From first principles, the transition from electrical to optical interconnects represents a fundamental change in information carrier physics, reminiscent of the historical shift from vacuum tubes to transistors in computing architecture. This near-term industry transition is being driven by the scaling demands of artificial intelligence and high-performance computing infrastructure, where bandwidth density and power efficiency have become critical limiting factors 6,13. The emerging ecosystem is characterized not by convergence toward a single solution, but by multiple, partially competing optical approaches—each making distinct trade-offs between integration complexity, reach, power efficiency, and implementation maturity 15,8.
Co-packaged optics (CPO) represents the highest-integration option for AI datacenter interconnects, focusing on delivering superior bandwidth density and energy efficiency through closer physical integration with compute silicon. This approach has spawned a collaborative multi-source effort to define a common physical layer specification, with the OCI MSA establishing a roadmap that begins at 4 wavelengths × 50 Gb/s = 200 Gb/s per fiber and scales to 800 Gb/s and ultimately 3.2 Tb/s+ per fiber 13,10. Crucially, the specification explicitly supports pluggable, on-board, and co-packaged implementations, reflecting the industry's recognition that different deployment scenarios will require different integration levels 10,2.
A Bifurcated Technology Landscape: CPO, LPO, and Coherent Approaches
The industry is actively developing three primary optical interconnect pathways, each with distinct value propositions and implementation challenges. Alongside CPO, Low-Power Optics (LPO) approaches—most notably Microsoft's MOSAIC MicroLED architecture—offer alternative power efficiency profiles, while coherent pluggables address different reach requirements and implementation complexities 13,15.
The MOSAIC proposal claims markedly lower per-link energy consumption compared with conventional laser optics, citing 3.1–5.3 W per 800G link and a 56–68% per-link power reduction 15. Positioned for intra-facility use with a commercialization target in late 2027, MOSAIC requires multicore imaging fiber and specialized cabling while maintaining compatibility with QSFP/OSFP form factors in prototype implementations 15. This creates a clear technology tension: MOSAIC/LPO prioritizes power efficiency and leverages commodity MicroLED/CMOS supply chains, whereas laser-based CPO and coherent solutions trade different reach, interoperability, and maturity characteristics 15.
Conventional laser-based optics retain advantages in reach (100+ meters) and benefit from broader existing production supply chains 15. This bifurcation suggests datacenter operators will face complex architecture decisions, weighing energy savings against reach requirements and operational disruption considerations 15.
Component-Level Dynamics: Where Value Capture Occurs
From a supply chain perspective, the optical ecosystem reveals a systematic stratification of value capture. The optical transceiver/module layer has become largely commoditized, experiencing rapid volume growth but persistent margin compression due to intense competition—particularly from large Asian suppliers—and significant buyer leverage from hyperscalers 5. Conversely, optical components—including lasers, modulators, advanced photonics building blocks, and increasingly, digital signal processors (DSPs)—represent more technically complex layers with fewer qualified suppliers and stronger pricing leverage, especially during periods of supply tightness 5.
This structural distinction has material implications for silicon-centric vendors like Broadcom. Emerging digital signal processors and merchant silicon are repeatedly identified as critical enablers of optical scaling, with industry venues like OFC highlighting them as material technology trends 7,12. Broadcom's recent announcement of an optical digital signal processor product places the company directly in this higher-value component/merchant silicon layer, aligning with the strategic opportunity to capture better structural economics than those available in the low-margin pluggable module assembly business 11,5.
Adoption Timelines and Power Economics: A Mixed Near-Term Picture
Several supply-side constraints and timing signals indicate a protracted transition window for optical interconnect adoption. Industry analysts have deferred expectations for mass optics adoption in AI clusters toward 2028, with material constraints—particularly T-glass and laser supply—flagged as bottlenecks stretching into 2027 8,15,1. These constraints create tangible risks for laser-based optical roadmaps and industry-wide ramp timing.
