Nvidia's Q2 2026: Contradictions Emerge on Blackwell Production, AI Market Share, and Margins

The above is the analysis of the conflicting points in this earnings call

Date of Call: August 27, 2025

Financials Results

• Revenue: $46.7B; exceeded outlook and grew sequentially across all market platforms (total YOY growth not disclosed)
• Gross Margin: GAAP 72.4%; non-GAAP 72.7% including ~40 bps benefit from H20 reserve release (ex-benefit non-GAAP 72.3%; above outlook)

Guidance:

• Q3 revenue expected at $54B ±2%; excludes H20 shipments to China.
• Potential incremental H20 revenue of $2–$5B in Q3 if geopolitical/licensing issues resolve.
• GAAP gross margin: 73.3% ±50 bps; non-GAAP gross margin: 73.5% ±50 bps.
• Expect to exit year with non-GAAP gross margin in the mid-70s.
• GAAP opex ≈ $5.9B; non-GAAP opex ≈ $4.2B for Q3.
• Full-year opex growth to be high-30s % YoY (up from prior mid-30s outlook).
• Other income (GAAP/non-GAAP) ≈ $0.5B, excluding securities gains/losses.
• Tax rate (GAAP/non-GAAP): ~16.5% ±1%.

Business Commentary:

• Record Revenue and Data Center Growth:
• NVIDIA Corp reported record revenue of $46.7 billion for Q2, exceeding their outlook.

• Growth was driven by strong performance in data center revenue, which grew 56% year-over-year, and the Blackwell platform reaching record levels with a sequential increase of 17%.

• Inference Performance Enhancement and AI Infrastructure:

  The transition to the Gb 300 has been seamless, achieving a 10x improvement in token per watt energy efficiency compared to the Hopper architecture.

  The focus on improving inference performance is driven by the increasing demand for AI inference capabilities, particularly from reasoning and agentic AI models.

• Networking Expansion:

  NVIDIA's networking business reported record revenue of $7.3 billion, up 46% sequentially and 98% year-on-year.

  The growth is attributed to strong demand for Spectrum X Ethernet, InfiniBand, and NVLink, driven by the high efficiency and low latency networking requirements for AI compute clusters.

• Geopolitical Challenges and China Market Potential:

  The licenses for H20 sales to China customers have been reviewed, with anticipation of $2,000,000,000 to $5,000,000,000 in H20 revenue in Q3 if geopolitical issues are resolved.

  The China market represents a significant opportunity, with estimated $50,000,000,000 in opportunity for NVIDIA this year, growing at 50% per year.

Sentiment Analysis:

• Positive: Management reported a record quarter with total revenue above outlook; data center up 56% YOY and networking at a record $7.3B (+98% YOY). Q3 revenue guide is $54B (+~$7B seq.). Blackwell Ultra ramping at full speed with extraordinary demand; Rubin already in fab. Sovereign AI revenue expected to exceed $20B this year (more than double YoY).

Q&A:

• Question from C.J. Muse (Cantor Fitzgerald): With 12-month lead times and Rubin ramping in 2H, how do you view growth into 2026, and how do networking and data center contribute?
• Response: Agentic AI drives orders-of-magnitude more compute; NVLink 72 rack-scale systems and annual cadences (Blackwell→Rubin) position NVIDIA to scale into a $3–$4T AI infrastructure buildout, with networking integral to performance and efficiency.

• Question from Vivek Arya (Bank of America Securities): What must happen for the $2–$5B China H20 shipments, and could ASICs displace NVIDIA GPUs?
• Response: H20 shipments hinge on licenses and geopolitics; supply is ready. ASICs are limited—NVIDIA’s full-stack, ubiquitous platform delivers superior perf/W and ROI across rapidly changing models, keeping GPUs the preferred choice.

• Question from Ben Reitzes (Melius): Clarify the $3–$4T AI infrastructure by decade-end and NVIDIA’s share; any bottlenecks like power?
• Response: $3–$4T is sensible given ~$600B annual hyperscaler CapEx; NVIDIA represents ~35% of a ~$50–$60B per gigawatt AI factory. Power is the key limiter; NVIDIA focuses on perf/W to maximize factory revenue.

• Question from Joe Moore (Morgan Stanley): Long-term China prospects and importance of licensing Blackwell there?
• Response: China is a ~$50B opportunity this year and could grow ~50% annually; with many AI researchers and strong open-source models, NVIDIA is advocating to bring Blackwell to China beyond current H20 licenses.

• Question from Aaron Rakers (Wells Fargo): What is the opportunity for Spectrum XGS within Ethernet networking?
• Response: NVIDIA offers NVLink (scale-up), InfiniBand and Spectrum X Ethernet (scale-out), and Spectrum XGS (scale-across) to link AI factories; networking efficiency yields outsized ROI, with Spectrum X already >$10B annualized.

• Question from Stacy Raskin (Bernstein Research): How should we apportion the ~$7B sequential Q3 growth across Blackwell, Hopper, and networking?
• Response: Blackwell will be the lion’s share of growth; Hopper (H100/H200) continues; networking also rises as NVLink-based systems ship.

• Question from Jim Schneider (Goldman Sachs): How should we view the Rubin transition and its performance step relative to Blackwell?
• Response: NVIDIA’s annual cadence boosts perf/W to drive customers’ revenue and margins; Rubin brings significant new ideas (details to come), with record ramps expected following Blackwell and Blackwell Ultra.

• Question from Timothy Arcuri (UBS): Visibility on next year—can data center growth match a ~50% AI market CAGR?
• Response: Visibility is strong with large customer forecasts, heavy startup funding, and sold-out capacity; growth at least in line with market is plausible amid broadening AI demand.