#InteropTokyo2026 2日目 アイランドシックスメンバー一同、 ブース【7S06】で皆さまをお迎えしています！ 今回展示中の PacketLight「PL-2000ADS/GA」は、 🏅 Best of Show Award 2026APN部門ファイナリスト に選出されました。 #Interop26 #PacketLight #APN #ShowNet #OpenROADM

$NVDA NVIDIA Q2 FY2026 Earnings Call - Comprehensive Networking Analysis Executive Summary: Networking as Core Growth Driver NVIDIA’s networking business delivered record revenue of $7.3 billion in Q2, representing 46% sequential and 98% year-over-year growth. This explosive growth reflects networking’s critical role in AI infrastructure, with the company now offering three distinct networking technologies: NVLink for scale-up, InfiniBand/Spectrum-X for scale-out, and the newly announced Spectrum-XGS for scale-across. Management emphasized that choosing the right networking can improve AI factory efficiency by tens of percent, effectively making the networking investment “free” given the $50 billion cost of a gigawatt data center. 1. OVERALL NETWORKING BUSINESS PERFORMANCE 1.1 Financial Metrics Q2 Revenue: $7.3 billion (record quarter) Growth Rates: 46% sequential, 98% year-on-year Revenue Mix Contributing to Growth: Spectrum-X Ethernet InfiniBand NVLink 1.2 Strategic Positioning Colette Kress emphasized: “Escalating demands of AI compute clusters necessitate high-efficiency and low-latency networking… with strong demand across Spectrum-X Ethernet, InfiniBand, and NVLink.” 1.3 Historical Context Jensen referenced the Mellanox acquisition: “It’s the reason why NVIDIA dedicates so much in networking. That’s the reason why we purchased Mellanox five and a half years ago.” 1. THREE-TIER NETWORKING ARCHITECTURE 2.1 Jensen’s Framework Jensen Huang articulated NVIDIA’s comprehensive networking strategy: “We now offer three networking technologies. One is for scale-up, one is for scale-out, and one for scale-across.” 2.2 The Economics of Networking Critical insight from Jensen: “Choosing the right networking, the performance, the throughput improvement going from 65% to 85% or 90%, that kind of step-up because of your networking capability effectively makes networking free… the ability to improve the efficiency of that factory by tens of percents results in $10 billion, $20 billion worth of effective benefit.” 1. NVLINK - SCALE-UP NETWORKING 3.1 Revolutionary Architecture Evolution Previous Generation: NVLink 8: Node-scale computing where “each node is a computer” Current Generation: NVLink 72: Rack-scale computing where “each rack is a computer” Jensen on the achievement: “That disaggregation of NVLink 72 into a rack-scale system was extremely hard to do, but the results are extraordinary” 3.2 Performance Specifications Bandwidth: “14x the bandwidth of PCIe Gen 5” Impact: “We’re seeing orders of magnitude speed up and therefore energy-efficiency and therefore cost-effectiveness of token generation because of NVLink 72” 3.3 Strategic Importance for AI Evolution Jensen linked NVLink directly to the AI paradigm shift: “We built the Blackwell NVLink 72 system, a rack-scale computing system for this moment. We’ve been working on it for several years.” Critical for Reasoning Systems: “At a time when we have long-thinking, thinking models, agentic AI, reasoning systems, the NVLink basically amplifies the memory bandwidth, which is really critical for reasoning systems” “NVLink 72 is what made it possible for Blackwell to deliver such an extraordinary generational jump over Hopper’s NVLink 8” 3.4 NVLink Fusion Initiative “Positive reception to NVLink Fusion, which allows semi-custom AI infrastructure, has been widespread” Example deployment: “Japan’s upcoming FugakuNEXT will integrate Fujitsu’s CPUs with our architecture via NVLink Fusion” Multi-workload capability: “It will run a range of workloads including AI, supercomputing and quantum computing” 3.5 Future Generation Rubin platform includes “NVLink 144 scale-up switch” (already taped out to TSMC) Represents continued evolution of rack-scale architecture 3.6 Revenue Impact “Strong growth as customers deployed Grace Blackwell NVLink rack-scale systems” Contributing significantly to the $7.3B networking revenue 1. SPECTRUM-X ETHERNET - ENHANCED SCALE-OUT NETWORKING 4.1 Product Positioning & Differentiation Jensen clarified the unique nature of Spectrum-X: “Spectrum Ethernet is not off the shelf. It has a whole bunch of new technologies designed for low latency and low jitter and congestion control.” 