SemiconductorsFEATURE

NVIDIA, Intel, and imec Chart Three Layers of Silicon Photonics for AI Data Centers

E
EffectStory 編輯部Editorial Team
Published · Updated
NVIDIA and Intel are moving optics from pluggable modules into co-packaged, chip-integrated designs to cut per-bit power in AI data-center networking. Intel's Optical Compute Interconnect reaches 4 Tb/s bidirectional at about 5 pJ/bit, versus roughly 15 pJ/bit for pluggable transceivers, while NVIDIA's Quantum-X Photonics switch delivers 144 ports of 800Gb/s InfiniBand with 3.5x power efficiency and 4x fewer lasers. imec's CMOS-based manufacturing and DWDM roadmap toward 400Gb/s underpin the cost and scaling path both vendors build on.

How does silicon photonics break the power bottleneck in AI data centers?

Intel's co-packaged silicon photonics solution consumes about 5 picojoules per bit, compared with roughly 15 pJ/bit for pluggable optical transceiver modulesCITE:E5. NVIDIA's co-packaged optics switches integrate photonics innovations with 4x fewer lasers, delivering 3.5x more power efficiency, 63x greater signal integrity, and 10x better network resiliency at scale, along with 1.3x faster deployment compared with traditional methods, according to NVIDIACITE:E1.

What performance breakthroughs define NVIDIA's Quantum-X Photonics switch design?

NVIDIA's Quantum-X Photonics switches provide 144 ports of 800Gb/s InfiniBand connectivity built on 200Gb/s SerDes technologyCITE:E2. The switches use a liquid-cooled design to efficiently cool the onboard silicon photonics, NVIDIA statesCITE:E2. NVIDIA announced the Quantum-X Photonics switch as part of its Spectrum-X co-packaged optics networking line on March 18, 2025CITE:E1.

How does Intel's Optical Compute Interconnect integrate optics on chip?

Intel's first Optical Compute Interconnect (OCI) chiplet supports up to 4 terabits per second of bidirectional data transfer and is compatible with PCIe Gen5CITE:E3. The fully integrated OCI chiplet combines a silicon photonics integrated circuit (PIC) — which includes on-chip lasers and optical amplifiers — with an electrical IC, Intel saysCITE:E4. Intel unveiled this chiplet on June 26, 2024CITE:E3.

Why does CMOS manufacturing give silicon photonics a cost advantage?

Integrated silicon photonics uses existing CMOS infrastructure to manufacture optical circuits, imec states, realizing cost benefits through economies of scaleCITE:E6.

What is the roadmap for scaling bandwidth with DWDM silicon photonics?

imec's dense wavelength-division multiplexing (DWDM) silicon photonics technology offers a path to scale optical links toward 400Gb/s, using eight channels of 50Gb/s each (8x50Gb/s)CITE:E7.

Key figures at a glance

SourcePower EfficiencyBandwidthOther Metrics
NVIDIA Quantum-X Photonics / Spectrum-X co-packaged optics3.5x power efficiency, 4x fewer lasersCITE:E1144 ports x 800Gb/s InfiniBand, 200Gb/s SerDesCITE:E263x signal integrity, 10x network resiliency, 1.3x faster deploymentCITE:E1
Intel Optical Compute Interconnect (OCI) chiplet5 pJ/bit vs. 15 pJ/bit for pluggable transceiversCITE:E54 Tb/s bidirectional, PCIe Gen5-compatibleCITE:E3PIC integrates on-chip lasers and optical amplifiers with an electrical ICCITE:E4
imec DWDM silicon photonicsCost benefits via CMOS economies of scaleCITE:E6Path to 400Gb/s (8x50Gb/s)CITE:E7

What this means

NVIDIA and Intel are pursuing the same underlying problem — per-bit energy consumption in AI data-center interconnects — from two different layers. Intel's 5 pJ/bit figure is measured against pluggable transceivers at about 15 pJ/bitCITE:E5, while NVIDIA's 3.5x efficiency and 4x laser-count reduction are measured against unspecified "traditional methods"CITE:E1; the two claims use different baselines but point in the same direction: co-packaging optics with compute silicon rather than routing through separate optical modules. imec's CMOS-based manufacturing processCITE:E6 and its DWDM roadmap toward 400Gb/sCITE:E7 describe the process economics and bandwidth-scaling path that underlie vendor-specific products such as NVIDIA's 144-port Quantum-X Photonics switchCITE:E2 and Intel's 4 Tb/s OCI chipletCITE:E3CITE:E4.

