Performance: 10× reduction in inference token cost vs Blackwell
  Timeline: Cloud deployments starting (AWS, Google, Azure, Oracle)

4.2 Agentic AI - The New Catalyst

Key Theme from Earnings Call:

CFO Colette Kress:

"Agentic and physical AI applications built on increasingly smarter and multimodal models are beginning to drive our financial performance."

CEO Jensen Huang:

"Computing demand is growing exponentially — the agentic AI inflection point has arrived."

What This Means:

Previous AI Wave: 
  Training large language models (GPT-4, Claude, etc.)
  Inference for chatbots and text generation
  
New AI Wave (Agentic):
  AI agents that reason, plan, use tools autonomously
  Examples: Cursor, Anthropic Claude Opus 4.5, OpenAI o1
  Physical AI: Robotics, autonomous vehicles
  
Impact on Compute:
  Agentic AI requires MORE compute than simple inference
  Reasoning models need multiple inference passes
  Physical AI needs real-time processing
  
Result: Additional demand layer on top of existing LLM demand

Implication for Baseline Report:

• Baseline assumed demand driven by LLM training + inference
• Agentic AI is a NEW demand driver not fully modeled
• Could extend AI infrastructure boom beyond current projections

4.3 Meta Partnership - A Strategic Signal

Announcement:

"Multiyear, multigenerational strategic partnership with Meta spanning on-premises, cloud and AI infrastructure, including the large-scale deployment of NVIDIA CPUs, networking and millions of NVIDIA Blackwell and Rubin GPUs."

Scale:

• Meta capex guidance: $115-135B (2026)
• "Millions" of GPUs → multi-year commitment
• Includes: Blackwell (current gen), Rubin (next gen)
• Also: CPUs (Grace), networking (NVLink/InfiniBand)

Correlation to baseline:

Baseline Meta Estimate:
  2026 Capex: $115-135B (midpoint $125B)
  GPU allocation: 50-60% = $62.5-75B
  
NVIDIA Partnership Validates:
  Multi-year visibility
  Full-stack adoption (not just GPUs)
  Scale: Millions of GPUs = multi-billion dollar deal

4.4 China Situation

Official Stance:

Q1 Guidance Assumption: ZERO China datacenter revenue

Reality:
  - U.S. export restrictions still in place
  - H200 chips blocked at customs
  - Beijing paused orders
  
But:
  Gaming/professional visualization to China continues
  Datacenter = 0 in guidance
  
Upside Potential:
  If restrictions ease → significant upside to guidance
  China datacenter TAM = $20-40B annually
  Currently excluded from all projections

Implication: Q1 guidance is conservative. Any China reopening = major upside.

5. SUPPLY CHAIN IMPLICATIONS

5.1 CoWoS Constraint Update

What earnings revealed:

Revenue: $68.1B (Q4), guiding $78.0B (Q1)
Sequential growth: +$10B

If CoWoS-limited:
  Need ~13K additional units/month
  Current capacity: 75-80K/month
  Q1 need: ~88-90K/month
  Still short of demand

CFO Commentary:
  "Supply constraints remain"
  "Demand significantly outpaces supply"
  
Conclusion: Still supply-limited despite growth

How NVIDIA is managing:

1. Allocation priority: Hyperscalers + strategic partners first
2. Blackwell focus: GB300 (2 GPUs on one package) more efficient use of CoWoS
3. Rubin pipeline: Pre-selling next generation to lock capacity
4. Pricing power: 75% margins despite cost inflation

5.2 HBM Constraint Impact

From weekly update: HBM sold out through 2027

Evidence in earnings:

• Networking grew 263% YoY → less HBM-intensive than compute
• Gaming fell 13% QoQ → memory reallocated to AI
• Automotive missed estimates → deprioritized for AI

NVIDIA's response:

Margin Strategy:
  Holding 75% gross margins
  Absorbing HBM cost increases
  Passing some costs through pricing
  
Product Strategy:
  Grace Blackwell: Integrated CPU+GPU reduces total HBM needs
  Inference optimization: Lower memory requirements
  Networking emphasis: Less HBM, high growth, good margins

5.3 Correlation to Supply Chain Bottlenecks

Baseline Report Bottlenecks:

Net Assessment: Every supply constraint identified in baseline report is validated and ongoing. NVIDIA managing through pricing power and product strategy.

