NVIDIA: Data Center Revenue Trajectory Points to $180B TAM by FY2027

Core Thesis

I maintain NVIDIA's data center business will capture 65-70% of the $180 billion AI infrastructure TAM by FY2027, driven by Blackwell architecture's 5x inference efficiency gains and sustained hyperscaler deployment rates exceeding 40% quarterly growth. Current valuation at 24.8x forward earnings reflects temporary margin compression during the H100-to-B100 transition, creating tactical accumulation opportunity.

Data Center Revenue Analysis

NVIDIA's data center segment generated $47.5 billion in FY2024, representing 300% year-over-year growth. Q4 FY2024 data center revenue of $18.4 billion exceeded guidance by 12%, with sequential growth of 28%. I project Q1 FY2025 data center revenue of $21.2 billion, maintaining the 15-20% quarterly growth trajectory despite supply chain normalization.

Hyperscaler customers (Microsoft, Meta, Google, Amazon) comprised 74% of data center revenue in Q4, with enterprise and sovereign AI contributing the remaining 26%. Microsoft Azure's AI workload revenue grew 150% year-over-year, while Meta's Reality Labs capex allocation increased to $4.2 billion quarterly, directly benefiting NVIDIA's compute platform penetration.

Blackwell Architecture Economics

Blackwell B100 and B200 GPUs deliver 2.5x memory bandwidth (8TB/s vs 3.35TB/s for H100) and 5x inference performance per dollar compared to Hopper. Manufacturing on TSMC's 4NP process node reduces power consumption by 25% while increasing transistor density to 208 billion per chip.

Blackwell gross margins will compress to 71-73% in FY2025 from H100's 78-80% peak, reflecting higher CoWoS packaging costs and initial yield curves. However, I calculate total addressable margin dollars expanding 45% as Blackwell ASPs range from $35,000-$70,000 per unit versus H100's $25,000-$40,000 range.

Inference Market Penetration

Inference workloads represent 85% of AI compute demand, growing at 47% CAGR through 2027. NVIDIA's CUDA ecosystem maintains 92% market share in training, but inference competition from AMD's MI300X and Intel's Gaudi3 intensifies margin pressure.

Blackwell's FP4 precision support enables 16x higher throughput for large language model inference compared to training-optimized architectures. I estimate inference-specific revenue reaching $28 billion by FY2026, representing 38% of total data center segment.

Supply Chain and Manufacturing

TSMC allocated 60% of its advanced packaging capacity to NVIDIA through 2025, securing CoWoS supply for 2.5 million Blackwell units annually. HBM3E memory from SK Hynix and Samsung remains the primary bottleneck, with supply agreements covering 1.8 million units through Q2 FY2025.

I calculate NVIDIA's inventory turns improving to 4.2x in FY2025 from 3.8x in FY2024 as supply chain execution normalizes. Days sales outstanding decreased to 28 days in Q4 FY2024, reflecting improved customer payment terms and reduced channel inventory.

Competitive Dynamics

AMD's MI300X delivers competitive FP16 performance at 30% lower pricing, capturing 8% training market share in Q4 2024. However, CUDA's software moat remains intact with 4.8 million registered developers and 3,200 GPU-accelerated applications.

Intel's Gaudi3 targets inference workloads with 40% better price-performance than H100 for specific transformer models. I estimate Intel capturing 3-5% inference market share by 2025, primarily in cost-sensitive enterprise deployments.

Financial Projections

I project FY2025 revenue of $118 billion (85% data center, 8% gaming, 7% other segments) with non-GAAP EPS of $28.50. Free cash flow will reach $67 billion, supporting $15 billion in share repurchases and $2.8 billion in quarterly dividends.

Gross margins compress to 73.2% in FY2025 from 76.8% in FY2024, while operating margins remain above 60% due to operating leverage on the $7.2 billion quarterly revenue base.

Risk Assessment

China revenue exposure remains at 15-20% of total, vulnerable to export control escalation. Blackwell production ramp delays could compress Q2-Q3 FY2025 guidance, while hyperscaler capex normalization poses 2025-2026 growth deceleration risk.

Bottom Line

NVIDIA trades at 25.2x FY2026 earnings despite controlling 70% of a $180 billion TAM expanding at 35% annually. Blackwell's inference efficiency gains and sustained hyperscaler demand support 22% revenue CAGR through FY2027. Current weakness creates accumulation opportunity below $200 with 12-month target of $245.