📊 Evidence

FAQ

How does silicon photonics break the power bottleneck in AI data centers?

Intel's co-packaged silicon photonics solution consumes about 5 picojoules per bit, compared with roughly 15 pJ/bit for pluggable optical transceiver modulesCIT…

What performance breakthroughs define NVIDIA's Quantum-X Photonics switch design?

NVIDIA's Quantum-X Photonics switches provide 144 ports of 800Gb/s InfiniBand connectivity built on 200Gb/s SerDes technologyCITE:E2.

How does Intel's Optical Compute Interconnect integrate optics on chip?

Intel's first Optical Compute Interconnect (OCI) chiplet supports up to 4 terabits per second of bidirectional data transfer and is compatible with PCIe Gen5CIT…

Why does CMOS manufacturing give silicon photonics a cost advantage?

Integrated silicon photonics uses existing CMOS infrastructure to manufacture optical circuits, imec states, realizing cost benefits through economies of scaleC…

📎 Sources

  1. nvidianews.nvidia.com
  2. intel.com
  3. imec-int.com

Related data

Author's TakeEffectStory 編輯部

The numbers here point to a common architectural shift: moving optics from pluggable modules into co-packaged, on-chip integration. Intel's 5 pJ/bit figure against 15 pJ/bit for pluggable transceivers, and NVIDIA's 4x reduction in laser count, both target the same cost driver — the optical-electrical conversion overhead that dominates interconnect power budgets at scale. What stands out is that NVIDIA's Quantum-X Photonics switch (144 ports x 800Gb/s) and Intel's OCI chiplet (4 Tb/s bidirectional) sit at different layers of the stack — switch fabric versus chip-to-chip interconnect — rather than competing head-to-head. The metric worth tracking next is whether Intel's OCI chiplet moves from a standalone chiplet announcement into a shipping product reporting port-density and bandwidth figures comparable to NVIDIA's already-specified Quantum-X switch, which would signal the interconnect layer converging on production timelines as fast as the switch layer already has.

E
EffectStory 編輯部Editorial Team

Related

BRIEF

CNA Launches Taiwan's First News MCP Tool, AskCNA, Priced at NT$200 a Month

Central News Agency (中央社) launched CNA MCP on August 31, 2026, Taiwan's first news tool built on Anthropic's Model Context Protocol (released November 2024), letting AI agents such as Claude, ChatGPT, and Grok retrieve and cite its archives in real time. The tool integrates nearly 5 million newswire stories, 3.5 million photos, and open data from about 150 government agencies, priced at NT$200 a month with an early-bird bonus-quota plan, and received funding from Google Taiwan's nDX Digital Innovation Grant Program.

EffectStory 編輯部 ·
BRIEF

Sony Music and Warner Chappell Sue Anthropic Over Alleged 'Brazen Campaign' of Copyright Theft

Sony Music Publishing and Warner Chappell, joined by other music publishers, sued Anthropic and co-founders Dario Amodei and Benjamin Mann in the U.S. District Court for the Northern District of California, alleging illegal torrenting, scraping, and downloading of copyrighted lyrics and sheet music. The publishers seek up to $150,000 per work and $25,000 per instance of stripped copyright data, a total that could reach several billion dollars. The filing follows Anthropic's earlier $1.5 billion settlement in the Bartz case.

EffectStory 編輯部 ·
BRIEF

Why Anthropic Turned to Nscale and Lambda for $45B and $35B GPU Compute Deals

Anthropic has assembled compute capacity across at least four NVIDIA-linked providers: a $35 billion contract with Lambda tied to a Hut 8-built Texas data center, a $45 billion, six-year deal with Nscale for a West Virginia campus running NVIDIA Vera Rubin systems, a $10 billion contract with startup Volta in Norway, and a reported (unconfirmed) tenancy at Riot Platforms' Rockdale, Texas site. NVIDIA sits inside nearly every arrangement — as investor, lessor, or chip supplier.

EffectStory 編輯部 ·