Skip to content
    NVDA
    Earnings call· Jul 2025(Q2 FY26)

    NVIDIA CORP NVDA

    Aug 27, 2025 Source

    Executive summary

    NVIDIA Q2 FY26 — Record Revenue Driven by Blackwell and AI Infrastructure Build-Out

    NVIDIA delivered record Q2 FY26 results, driven by strong demand for its Blackwell platform and full-stack AI solutions across cloud, enterprise, and sovereign AI initiatives. The company is accelerating investments to capitalize on the immense AI infrastructure opportunity, which is projected to reach $3 trillion to $4 trillion by the end of the decade. Geopolitical issues continue to impact H20 shipments to China, leading to its exclusion from the Q3 outlook, though management is advocating for broader market access for its advanced architectures.

    Highlights

    5

    • Total revenue reached a record $46.7 billion, exceeding outlook and growing sequentially across all market platforms.

    • Data Center revenue grew 56% year-over-year, with the Blackwell platform growing 17% sequentially and generating tens of billions in revenue.

    • Networking revenue reached a record $7.3 billion, increasing 46% sequentially and 98% year-on-year, with Spectrum-X Ethernet annualized revenue exceeding $10 billion.

    • Gaming revenue hit a record $4.3 billion, up 14% sequentially and 49% year-on-year, driven by Blackwell GeForce GPUs.

    • Returned $10 billion to shareholders in Q2 through share repurchases and cash dividends, with a new $60 billion share repurchase authorization approved.

    Concerns

    4

    • Data Center revenue declined sequentially due to a $4 billion drop in H20 revenue.

    • China data center revenue declined sequentially to a low single-digit percentage of total data center revenue, with no H20 shipments included in Q3 outlook due to geopolitical issues.

    • Inventory increased sequentially from $11 billion to $15 billion in Q2 to support the ramp of Blackwell and Blackwell Ultra.

    • Full-year operating expenses growth expectation raised to high 30s range year-over-year, up from prior mid-30s expectation.

    Guidance & targets

    10
    CategoryTargetConfidence
    Total Revenue
    $54 billion, plus or minus 2%
    high materiality
    High
    GAAP Gross Margin
    73.3%
    medium materiality
    High
    Non-GAAP Gross Margin
    73.5%
    high materiality
    High
    Non-GAAP Gross Margin
    mid-70s
    high materiality
    Medium
    GAAP Operating Expenses
    $5.9 billion
    medium materiality
    High
    Non-GAAP Operating Expenses
    $4.2 billion
    medium materiality
    High
    Full-year Operating Expenses Growth
    high 30s range year-over-year
    medium materiality
    High
    Other Income and Expenses (GAAP and Non-GAAP)
    $500 million income
    low materiality
    High
    Tax Rates (GAAP and Non-GAAP)
    16.5%, plus or minus 1%
    low materiality
    High
    H20 Revenue to China
    $2 billion to $5 billion
    high materiality
    Medium

    Segment performance

    5
    SegmentRevenueYoYQoQMargin
    Data Center
    Revenue grew significantly year-over-year but declined sequentially due to a $4 billion drop in H20 revenue. Blackwell platform saw strong sequential growth. Sovereign AI revenue is expected to more than double this year. China revenue declined sequentially.
    Blackwell platform sequential growth: 17%H20 revenue decline: $4 billionH20 sales outside China: $650 millionSovereign AI revenue (FY26): >$20 billionChina revenue share: low single-digit percentage
    		—56%——
    Networking
    Delivered record revenue with strong sequential and year-on-year growth across Spectrum-X Ethernet, InfiniBand, and NVLink. Spectrum-X Ethernet is a significant contributor.
    Spectrum-X Ethernet annualized revenue: >$10 billion
    		$7.3 billion98%46%—
    Gaming
    Record revenue driven by the ramp of Blackwell GeForce GPUs and increased supply availability. GeForce RTX 5060 desktop GPU shipped.
    		$4.3 billion49%14%—
    Professional Visualization
    Growth driven by adoption of high-end RTX Workstation GPUs and AI-powered workloads.
    		$601 million32%——
    Automotive
    Revenue includes only in-car compute, primarily driven by self-driving solutions. Began shipments of NVIDIA Thor SoC.
    		$586 million69%——

    Operational metrics

    20
    Non-GAAP Gross Margin
    72.7%
    Q2 FY26

    Exceeded outlook, even when excluding the benefit from releasing previously reserved H20 inventory.

    Inventory
    $15 billionup from $11 billion sequentially
    Q2 FY26 end

    Increased to support the ramp of Blackwell and Blackwell Ultra.

    Capital Returned to Shareholders
    $10 billion
    Q2 FY26

    Returned to shareholders in Q2 FY26.

