YK Research

NVIDIA Corporation (NVDA)

The Inference Era: GPU Training Gives Way to Full-Stack AI Infrastructure

Last updated: 14 March 2026 · YK Research

1. March 2026 Update

Current Price: $180.25  |  Market Cap: ~$4.5T

Key Development: NVIDIA closed its $20B Groq acquisition in December 2025. Largest deal in company history. This is a hard pivot to inference computing. NVIDIA is no longer just a GPU company. It is a full-stack AI infrastructure platform for training and inference.

Data Center Revenue
$60.9B (TTM)
YoY Growth
217%
Gross Margin
75%
AI Training Share
92%

2. Company Snapshot & The Three-Layer Moat

Founded
1993
CEO Tenure
31 Years
CUDA Developers
4M+
R&D Investment
$8.7B

Layer 1: GPU Architecture

Five major GPU generations. Each delivered order-of-magnitude jumps in transistor density and compute. Competitors can't catch up because NVIDIA keeps moving the target.

Layer 2: CUDA network

Developer Lock-in

4 million CUDA developers. Largest GPU computing community ever built. Every AI researcher, every ML framework, every cloud provider runs deep CUDA dependencies. Switching takes years.

Software Stack Depth

CUDA is not just a language. It is cuDNN for deep learning, cuBLAS for linear algebra, TensorRT for inference, NCCL for multi-GPU comms. Hundreds of libraries refined over 18 years.

Layer 3: Full-Stack Platform

  • DGX systems: turnkey AI supercomputers
  • Networking (Mellanox/InfiniBand): GPU-to-GPU fabric
  • NVIDIA AI Enterprise: software licensing revenue
  • Omniverse: digital twin simulation platform
  • DRIVE: autonomous vehicle computing platform
  • Groq acquisition (2025): inference-optimized hardware
🔍
NVIDIA doesn't sell chips. They sell the entire AI development workflow. Leaving means rewriting your software stack, retraining your engineers, and eating 2-3 years of lost productivity.
Senior ML Engineer, Major Cloud Provider

3. The Lindy Effect: 60 Years of Semiconductor Survival

The semiconductor industry survived multiple extinction events over 60 years. Every crisis was supposed to end Moore's Law. Instead the industry adapted and got stronger.

Computing Waves

  • 1960s Mainframes: IBM dominated. Skeptics called semiconductors a fad. The industry thrived.
  • 1980s PC Revolution: Intel and AMD emerged. Critics said PCs would commoditize chips. Instead: massive new demand.
  • 2000s Internet/Mobile: Qualcomm and ARM rose. “End of Moore's Law” predicted. The industry found new scaling paths.
  • 2010s Cloud Computing: Custom chips (Google TPU) threatened. GPU computing emerged as the dominant model.
  • 2020s AI Revolution: Current wave. NVIDIA is the primary beneficiary.

NVIDIA's Resilience Factors

  • Founder-led: Jensen Huang has run NVIDIA for 31 years. One of the longest CEO tenures in tech. Founder-led companies outperform.
  • Platform, not product: NVIDIA sells a network, not hardware. Recurring revenue. Deep lock-in.
  • R&D intensity: $8.7B annual R&D. More than most competitors' total revenue.
  • Multiple growth vectors: AI training, inference, automotive, robotics, digital twins. If one slows, others accelerate.

4. The 18-Year CUDA Moat

CUDA launched in 2006. It is the de facto standard for GPU computing. The moat is not the technology. It is 18 years of developers, libraries, frameworks, and institutional knowledge built on top of it.

Developer network Comparison

Why Customers Can't Leave

Code Rewrite Cost

Millions of lines of CUDA code in production. Rewriting for another platform takes 2-3 years and costs hundreds of millions.

Talent Lock-in

Every ML engineer learns CUDA. Universities teach CUDA. Switching means retraining your entire workforce. That is a multi-billion dollar proposition.

Framework Dependencies

PyTorch, TensorFlow, JAX. All optimized for CUDA first. Alternative backends lag 6-12 months in performance and features.