4.2 Performance Metrics Annualized Revenue: “exceeding $10 billion” Growth: “double-digit sequential and year-over-year growth” Market Age: “only about a year and a half old” Market Success: Jensen declared it “a home run” 4.3 Technical Capabilities Designed specifically for “Ethernet AI workloads” Provides “the highest-throughput and lowest-latency network for Ethernet” Includes advanced congestion control mechanisms “Has the ability to come closer, much, much closer to InfiniBand than anything that’s out there” 4.4 Target Market For customers “who would like to use Ethernet, because their whole data center is built with Ethernet” 4.5 Spectrum-X in Rubin Platform The Rubin platform includes “Spectrum-X scale-out and scale-across switch” as one of its six chips 1. SPECTRUM-XGS - NEW SCALE-ACROSS TECHNOLOGY 5.1 Product Announcement “At Hot Chips, we introduced Spectrum-XGS Ethernet, a technology designed to unify disparate data centers into giga-scale AI superfactories” 5.2 Performance Promise “Projected to double GPU-to-GPU communication speed” 5.3 Early Adoption “CoreWeave is an initial adopter of the solution” 5.4 Strategic Purpose Jensen explained: “Spectrum-XGS, a giga-scale for connecting multiple data centers, multiple AI factories into a superfactory, a gigantic system” 5.5 Market Vision “You’re going to see that networking obviously is very important in AI factories” 1. INFINIBAND - HIGH-PERFORMANCE SCALE-OUT 6.1 Performance Leadership Jensen’s unequivocal statement: “InfiniBand, which is unquestionably the lowest latency, the lowest jitter, the best scale-out network” 6.2 Financial Performance “InfiniBand revenue nearly doubled sequentially” Driven by “adoption of XDR technology” 6.3 Technical Advantages XDR provides “double the bandwidth improvement over its predecessor” “Especially valuable for the model builders” Benchmark superiority: “If you were to benchmark an AI factory, the ones with InfiniBand are the best performance” 6.4 Target Market & Trade-offs Primary users: “For supercomputing, for the leading model makers, InfiniBand, Quantum InfiniBand is the unambiguous choice” Complexity consideration: “It does require more expertise in managing those networks” 6.5 Competitive Positioning Positioned as the premium option for those requiring absolute best performance, despite higher complexity 1. NETWORKING’S ROLE IN AI INFRASTRUCTURE 7.1 Critical Component of Blackwell Success Colette Kress noted: “Blackwell is still going to be the lion’s share of what we have in terms of data center. But keep in mind that helps both our compute side as well as it helps our networking side because we are selling those significant systems that are incorporating the NVLink” 7.2 Integrated System Sales Networking growth tied directly to rack-scale system deployments HGX systems driving both compute and networking revenue Seamless integration enabling customer adoption 7.3 Supply Chain Integration Six chips required for complete Rubin platform Networking components integral to platform, not add-ons Manufacturing coordination across multiple chip types 1. COMPETITIVE DYNAMICS & MARKET POSITIONING 8.1 Network Efficiency as Competitive Advantage Jensen’s framework on networking ROI: “The AI factory, a gigawatt as I mentioned before, could be $50 billion. And so the ability to improve the efficiency of that factory by tens of percents is – results in $10 billion, $20 billion worth of effective benefit.” 8.2 Performance Benchmarking Throughput improvements from 65% to 85-90% with right networking choice Direct revenue impact in power-limited environments Efficiency gains “effectively makes networking free” 8.3 Full-Stack Integration Advantage Jensen emphasized the complexity: “In order to build Blackwell, the platform; and Rubin, the platform, we had to build CPUs that connect fast memory… to the GPU; to a SuperNIC, to a scale-up switch, we call NVLink, completely revolutionary, we’re in our fifth generation now; to a scale-out switch, whether it’s Quantum or Spectrum-X Ethernet; to now scale-across switches” 1. CUSTOMER ADOPTION & USE CASES 9.1 Cloud Service Providers Major CSPs deploying all three networking tiers CoreWeave as early adopter of Spectrum-XGS Seeing 10x inference improvements with proper networking 9.2 AI Model Builders OpenAI, Meta, Mistral using GB200 NVL72 systems Leveraging NVLink 72 for training and inference InfiniBand preferred for leading model makers 9.3 Quantum & Supercomputing Centers FugakuNEXT (Japan) - NVLink Fusion integration Julich, AIST, NNF, NERSC running on CUDA-Q platform 300 ecosystem