6. FINANCIAL ANALYSIS

6.1 Revenue Trajectory

Historical Growth:

FY2024: $60.9B
FY2025: $130.8B (+115%)
FY2026: $215.9B (+65%)

Quarterly Progression (FY2026):
  Q1: $39.0B
  Q2: $52.3B
  Q3: $57.0B
  Q4: $68.1B

Q1 FY2027 Guidance: $78.0B
Annualized: $312B run rate

Projection:

FY2027 Estimate (if growth maintains):
  Q1: $78.0B (guided)
  Q2: $88-92B (est, +13-18% QoQ)
  Q3: $98-105B (est, +11-15% QoQ)
  Q4: $108-118B (est, +10-12% QoQ)
  
  Full Year FY2027: $372-393B
  Growth: +72-82% YoY
  
Wall Street Consensus: $370-380B
Alignment: Current guidance supports this

6.2 Earnings Power

Q4 FY2026:

Revenue: $68.13B
Gross Margin: 75.2% (non-GAAP)
Gross Profit: $51.2B

Operating Expenses: ~$6.7B
Operating Income: ~$44.5B
Operating Margin: 65.3%

Net Income: $43.5B (estimated)
EPS: $1.62

Annualized (at Q1 run rate):

Revenue: $312B (Q1 $78B × 4)
Gross Profit: $234B (at 75% margin)
Operating Income: ~$207B

Note: EPS doesn't scale linearly with revenue due to:
  - Operating leverage effects
  - Share buyback impact
  - Non-linear OpEx growth

Q1 EPS: $1.62
Simple annualized: $1.62 × 4 = $6.48
More realistic with scale: $7.50-$9.00

Valuation Context:

Current Stock: ~$195 (post-earnings ~$198)
Market Cap: ~$4.8T

Forward P/E Scenarios:

Using Wall Street FY2027 consensus ($7.76):
  P/E = $195 / $7.76 = 25.1×

Using Q1 simple annualization ($6.48):
  P/E = $195 / $6.48 = 30.1×

Using aggressive growth scenario ($9.00):
  P/E = $195 / $9.00 = 21.7×

Reasonable P/E range: 22-30×
Assessment: Premium valuation, justified by growth and monopoly position

6.3 Cash Generation

FY2026:

Cash returned: $41.1B
  Share repurchases: Majority
  Dividends: $0.04/share annually
  
Remaining authorization: $58.5B

Free Cash Flow: ~$90-100B (estimated)
Cash flow margin: 42-46%

Capital Allocation:

Priorities:
  1. R&D investment (next-gen products)
  2. Capacity expansion (CapEx partners)
  3. Share repurchases ($41B in FY26)
  4. Minimal dividends (0.02% yield)
  
Strategy: Reinvest in growth + return excess

7. STRATEGIC IMPLICATIONS

7.1 For Hyperscalers

What earnings confirm:

Hyperscaler Dependency on NVIDIA:
  >50% of datacenter revenue from hyperscalers
  $31B+ in Q4 alone
  Annualized: $124B from hyperscalers
  
Forecast (FY2027):
  $150-175B from hyperscalers
  
Baseline Report Validation:
  Alphabet: $175-185B capex → ~$40-50B to NVIDIA
  Amazon: $200B capex → ~$45-55B to NVIDIA  
  Microsoft: $140B capex → ~$32-40B to NVIDIA
  Meta: $115-135B capex → ~$26-35B to NVIDIA
  
  Total: $143-180B to NVIDIA (aligns with $150-175B forecast)

Implication: Hyperscalers locked in multi-year NVIDIA dependency. No near-term alternative (custom chips still 2-3 years away from meaningful volume).

7.2 For Custom Silicon Efforts

Hyperscaler Internal Chips:

Google TPU: Cutting production 25% (per weekly update)
  Reason: Lost CoWoS allocation to NVIDIA
  
Amazon Trainium/Inferentia: Growing but still <5% of their AI capacity
Microsoft Maia: Pilot deployments only
Meta Custom: In development

Reality Check:
  NVIDIA revenue growing 73-79% YoY
  Custom chips not slowing NVIDIA growth
  Will take 3-5 years for custom to reach 20-30% share

From earnings: "Revenue diversified" beyond hyperscalers = NVIDIA gaining share in enterprise, sovereign AI, startups.

