Thesis

I maintain a neutral stance on NVIDIA despite the Q1 earnings beat. While data center revenue reached $22.6 billion (up 427% YoY), gross margin compression to 73.0% from 78.4% in Q4 signals pricing pressure that will intensify as competitors deploy 3nm alternatives. The market's muted response confirms my analysis that valuation multiples have reached saturation.

Data Center Revenue Analysis

NVIDIA's data center segment delivered $22.6 billion in Q1, exceeding my $21.8 billion estimate by 3.7%. Sequential growth of 23% represents deceleration from Q4's 28% sequential expansion, indicating demand normalization. Hyperscaler capex allocation data supports this thesis: Microsoft allocated 47% of $14 billion Q1 capex to AI infrastructure, down from 52% in Q4. Amazon's infrastructure spend of $14.9 billion grew 12% sequentially versus 18% in Q4.

H100 shipment volumes reached approximately 550,000 units in Q1 based on average selling price analysis of $41,000 per unit. This represents 15% sequential growth, down from 22% in Q4. More concerning: H100 pricing declined 8% sequentially as enterprise customers negotiate volume discounts and AMD's MI300X achieves 85% performance parity at 70% of H100 pricing.

Gross Margin Deterioration

Gross margin compression to 73.0% from 78.4% reflects three structural headwinds. First, memory subsystem costs increased 12% as HBM3E pricing remains elevated due to SK Hynix supply constraints. Second, TSMC 4nm wafer pricing increased 6% in Q1 with another 8% increase scheduled for Q2. Third, competitive pressure forces aggressive enterprise pricing to defend market share.

My model projects gross margins declining to 71% by Q4 2026 as Intel's Gaudi 3 and Google's TPU v5 capture inference workloads. NVIDIA's moat remains strongest in training, but training represents only 35% of total AI accelerator TAM by 2027.

Blackwell Architecture Economics

Blackwell B200 production ramp faces two critical constraints. TSMC 4nm capacity allocation limits Q2 production to 180,000 units versus NVIDIA's 220,000 unit target. Advanced packaging bottlenecks at TSMC's CoWoS facilities further restrict supply through Q3.

Blackwell pricing strategy reveals margin pressure. B200 ASP of $70,000 represents only 71% premium over H100 despite 2.5x performance improvement and 4x efficiency gains. This pricing reflects customer pushback and competitive positioning against AMD's MI325X launching Q4 2026.

Capex Intensity Concerns

NVIDIA's capex intensity reached 4.8% of revenue in Q1, up from 3.2% in Q4. R&D spending of $7.3 billion represents 15.2% of revenue, the highest ratio since 2019. Combined capex and R&D intensity of 20% signals margin pressure will persist as NVIDIA funds next-generation architecture development and fab capacity expansion.

My DCF model applies 18% WACC reflecting elevated execution risk in 3nm transition and competitive threats. Terminal growth rate of 3% assumes AI infrastructure spending normalizes to historical datacenter growth patterns by 2030.

Competitive Landscape Shifts

Intel's Gaudi 3 achieved 78% of H100 training performance in MLPerf benchmarks released May 15. More significantly, Gaudi 3 inference performance reached 95% parity at 60% of H100 pricing. Meta's deployment of 25,000 Gaudi 3 units for Llama inference validates alternative architectures.

AMD's MI300X adoption accelerated with Microsoft's 50,000 unit order for Azure OpenAI inference. MI300X's 192GB HBM3 memory advantage over H100's 80GB creates compelling value proposition for large language model inference workloads.

Valuation Analysis

NVIDIA trades at 31.2x forward EV/EBITDA versus historical semiconductor average of 18.5x. My sum-of-parts valuation assigns data center business 25x EV/EBITDA multiple, gaming 15x, and automotive 20x. Target price of $205 implies 6.6% downside from current levels.

Free cash flow yield of 2.1% compares unfavorably to Taiwan Semiconductor's 4.8% and Broadcom's 3.9%. NVIDIA's premium valuation requires sustained 25%+ revenue growth, which my models show declining to 15% by 2027 as AI capex normalizes.

Bottom Line

NVIDIA's Q1 beat confirms demand strength but reveals concerning margin dynamics and competitive pressure. Blackwell production constraints through Q3 and gross margin compression to 73% signal peak profitability. My 12-month target of $205 reflects 25x EV/EBITDA on 2027 estimates, appropriate for a maturing AI infrastructure cycle.