    Share Repurchase Authorization
    $60 billion
    new authorization

    Board of Directors approved a new authorization, adding to the remaining prior authorization at the end of Q2.

    AI Infrastructure Spend
    $3 trillion to $4 trillion
    by end of decade

    Projected total AI infrastructure spend by the end of the decade.

    Hyperscaler CapEx
    $600 billiondoubled in 2 years
    CY25

    Annual CapEx for the top 4 CSPs, nearly doubling in 2 years.

    AI-native Start-up Funding
    $180 billionup from $100 billion last year
    this year

    Total funding for AI-native start-ups this year.

    AI-native Start-up Revenue
    $20 billionup from $2 billion last year
    this year

    Revenue generated by top AI-native start-ups this year.

    GB300 Rack Production
    1,000 racks per week
    current run rate

    Current production run rate, expected to accelerate further throughout Q3.

    AI Inference Performance (GB300 vs H100)
    10xmore inference performance
    current

    CoreWeave seeing 10x more inference performance on reasoning models compared to H100 with GB300 infrastructure.

    Energy Efficiency (GB300 NVL72 vs Hopper)
    50xincrease
    current

    Delivered by new NVFP4 4-bit precision and NVLink 72 on the GB300 platform.

    Blackwell Performance Improvement
    2xmore performance
    since launch

    Improved by NVIDIA software innovation and developer ecosystem contributions.

    LLM Pretraining Speed (GB300 NVFP4 vs H100 FP8)
    7xfaster training
    current

    Using NVFP4 computations on the GB300 compared to H100 using FP8.

    Sovereign AI Revenue
    $20 billionmore than double last year
    FY26

    Expected revenue from Sovereign AI initiatives this fiscal year.

    European AI Compute Infrastructure Increase
    tenfold
    future

    European Union plans to invest EUR 20 billion to establish 20 AI factories to increase its AI compute infrastructure by tenfold.

    Isambard-AI Supercomputer Performance
    21 exaflops
    current

    Powered by NVIDIA, unveiled as the U.K.'s most powerful AI system.

    Singapore Revenue Share
    22%
    Q2 FY26

    Customers have centralized their invoicing in Singapore, with the vast majority of billed revenue for U.S.-based customers.

    GeForce NOW Catalog Size
    4,500 titlesdoubled
    current

    Largest library of any cloud gaming service.

    Robotics Developers
    2 million
    current

    Number of developers adopting NVIDIA's robotics full stack platform.

    Robotics Hardware/Software/Sensor Partners
    1,000+
    current

    Number of partners taking NVIDIA's robotics platform to market.

    Industry KPIs

    4
    MetricValueDetails
    Lead times12 monthsmonths
    Ai data center revenue—
    Inventory channel inventory$15 billionUSD
    Node platform ramp scheduleRubin platform

    Product announcements

    5
    ProductTypeDetails
    GeForce RTX 5060 desktop GPUlaunch
    Blackwell on GeForce NOWexpansion
    Spectrum-XGS Ethernetlaunch
    Jetson Thorlaunch
    NVIDIA Thor SoClaunch

    Deals & partnerships

    4
    CoreWeaveInitial adopter of Spectrum-XGS Ethernet solution

    CoreWeave is an initial adopter of the Spectrum-XGS Ethernet solution and prepares to bring their GB300 instance to market, already seeing 10x more inference performance on reasoning models compared to H100.

    OpenAIOptimization of open source GPT models for RTX GPUs; top contributor to OpenAI models, data, and software.

    NVIDIA partnered with OpenAI to optimize their open source GPT models for RTX GPUs. NVIDIA is also a top contributor to OpenAI models, data, and software.

    SiemensMajor expansion of partnership to enable AI automatic factories using Omniverse.

    Announced a major expansion of the partnership to enable AI automatic factories, leveraging NVIDIA Omniverse with Cosmos.

    FujitsuIntegration of Fujitsu's CPUs with NVIDIA's architecture via NVLink Fusion for FugakuNEXT supercomputer.

    Japan's upcoming FugakuNEXT will integrate Fujitsu's CPUs with NVIDIA's architecture via NVLink Fusion.

    Risks & headwinds

    3
    Geopolitical issues impacting H20 shipments to ChinaQ3 FY26

    $2 billion to $5 billion in H20 revenue not included in Q3 outlook

    Mitigation: Working through geopolitical issues; advocating for U.S. government approval of Blackwell for China; seeking licenses for H20 sales.

    Increased operating expensesFY26

    Full-year operating expenses expected to grow in high 30s range year-over-year, up from prior mid-30s expectation

    Mitigation: Accelerating investments in the business to address the magnitude of growth opportunities.