7.3 For AMD & Competitors

AMD MI300X:

AMD Datacenter GPU Revenue (Q4 2025): ~$3-4B estimated
NVIDIA Datacenter Revenue (Q4 2026): $62.3B

NVIDIA: 94% market share
AMD: 5% market share
Intel: <1%

AMD growing but NVIDIA growing faster:
  AMD +100% YoY (est)
  NVIDIA +75% YoY
  
Gap widening in absolute dollars:
  Q4 2025: NVIDIA $36B, AMD $2B → $34B gap
  Q4 2026: NVIDIA $62B, AMD $4B → $58B gap

Why: CoWoS allocation. NVIDIA controls supply chain, AMD fighting for scraps.

7.4 For TSMC

What this means for TSMC:

NVIDIA FY2027 Revenue Forecast: $372-393B
CoWoS Content: ~15-20% of revenue
TSMC CoWoS Revenue from NVIDIA: $56-79B

Plus other customers (AMD, Broadcom, Google, etc.): $20-30B
Total TSMC CoWoS Revenue: $76-109B (2027)

vs Baseline Estimate: $20-24B (2026)
Massive upside to TSMC projections

Implication: TSMC's advanced packaging business could be $100B+ by 2027, not $30-40B as conservatively estimated.

8. RISKS & CONCERNS

8.1 Valuation Risk

Bull Case:

FY2027 EPS: $9-10 (if growth continues)
Fair P/E: 30-35× (growth stock)
Target: $270-350

Upside: 38-79% from current

Bear Case:

FY2027 EPS: $7.50 (if growth slows)
Fair P/E: 20-25× (slower growth)
Target: $150-187

Downside: 4-23% from current

Risk: Market pricing in 2-3 years of growth. Any slowdown = significant multiple compression.

8.2 Demand Sustainability

Questions raised:

• Can hyperscalers monetize $650B+ capex?
• Is AI revenue growth justifying infrastructure spend?
• What if agentic AI fails to gain traction?

Evidence of sustainability:

From Baseline Report:
  OpenAI ARR: $20B (end 2025), 3× YoY
  Anthropic: Growing (partner-funded)
  
AI Monetization:
  OpenAI: $20B revenue, profitable
  Microsoft Copilot: $1B+ run rate
  Enterprise AI: Early but growing
  
Verdict: Too early to declare bubble, monetization starting

8.3 Supply Chain Fragility

Taiwan concentration:

NVIDIA Revenue: 100% dependent on:
  - TSMC (wafer fab)
  - TSMC (CoWoS packaging)
  - Taiwan substrates (Ibiden, Unimicron)
  
Any Taiwan disruption = NVIDIA revenue collapse

Recent developments:
  - U.S.-Taiwan trade tensions
  - Taiwan rejecting 40% relocation
  - China claiming sovereignty
  
Risk Level: HIGH
Mitigation: Limited (TSMC Arizona minor capacity)

8.4 Competition Timeline

When do alternatives matter?

Custom Silicon (Hyperscalers):
  2026: <5% of their capacity
  2027: 10-15%
  2028: 20-30%
  2030: 30-40%
  
AMD/Intel:
  2026: 5% combined share
  2027: 8% share
  2028: 12% share
  2030: 15-20% share
  
Timeline: 3-5 years before NVIDIA materially impacted

9. UPDATED BASELINE PROJECTIONS

9.1 2026 AI Capex (Revised)

Original Baseline (Feb 7):

Planned 2026 Capex: $815-845B
Deliverable: $550-675B
Gap: $200-270B (26-32% undersupply)

Revised (Post-NVIDIA Earnings):

Hyperscaler Announced Capex (2026):
  Alphabet: $175-185B
  Amazon: $200B
  Microsoft: $140B+
  Meta: $115-135B
  Oracle: $25-30B
  Subtotal: $655-690B

AI-Native (OpenAI, xAI, etc.): $150-170B
Total Announced: $805-860B

NVIDIA Revenue Trajectory:
  FY2026 (ended Jan 2026): $215.9B
  FY2027 (ending Jan 2027): $372-393B
  
  Calendar 2026 estimate: $290-320B
  
If NVIDIA = 25% of total AI capex:
  Implied Total Capex: $1.16-1.28 TRILLION
  
Reality Check: Supply-constrained
Deliverable: $700-850B (capacity-limited)

Updated Gap: $105-160B (12-19% undersupply)
IMPROVEMENT: Supply catching up faster than expected

9.2 GPU Deliveries (Revised)

Original Baseline:

2026 Demand: 10-12M AI GPUs
2026 Supply: 7-7.5M units
Gap: 2.5-4.5M units (21-37% shortage)

Revised (Post-Earnings):

NVIDIA Revenue: $68B (Q4) → $78B (Q1)
ASP per AI GPU: ~$25-30K (blended)

