Technology and Innovation Community

 View Only

  • 1.  GS Equity Research update on Optical Networking

    Posted 2 days ago

    Recent update fro the team @ GS Research on Optical Networking.

    report

    Key summary:

    Networking is the next frontier in AI infrastructure, poised to enhance computing capability through seamless data exchange and low latency. While an investor concern is on networking configuration replacing another, we expect all configuration will enjoy strong growth. Specifically: 

    • Dollar content increase by 16x / 45x in Scale Out / Scale Up per computing unit from GB300 NVL72 (per computing unit means 72 GPUs per rack to reach NVL72) to Rubin Ultra NVL576 (per computing unit means 72 GPUs per rack, and 8 racks together to reach NVL576), with opportunities across pluggable optical modules, optical engines in CPO, copper cables, and PCB midplanes.
    • A 13x larger addressable market for optical modules / optical engines expanding from scale out (e.g. GB300 NVL72) to scale up (e.g. Rubin Ultra NVL576 level 2 scale up via CPO) per computing unit.
    • A 10x larger value market for pluggable optical modules in scale out per computing unit from GB300 NVL72 to Rubin Ultra NVL576, even with a 29% CPO penetration rate. The numbers of pluggable optical module (1.6T equivalent) per computing unit would increase from 216 units in GB300 NVL72 to 2.5k units in Rubin Ultra NVL576.

    #AI #innovationcommunity

    - Todor



    ------------------------------
    Todor Kostov
    Director
    ------------------------------


  • 2.  RE: GS Equity Research update on Optical Networking

    Posted 2 days ago

    Thanks Todor.  A useful acronym cheatsheet for those readers a bit lost with some of them in the report:

    Acronym Meaning Why It Matters
    AI Artificial Intelligence Drives explosive demand for compute + networking
    ASIC Application-Specific Integrated Circuit Custom AI chips used by hyperscalers
    ASP Average Selling Price Higher speeds (1.6T/3.2T) increase ASPs
    BoM Bill of Materials Key for margin analysis
    CPO Co-Packaged Optics Next-gen optical architecture integrating optics near switch/GPU
    NPO Near-Packaged Optics / Onboard Optics Intermediate step between pluggables and CPO
    SiPh Silicon Photonics Optical technology expected to gain major market share
    EML Electro-Absorption Modulated Laser Traditional optical laser technology
    CW Laser Continuous Wave Laser Common light source for silicon photonics
    VCSEL Vertical-Cavity Surface-Emitting Laser Short-range optical laser technology
    FAU Fiber Array Unit Connects optical engines to fiber
    ELS External Laser Source Supplies laser light to CPO systems
    DSP Digital Signal Processor Signal conditioning inside optical modules
    PIC Photonic Integrated Circuit Integrates optical functions onto chips
    EIC Electronic Integrated Circuit Traditional electronic chip circuitry
    GPU Graphics Processing Unit Main AI accelerator
    XPU General term for AI accelerators Includes GPUs, CPUs, NPUs, ASICs
    CPU Central Processing Unit General-purpose compute
    NPU Neural Processing Unit AI-focused processor
    NIC Network Interface Card Connects servers to network fabric
    NVL72 / NVL144 / NVL576 Nvidia rack-scale GPU architectures Defines cluster scale and networking complexity
    NVSwitch Nvidia high-speed GPU interconnect switch Enables GPU-to-GPU communication
    ToR Top-of-Rack Switch Primary networking switch inside racks
    EoR End-of-Row Switch Aggregates rack networking
    InfiniBand Nvidia low-latency networking protocol Dominant in AI training clusters
    Ethernet Standard networking protocol Increasingly used for AI networking
    AEC Active Electrical Cable Copper cable with retimers for longer distances
    ACC Active Copper Cable Copper cable with redrivers
    DAC Direct Attach Copper Simple low-cost short-distance copper cable
    AOC Active Optical Cable Optical cable with integrated transceivers
    MPO Multi-Fiber Push-On connector High-density fiber connector
    PCB Printed Circuit Board Used for electrical interconnects
    PCB Midplane Large internal PCB connection architecture New scale-up design in Rubin Ultra
    CCL Copper Clad Laminate Material used in advanced PCBs
    TAM Total Addressable Market Core valuation framework
    TCO Total Cost of Ownership Important for hyperscaler adoption decisions
    Tb/s Terabits per second Measures networking bandwidth
    Gb/s Gigabits per second Lower bandwidth measurement
    800G / 1.6T / 3.2T Optical networking speeds Core upgrade cycle driving AI networking demand
    Scale-Up Expanding compute within a rack/system Requires ultra-fast low-latency interconnects
    Scale-Out Connecting racks/clusters together Major driver of optical demand
    Hyperscaler Large cloud provider (AWS, Google, Meta, Microsoft) Biggest AI infrastructure spenders
    CSP Cloud Service Provider Another term for hyperscalers
    MRM Micro Ring Modulator Silicon photonics modulation technology
    MZM Mach-Zehnder Modulator Alternative optical modulation technology
    InP Indium Phosphide Critical semiconductor material for lasers
    GTC Nvidia GPU Technology Conference Major roadmap announcement event
    SerDes Serializer/Deserializer Converts parallel ↔ serial data for high-speed links
    DR / FR Data center reach optical standards Different transmission distance classes



    ------------------------------
    Carlos Salas
    Portfolio Manager & Freelance Investment Research Consultant
    ------------------------------

    Original Message


Related Content