    Power limitations in data centersOngoing

    Data centers are power limited

    Mitigation: Focus on maximizing performance per watt (e.g., GB300 NVL72's 10x improvement in token per watt energy efficiency) to drive revenue generation for customers.

    What to watch in Q3 FY26

    5

    H20 Shipments to China

    Q3 FY26
    CurrentNone included in Q3 outlook
    Target$2 billion to $5 billion (potential)

    Why it matters

    Resolution of geopolitical issues and license approvals could significantly impact Q3 revenue and future China market access.

    If geopolitical issues reside, we should ship $2 billion to $5 billion in H20 revenue in Q3. And if we had more orders, we can bill more.

    Q&A highlights

    7

    What is NVIDIA's vision for growth into 2026 and beyond, particularly regarding reasoning agentic AI and the network/data center split?

    Jensen Huang highlighted the evolution of reasoning agentic AI, which requires 100x to 1,000x more computation than traditional chatbots, driving tremendous growth. He emphasized the Blackwell NVLink 72 rack-scale system's role in this, offering orders of magnitude speed-up and energy efficiency. He reiterated the $3 trillion to $4 trillion AI infrastructure opportunity over the next 5 years, with NVIDIA contributing significantly to the AI infrastructure build-out.

    “The amount of computation necessary for 1 shot versus reasoning agentic AI models could be 100x, 1,000x and potentially even more as the amount of research and basically reading and comprehension that it goes off to do.”

    asked by Christopher Muse · answered by Jen-Hsun Huang

    2 min read6 chapters

    Detailed Narrative

    01

    AI Infrastructure Market Opportunity

    NVIDIA projects the AI infrastructure market to reach $3 trillion to $4 trillion by the end of the decade, driven by the evolution of reasoning agentic AI, global sovereign AI build-outs, enterprise adoption, and physical AI/robotics. Hyperscaler CapEx has doubled in two years to $600 billion annually, indicating the early stages of this massive build-out. NVIDIA's full-stack AI solutions, including the Blackwell and upcoming Rubin platforms, are positioned to capture a significant portion of this investment, with NVIDIA's components representing approximately $35 billion of a $50 billion gigawatt AI factory.

    02

    Blackwell Platform Ramp and Performance

    The Blackwell platform, including GB200 and GB300, is experiencing widespread adoption and is ramping at full speed, with production of GB300 racks at approximately 1,000 per week, expected to accelerate further in Q3. The GB300 NVL72 AI factories offer a 10x improvement in token per watt energy efficiency compared to Hopper, redefining inference economics. Software innovations like CUDA, TensorRT-LLM, and Dynamo have improved Blackwell's performance by over 2x since launch, and new NVFP4 computations achieve 7x faster training than H100.

    03

    Networking Innovation and Growth

    Networking revenue reached a record $7.3 billion in Q2 FY26, driven by strong demand across Spectrum-X Ethernet, InfiniBand, and NVLink. Spectrum-X Ethernet, designed for AI workloads, delivered double-digit sequential and year-over-year growth, with annualized revenue exceeding $10 billion. The introduction of Spectrum-XGS Ethernet aims to unify disparate data centers into giga-scale AI super factories, further expanding NVIDIA's networking capabilities for scale-across applications.

    04

    China Market Dynamics and H20 Shipments

    China's data center revenue declined sequentially to a low single-digit percentage of the total. While the U.S. government began reviewing licenses for H20 sales to China, and some licenses were received, no H20 shipments based on these licenses have occurred. NVIDIA's Q3 outlook excludes H20 shipments to China, though $2 billion to $5 billion in revenue is possible if geopolitical issues resolve. Management continues to advocate for approval of Blackwell for the China market, which represents a significant opportunity estimated at $50 billion this year if fully addressable.

    05

    Rubin Platform and Annual Cadence

    The Rubin platform, NVIDIA's third-generation NVLink rack-scale AI supercomputer, is on schedule for volume production next year, maintaining an annual product cadence. Rubin will feature new chips, including the Vera CPU, Rubin GPU, CX9 SuperNIC, NVLink 144 switch, Spectrum-X switch, and silicon photonics processor, all currently in fab. This annual rhythm aims to accelerate cost reduction and maximize revenue generation for customers by continuously improving performance per watt and AI capabilities.

    06

    Robotics and Physical AI Expansion

    NVIDIA's robotics platform is gaining significant traction, with the new Jetson Thor computing platform now available, offering an order of magnitude greater AI performance and energy efficiency for edge AI. The Omniverse with Cosmos platform serves as a physical AI digital twin platform for robot development, with expanded partnerships like Siemens for AI automatic factories. The company sees robotics and physical AI as a significant long-term demand driver for its data center platform.

    AI-generated summary of the company’s earnings call. Not investment advice.