Despite these challenges, specific product availability claims exist. Coherent pluggables are expected commercially in 2026 according to one source, and companies are currently shipping or testing new pluggable and near-package optical products 15,7,4,3. This creates a tension between technological readiness—demonstrated through product announcements and availability—and the timing of meaningful hyperscale procurement, which appears to lag behind standards harmonization and implementation maturity 15,9,8.
Power economics serve as a decisive factor in architecture selection for next-generation AI clusters. Pluggable 800G transceivers are estimated at roughly ~10 W per port, representing approximately 10% of rack power in ultra-dense AI configurations 14. Alternative approaches cite material per-link power reductions: CPO demonstrates ~25–30% improvements, while MOSAIC claims 56–68% reductions 15. The varying magnitude of these claims highlights both opportunity and measurement inconsistency across vendor materials—a discrepancy that buyers will need to reconcile when making architecture decisions 15.
Implications for Broadcom's Strategic Positioning
Product Alignment with High-Value Technology Layers
Broadcom's optical DSP announcement positions the company in a technology segment that industry sources consistently identify as high-value and enabling for scaling beyond current pluggable approaches 11,7,12,7. Given that optical modules face commoditization with limited margin upside, this focus on higher-function DSPs and merchant silicon represents a strategic choice to capture superior structural economics within the optical stack 5. The alignment is particularly relevant as DSPs become essential for enabling transitions from 800G to 1.6T and beyond.
Opportunity Versus Timing and Supply Risks
Broadcom's DSP roadmap stands to benefit if adoption of advanced optics—whether coherent pluggables, CPO implementations, or LPO solutions requiring supporting merchant silicon—accelerates 11. However, mass purchase cycles by hyperscalers are projected by some industry analysts to coalesce around 2028, creating timing uncertainty for volume-driven revenue 8. Additional risks emerge from material shortages (lasers, T-glass) and implementation frictions, including cabling complexities, rack redesigns for MOSAIC architectures, and reliability challenges at extreme GPU scale 15,1,15. These factors collectively create tail risks to volume ramp timing.
Competitive Positioning in a Crowded Landscape
The optical module layer's competitive intensity—driven by Asian competitors and OEM buyer leverage—reinforces the strategic wisdom of Broadcom's emphasis on differentiated silicon/DSP products and merchant silicon platforms 5. Rather than competing directly in commodity pluggable modules, Broadcom's public content cadence and marketing activity indicate active market engagement focused on higher-value technology layers 9,4. This approach supports ecosystem visibility while avoiding the margin compression characteristic of the module assembly business.
Key Tensions and Monitoring Points
MOSAIC Versus Laser/CPO: A Fundamental Trade-off
The technology bifurcation between MOSAIC and laser-based approaches represents a fundamental trade-off that datacenter operators must navigate. MOSAIC claims substantially lower per-link power and leverages commodity MicroLED/CMOS supply chains, but reports shorter reach (up to ~50 meters) and requires new multicore imaging fiber cabling 15. Laser-based optics and hollow-core fiber retain longer reach advantages and benefit from existing production maturity, yet face laser supply shortages and T-glass constraints that pose tail risks to scaling roadmaps 15,1. This creates a market environment where different operators will weigh energy savings, reach requirements, and operational disruption differently 15.
Product Availability Versus Adoption: The Timing Disconnect
Assertions of commercial availability for technologies—such as coherent pluggables in 2026—coexist with analyst projections that mass adoption across AI clusters will materialize later, around 2028 15,8. This disconnect suggests that early commercial products may face extended ramp periods before hyperscale procurement drives sustained volume economics, creating potential challenges for revenue forecasting and investment timing.
Strategic Considerations Moving Forward
Prioritize Merchant Silicon and DSP Leadership
Broadcom's announced optical DSP aligns with the component/merchant silicon layer that industry sources identify as having stronger structural economics and pricing leverage than commoditized pluggable modules 11,5. This advantage represents a systematic opportunity that Broadcom should continue to exploit through focused R&D investment and ecosystem partnerships.