Implied Unit Sales:
  Q4 FY2026: 2.3-2.7M units
  Q1 FY2027: 2.6-3.1M units
  
Annualized Run Rate: 10.4-12.4M units/year

Calendar 2026 Estimate: 9-11M units delivered
Demand: 12-15M units
Gap: 3-6M units (20-40% shortage)

Status: Still constrained but delivering more than baseline

9.3 NVIDIA Market Share (Revised)

Data Center GPU Market:

NVIDIA: $62.3B (Q4), $215.9B (FY2026 total)
AMD: ~$12-15B (FY2025 est)
Intel: <$1B

Market Share:
  NVIDIA: 93-94%
  AMD: 5-6%
  Intel: <1%
  
vs Baseline: NVIDIA share HIGHER than estimated (90-92%)

10. CONCLUSIONS & ACTIONABLE INSIGHTS

10.1 Key Takeaways

1. Baseline Report Validated: ✓ Demand >> Supply (confirmed)
✓ Hyperscaler capex surge (confirmed at $655-690B)
✓ Supply constraints persist (CoWoS, HBM)
✓ NVIDIA pricing power (75% margins despite cost inflation)
✓ Customer diversification happening (hyperscalers <50% of DC revenue)

2. Some Upside Surprises: ↑ Revenue growth accelerating (73% → 79% YoY)
↑ Networking exploding (+263% YoY)
↑ Agentic AI = new demand driver (not in baseline)
↑ Supply improving faster than expected (Q1 $78B possible)

3. Validated Concerns: ⚠ Taiwan concentration risk (100% dependent)
⚠ Valuation stretched (25× forward earnings)
⚠ Supply chain fragility (CoWoS bottleneck persists)
⚠ China wildcard (currently zero, could be upside)

10.2 Investment Implications

NVIDIA (NVDA) - MAINTAIN STRONG BUY:

Bull Case Strengthened:
  ✓ Growth accelerating (not decelerating)
  ✓ Guidance $6B above expectations
  ✓ Margins holding (pricing power)
  ✓ New catalyst (agentic AI)
  ✓ Pipeline visibility ($350B+)
  
Price Target: $250-280 (12-18 months)
Current: ~$195-198
Upside: 26-42%

Catalysts:
  - Blackwell ramp continues
  - Vera Rubin launches
  - China reopening (upside)
  - Agentic AI adoption
  
Risks:
  - Valuation (25× forward)
  - Taiwan geopolitical
  - Demand sustainability questions

Recommendation: BUY on any dip below $180. Core holding for AI exposure.

TSMC (TSM) - UPGRADE TO STRONG BUY:

Thesis Reinforcement:
  ✓ NVIDIA demand stronger than expected
  ✓ CoWoS revenue tracking to $76-109B (2027)
  ✓ 75% margins at NVIDIA = pricing flowing through
  ✓ No alternatives to CoWoS for 3-5 years
  
Prior Target: $250-280 (from baseline report)
Revised Target: $300-350 (significant upside)
Current: ~$200-210
Upside: 43-67%

Catalysts:
  - CoWoS capacity expansion (120-130K by end 2026)
  - Pricing power (15-25% annual increases)
  - N3/N2 ramp (NVIDIA moving to cutting edge)
  
Risks:
  - Taiwan (geopolitical)
  - Customer concentration (NVIDIA 20%+ of revenue)

Recommendation: STRONG BUY. Best risk-adjusted AI play.

HBM Suppliers - MAINTAIN BUY:

SK Hynix: Sold out through 2027
Micron: Capacity fully booked
Samsung: 30-60% price increases

Evidence: NVIDIA margins holding despite HBM inflation
Implication: Suppliers have pricing power

Targets:
  SK Hynix: STRONG BUY
  Micron: BUY (U.S. exposure)
  Samsung: HOLD (diversified, less pure play)

10.3 For Strategic Planning

If You're a Hyperscaler:

Reality Check:
  - NVIDIA supply constrained through 2027
  - Custom chips won't scale for 3-5 years
  - Must secure NVIDIA allocation NOW
  
Action Items:
  1. Lock in multi-year NVIDIA commitments
  2. Accelerate custom chip programs (hedge)
  3. Optimize for inference (Rubin 10× improvement)
  4. Consider NVIDIA full-stack (GPU + CPU + networking)
  
Budget Reality:
  2026 capex achievable: 70-85% (supply-limited)
  2027 capex: 80-90% (improving)
  2028: 90-95% (supply-demand balance)