partners supporting quantum initiatives 9.4 Enterprise Deployments Growing adoption of Spectrum-X for Ethernet-based infrastructure RTX PRO servers integrating networking for standard IT environments 1. FUTURE ROADMAP & TECHNOLOGY EVOLUTION 10.1 Near-term Developments Rubin platform with enhanced networking (volume production next year) NVLink 144 representing next evolution Continued Spectrum-X market expansion 10.2 Long-term Vision Jensen’s projection: “Soon we’ll be building millions of GPU – millions of Rubin GPU platforms powering multi-gigawatt, multi-site AI superfactories” 10.3 Technology Integration Silicon Photonics processor in Rubin (taped out) Convergence of optical and electrical networking Scale-across becoming critical for superfactory architecture 1. KEY STRATEGIC INSIGHTS 11.1 Networking as Revenue Enabler Direct quote establishing the value proposition: “When we increase the perf per watt, the token generation per amount of usage of energy, we are effectively driving the revenues of our customers” 11.2 Annual Innovation Cadence Continuous improvement across all three networking tiers Coordinated development with compute platforms Maintaining technology leadership through rapid iteration 11.3 Market Education Need Jensen spent considerable time explaining networking importance, suggesting market still underappreciates networking’s role in AI infrastructure 11.4 Pricing Power Record revenue and growth rates indicate strong pricing Performance advantages justify premium pricing Integrated solutions prevent commoditization 1. ANALYST Q&A INSIGHTS ON NETWORKING 12.1 Aaron Rakers (Wells Fargo) Question Question: “I want to go back to the Spectrum-XGS announcement this week, and thinking about the Ethernet product now pushing over $10 billion of annualized revenue, just what is the opportunity set that you see for Spectrum-XGS?” Jensen’s Comprehensive Response: Detailed explanation of three-tier architecture Emphasized each tier serves distinct purpose Highlighted massive efficiency gains from proper networking Positioned Spectrum-XGS as essential for future superfactories 12.2 Stacy Rasgon (Bernstein) Question Question about Q3 growth allocation across products Colette’s Response: “Blackwell is still going to be the lion’s share of what we have in terms of data center. But keep in mind that helps both our compute side as well as it helps our networking side because we are selling those significant systems that are incorporating the NVLink” This confirms networking revenue is tightly coupled with compute platform sales. 1. TECHNICAL SPECIFICATIONS SUMMARY 13.1 Bandwidth Comparisons NVLink: 14x PCIe Gen 5 bandwidth InfiniBand XDR: 2x bandwidth over previous generation Spectrum-XGS: Expected to 2x GPU-to-GPU communication speed 13.2 Efficiency Metrics Network efficiency improvements: 65% → 85-90% Token generation improvement with NVLink 72: “orders of magnitude” ROI on networking: Can generate $10-20B benefit on $50B investment 13.3 Scale Metrics NVLink 8 → NVLink 72 → NVLink 144 (roadmap) Spectrum-X: $10B annualized revenue in ~1.5 years InfiniBand: Revenue nearly doubled sequentially 1. COMPETITIVE MOAT ANALYSIS 14.1 Integration Complexity The full quote revealing the moat: “The complications, the complexity of everything that we do is really quite extraordinary. It’s just done in a really, really extreme scale now.” 14.2 Multi-Generation Advantage Fifth generation NVLink Years of development on rack-scale architecture Continuous innovation across all networking tiers 14.3 Ecosystem Lock-in Software optimization for specific networking architectures Customer investments in training for InfiniBand management Platform-level integration preventing component substitution 1. FINANCIAL IMPACT & PROJECTIONS 15.1 Current Contribution $7.3B quarterly revenue (approximately 15.6% of total revenue) Growing faster than company average (98% YoY vs 56% for data center) Margin accretive given software/IP content 15.2 Future Growth Drivers Gigawatt superfactory buildouts requiring scale-across Continued Blackwell/Rubin deployments driving NVLink Enterprise Ethernet transitions to Spectrum-X Sovereign AI initiatives requiring complete networking stack 15.3 Investment Thesis Jensen’s articulation of networking value: “Choosing the right networking, you’re basically paying – you’ll get a return on it like you can’t believe” This positions networking not as a cost but as a revenue multiplier for AI infrastructure deployments.