Library network

cuDNN, cuBLAS, TensorRT, NCCL, Triton. Hundreds of optimized libraries. Each represents years of work. Nothing equivalent exists elsewhere.

Competitive Attempts to Challenge CUDA

  • AMD ROCm: Open-source alternative since 2016. Ten years later, still a fraction of CUDA's network. Missing libraries. Poor docs. Limited cloud support.
  • Intel oneAPI: Good vision, slow execution. Gaudi accelerators show promise but lack software maturity.
  • Google TPU/JAX: Works inside Google. Limited adoption outside. JAX is growing but niche compared to PyTorch.
  • OpenAI Triton: Promising open-source GPU language. Still requires NVIDIA GPUs today. Potentially disruptive in 5+ years.

5. The $20B Groq Deal

Deal Structure

Deal Value
$20B
Announced
Dec 2025
Expected Close
Q2 2026
Deal Type
All Cash

NVIDIA's largest acquisition ever. Groq's Language Processing Unit (LPU) is purpose-built for inference. It complements NVIDIA's GPU training dominance.

Why Inference Matters More Than Training

The market is shifting from training (building models) to inference (running them in production). Training is a one-time cost. Inference runs continuously. As AI scales, inference spending will dwarf training by 10x.

Groq's SRAM Advantage

Groq's LPU uses SRAM instead of HBM. SRAM is faster but more expensive per bit. For inference where latency beats capacity, this is the right trade-off.

Integration Plan

  • Phase 1 (2026): Integrate Groq LPU into DGX Cloud for inference-as-a-service. Seamless training-to-inference pipeline.
  • Phase 2 (2027): Hybrid GPU+LPU systems. Train on NVIDIA GPUs, auto-deploy to Groq LPUs for production.
  • Phase 3 (2028+): Unified architecture combining GPU compute with SRAM-based inference. The full-stack AI platform.
🔍
The Groq acquisition is NVIDIA's iPhone moment. Hardware innovation plus platform integration. They now own the entire AI deployment stack.
Industry Analyst

6. Market Dominance & Competition

AI Training Share
92%
Data Center GPUs
#1
Cloud GPU Instances
85%+
ML Framework Support
100%

Competitive Threats

AMD (MI300X)

Most credible GPU competitor. Strong hardware but ROCm software is years behind CUDA. Gaining share in inference where software matters less.

Google TPU

TPU v5 competes on training. But it only runs on Google Cloud. No on-prem option. JAX is growing but still niche.

Custom ASICs (Amazon, Microsoft)

Amazon Trainium and Microsoft Maia serve their own cloud workloads. Reduces NVIDIA dependency internally but does not serve the broader market.

Cerebras / Emerging Players

Wafer-scale computing has interesting performance numbers. High costs, limited software, and manufacturing challenges limit near-term impact.

NVIDIA's Response

  • Accelerate roadmap: Annual GPU cadence. Blackwell (2024) to Rubin (2025) to next-gen (2026).
  • Expand software moat: AI Enterprise software licensing creates recurring revenue and deeper lock-in.
  • Full-stack integration: DGX, networking, software, and Groq inference. No competitor matches the complete solution.
  • Strategic pricing: Aggressive cloud instance pricing to maintain volume while hardware ASPs rise.

7. Valuation & Scenario Analysis

Current Price
$180.25
Forward P/E
35×
PEG Ratio
0.8×
EV/Revenue
25×

12-Month Price Scenarios

Bear Case
$120
-33%

AI capex cycle peaks. Competition intensifies. Groq integration struggles. Revenue growth slows to 30%. Multiple compresses to 25x.

Base Case
$240
+33%

AI infrastructure buildout continues. Groq integration begins. Inference revenue ramps. Revenue growth 50%+. Multiple holds at 35x.

Bull Case
$360
+100%

AI spending accelerates beyond expectations. Groq creates a new market category. Sovereign AI demand surges. Revenue growth 80%+. Multiple expands to 45x.

TAM Expansion: 2030 Addressable Market

Groq Impact on Valuation

Groq adds $50-100B to NVIDIA's addressable market by 2030. 10% inference share means $25B+ annual revenue by 2030. The $20B price tag looks cheap.