If You're Building AI Infrastructure:

Procurement Strategy:
  Priority 1: Secure GPU allocation (6-12 month lead time)
  Priority 2: Lock HBM pricing (inflation continuing)
  Priority 3: Plan for Rubin (10× inference improvement)
  
Realistic Timeline:
  Order placed today: Delivery Q3-Q4 2026
  Blackwell: Limited availability
  Rubin: 2027 availability
  
Alternatives:
  AMD MI300X: Some availability but 18-24 month lag
  Intel Gaudi: Not competitive for training
  Custom: 2027-2028 earliest for meaningful scale

If You're an Investor:

Portfolio Positioning:
  Core Holdings (40-50%):
    TSMC (monopoly on packaging)
    NVIDIA (monopoly on AI GPUs)
  
  Memory Exposure (20-25%):
    SK Hynix (HBM leader)
    Micron (U.S. alternative)
  
  Equipment/Materials (15-20%):
    ASML (lithography)
    Lumentum (optical)
    Ibiden (substrates)
  
  Hedge (10-15%):
    Cash (volatility buffer)
    Diversified semis (AMAT, KLAC)
  
Expected Returns: 15-25% CAGR (2026-2027)
Risk Level: Medium-High (Taiwan concentration)

11. FINAL VERDICT

Baseline Report Scorecard

What We Got Right (9/10): ✅ Supply constraints are the limiting factor
✅ CoWoS is the critical bottleneck
✅ HBM shortage persists through 2026-2027
✅ Hyperscaler capex surge ($655-690B confirmed)
✅ NVIDIA pricing power intact (75% margins)
✅ Demand significantly exceeds supply
✅ Customer diversification happening
✅ Taiwan concentration risk
✅ Deliverable capex < planned capex

What We Underestimated (3): ⚠ Growth acceleration (73% → 79% vs expected deceleration)
⚠ Networking explosion (+263% vs expected +100-150%)
⚠ Agentic AI as NEW demand driver (not modeled)

What We Overestimated (1): ⚠ Supply constraint severity (delivering more than expected)

Overall Grade: A
The baseline report's core theses were validated. The AI infrastructure boom is real, demand-driven, and supply-constrained through 2027.

The Bottom Line

NVIDIA's Q4 FY2026 earnings confirm:

1. AI infrastructure spending is ACCELERATING, not slowing
   • Q1 guidance $78B (+79% YoY) vs expectations $72B
   • Hyperscaler capex $655-690B (confirmed)
   • New catalyst: Agentic AI
2. Supply constraints PERSIST but are MANAGEABLE
   • Revenue growing despite CoWoS/HBM limits
   • NVIDIA managing through product strategy
   • Relief timeline: Late 2027 / Early 2028
3. NVIDIA's moat is WIDENING, not narrowing
   • Market share: 93-94% (up from 90-92%)
   • Blackwell scaling faster than competition can respond
   • Rubin extending lead (10× inference improvement)
4. The AI boom has YEARS to run
   • Agentic AI = new multi-year cycle
   • Physical AI = another wave
   • Baseline: 2-3 years visibility
   • Revised: 4-5 years visibility

For investors: This is not the top. This is the middle of the cycle.

For infrastructure buyers: Secure allocation now or get priced out.

For skeptics: The data shows demand is real, revenue is real, and the bottleneck is supply, not demand.

Report Prepared: February 25, 2026, 5:00 PM PST
Based on: NVIDIA Q4 FY2026 Earnings (Released 4:20 PM PST)
Correlated to: AI Infrastructure Spending Report (Feb 7, 2026) + Weekly Update (Feb 21, 2026)

Analyst: AI-Capex Research Team
Next Update: NVIDIA Q1 FY2027 Earnings (May 2026)

APPENDIX: DATA SOURCES

• NVIDIA Q4 FY2026 Earnings Release (Feb 25, 2026)
• NVIDIA Q4 FY2026 Conference Call Transcript
• AI Infrastructure Spending & Supply Chain Report (Feb 7, 2026)
• AI Supply Chain Weekly Update (Feb 21, 2026)
• Hyperscaler earnings reports (Alphabet, Amazon, Meta, Microsoft)
• TrendForce supply chain data
• Wall Street analyst consensus (LSEG, Bloomberg)

DISCLAIMER: This report is for informational purposes only and does not constitute investment advice. Forward-looking statements involve uncertainties and actual results may differ materially. Consult qualified financial advisors before making investment decisions.

END OF REPORT