Account for Delayed Hyperscaler Ramp and Supply Risks
Public signals point to meaningful optics adoption by hyperscalers coalescing around 2028, even as specific products appear earlier in the timeline 8,15. Material constraints—particularly lasers and T-glass—combined with standards harmonization and implementation frictions can delay volume inflection points, creating timing risks to optical silicon and DSP revenue streams 1,15,8. Revenue modeling should incorporate these potential delays as realistic scenarios.
Maintain Architecture Flexibility Across Multiple Optical Approaches
The market has not yet converged on a single optical interconnect solution. MOSAIC/LPO, CPO, and coherent pluggables each claim distinct power/reach trade-offs, suggesting that hyperscalers may adopt different architectures based on specific workload requirements and operational constraints 13,15,10. Broadcom should maintain product compatibility across multiple optical architectures and continue ecosystem engagement to capture demand regardless of which approach gains traction in specific market segments.
Monitor Component Supply Dynamics and Customer Bargaining Power
The optical module layer demonstrates clear margin compression from competitive Asian suppliers and hyperscaler negotiating leverage 5. This dynamic reinforces the strategic importance of differentiated silicon/IP and careful management of exposure to constrained optical component materials with pricing leverage, including lasers and modulators 5. Supply chain resilience and diversified sourcing strategies will become increasingly important as adoption scales.
Conclusion: A Methodical Transition Requiring Ecosystem Collaboration
Much like the incremental advancement from bipolar to CMOS technologies in semiconductor history, the transition to optical interconnects represents a systematic evolution rather than an abrupt revolution. The fundamental physics—light's superiority over electrons for high-bandwidth, low-power data transmission—drives the long-term direction, but practical implementation requires addressing material constraints, supply chain bottlenecks, and integration complexities.
Success in this evolving landscape will depend on collaborative ecosystem development, with design houses, foundries, hyperscalers, and silicon vendors working together to overcome technical hurdles and establish viable production pathways. For Broadcom, the strategic emphasis on higher-value DSPs and merchant silicon aligns with historical patterns in technology adoption cycles, where differentiated components capture disproportionate value during transitional phases. The challenge lies in navigating timing uncertainties while maintaining the technical leadership and ecosystem partnerships necessary to capitalize on the optical interconnect opportunity as it unfolds over the coming years.
Sources
1. The Glass Bottleneck: How a Japanese Textile Maker Became AI's Unlikely Gatekeeper #AIChips #Supply... - 2026-03-09
2. ⚡️ #AMD, #Broadcom et #Nvidia rejoignent les géants du cloud pour définir l'interconnexion optique à... - 2026-03-13
3. LightMatter's path to low-cost datacenter optics sidesteps the co-packaging dilemma #Photonics #Dat... - 2026-03-11
4. 👏 Speed. Compatibility. Reliability. 👏 From 100G to 800G transceivers 📣 Tested for major platfor... - 2026-03-09
5. Look, the market has spent two years obsessing over the $NVDA bottleneck. And for good reason. GPUs ... - 2026-03-10
6. CPO (Co-Packaged Optics) Entire Supply Chain in One Chart _/ CPO is an emerging, highly technical ec... - 2026-03-12
7. Driven by the #AI infrastructure super cycle, the #optical #interconnect industry is at a pivotal mo... - 2026-03-13
8. Broadcom's CEO pumped the brakes on CPO, Jensen Huang says copper still matters, and analysts push t... - 2026-03-30
9. Broadcom - 2026-03-26
10. New Optical Standard for AI Clusters Forged by Tech Giants - 2026-03-12
11. Inside Broadcom's 102.4 Tbps chip rewiring AI data centers - 2026-03-12
12. Broadcom Taurus chip doubles AI bandwidth per optical lane - 2026-03-11
13. Three interconnect technologies are competing • CPO – highest integration • LPO – lower power, no DS... - 2026-03-16
14. Nvidia's Networking Division Hits $31B: Why a GPU Company Now Outsells Cisco in Data Center Switches - 2026-03-19
15. Microsoft MOSAIC MicroLED: How Laser-Free Cables Could Cut Data Center Networking Power by 50% - 2026-03-22