  • Revenue overlap: Inference-as-a-service adds high-margin recurring revenue.
  • TAM expansion: Opens inference segments where GPUs are too expensive or power-hungry.
  • Competitive moat: Owning training and inference hardware creates an unassailable position.
  • Integration risk: Culture clash, talent retention, and tech integration could delay results.

8. Risk Matrix

RiskSeverityProbabilityImpact on ThesisMitigant
CUDA network disruptionHIGHLOWOpen-source alternatives or new programming models erode CUDA's 18-year moat. Destroys NVIDIA's core competitive advantage.18 years of network depth. 4M+ developers. Massive switching costs. No credible alternative despite multiple attempts.
Custom ASICs gain major shareMEDIUMMEDIUMAmazon, Google, Microsoft custom chips reduce cloud market share. Could compress margins and slow growth.Custom ASICs serve first-party workloads only. Broader market needs general-purpose GPUs. Software network advantage holds.
AI capex cycle peaksHIGHLOW (near-term)Cloud providers cut AI infrastructure spending. Revenue growth decelerates. Multiple compresses.AI adoption still in early innings. Enterprise deployment barely started. Inference growth offsets training slowdown.
Groq integration failureMEDIUMMEDIUM$20B write-down risk. Culture clash between GPU and ASIC teams. Technology integration delays.Strong M&A track record (Mellanox). Jensen's hands-on management. Groq team motivated by NVIDIA resources.
Geopolitical tensions (Taiwan)HIGHLOWTSMC disruption halts GPU production. No alternative for leading-edge chips.TSMC building US fabs. Samsung diversification for some products. Diplomatic efforts ongoing.
Competition intensifies pricingMEDIUMMEDIUMAMD, Intel price competition compresses GPU margins. Cloud providers demand better pricing.Software network justifies premium. Customers pay for the platform, not just hardware. Gross margins are structurally higher.
End of Moore's Law accelerationMEDIUMMEDIUMPhysics limits slow transistor scaling. Next-gen GPUs deliver diminishing improvements.Investing in chiplet designs, advanced packaging, software optimization. Groq's SRAM approach sidesteps some limits.

9. Investment Framework

Bull Case

  • AI infrastructure spending is a multi-decade cycle. Not a bubble. Every enterprise will need AI compute.
  • The CUDA moat is deepening. 4M+ developers and growing. No competitor has cracked the network problem.
  • Groq creates an unassailable training + inference platform. First-mover in the inference era.
  • Sovereign AI (government infrastructure) creates $100B+ in new demand.
  • Robotics and autonomous vehicles represent massive TAM expansion beyond data center.

Bear Case

  • AI spending is a capex bubble. Cloud providers are over-building.
  • Custom ASICs from hyperscalers erode market share. The software moat is overrated.
  • $20B Groq deal is an expensive bet on unproven technology. Integration risks are real.
  • Taiwan/TSMC creates existential supply chain vulnerability.
  • At $4.5T market cap, future growth is priced in. Any miss causes severe compression.

Positioning Strategy

Scenario 1: New Position

Build in 3 tranches over 6 months. Start at 2% allocation. Add on 15%+ pullbacks. Target 5% weight.

Scenario 2: Existing Position

Hold core. Trim if it exceeds 10%. Sell covered calls on 20% for income. Add on 20%+ dips.

Scenario 3: Overweight

Reduce to 8% max. Take profits on highest-cost lots. Keep core for compounding. Hedge with put spreads.

Green Flags (Hold / Add)

  • Data center revenue grows 40%+ YoY
  • Groq integration hits milestones on schedule
  • Gross margins hold above 70%
  • CUDA developer network keeps expanding
  • New sovereign AI deals announced

Red Flags (Reduce / Exit)

  • Data center revenue growth falls below 20% YoY
  • A major cloud provider publicly shifts away from NVIDIA
  • Gross margins decline below 65% for two quarters
  • Key Groq engineers leave or milestones slip
  • Jensen announces retirement or succession