The optical connectivity market is moving from conventional copper and pluggable links toward linear pluggable optics (LPO), near-packaged optics (NPO), network-switch co-packaged optics (CPO), and, ultimately, compute scale-up optical connectivity and optical chiplet integration. The governing principle is straightforward: as AI clusters expand, they must move more data with lower power consumption, latency, and signal loss while preserving bandwidth density and reliability. Optical connectivity is consequently becoming a system-level constraint—and a potential source of competitive differentiation—for accelerator, switch, networking, packaging, and optical-component suppliers.
For NVIDIA, this is not merely an optics-revenue question. It is an architectural and platform issue. NVIDIA’s accelerated-computing systems, high-speed networking, switch silicon, optical interconnects, and rack-scale designs could all benefit from a greater optical bill of materials per cluster. At the same time, the transition introduces execution, qualification, supply-chain, standards, and technology-obsolescence risks across the ecosystem.
The claims considered here span July 28 through August 11, 2026, with the most recent material focused on optical chiplet integration, scale-up connectivity, and AI-infrastructure positioning. Corroboration is generally limited because most claims have one source. The CPO/NPO market forecast has two sources, while the potential FCC oversight issue has five, making those themes more robust than isolated company assessments.
The Architectural Shift: Bringing Optics Closer to Compute
The strongest industry consensus is that bandwidth growth is increasing optical content rather than eliminating the need for optical suppliers. Faster links require more advanced lasers, detectors, digital signal processors (DSPs), packaging, and thermal-management systems 2. Higher lane counts and transmission speeds also increase optical content per cluster for companies such as Coherent and Lumentum 1. Thus, higher bandwidth can support greater average selling prices and more dollar content even when physical fiber counts or discrete-component counts do not rise proportionally 8. For NVIDIA, the important distinction is that AI-cluster growth can expand the value of the interconnect bill of materials even as the architecture changes.
NPO places optical components closer to processing or networking chips than conventional pluggable modules 7. CPO goes further by integrating optical engines adjacent to, or within, the switch package 17. The longer-term development path is commonly described in three stages: pluggables, network-switch CPO, and compute scale-up optical connectivity or optical chiplet integration 19. Moving optics closer to compute can support larger scale-up domains without relying entirely on long electrical links 20. Scale-up optics also face stricter latency, power, reliability, and cost requirements than ordinary scale-out Ethernet optics 19. These requirements align directly with NVIDIA’s movement toward increasingly integrated, rack-scale AI systems and tightly coupled accelerator fabrics.
The near-term outcome is more likely to be coexistence than wholesale replacement. Pluggable, NPO, and CPO architectures are expected to coexist 15. NPO preserves many of the established pluggable ecosystem’s reliability, serviceability, and multisourcing advantages 14. It also shares a common photonic integrated circuit and many optical-engine components with CPO 15. This commonality reduces the risk that investment in NPO components becomes obsolete when CPO enters broader deployment 15.
The converse risk remains material. Rapid CPO or LPO adoption could reduce demand for conventional transceivers and stand-alone DSPs 18, shift value away from traditional module assemblers 15, and strand conventional pluggable-module investments 15. NVIDIA should therefore favor architectural flexibility and multiple qualified interconnect paths rather than dependence on a single packaging standard.
Economics: Attractive Physics, Difficult Manufacturing
CPO and NPO are designed to reduce power per bit, signal loss, and electrical reach while improving bandwidth density 7,17. LPO-supported platforms are cited as reducing interconnect power consumption by approximately 60% relative to traditional pluggable optics 13. These benefits are significant, but they do not eliminate the manufacturing problem. CPO introduces greater packaging, thermal, alignment, testing, and yield complexity 17.
Early deployments consequently face higher costs, difficult qualification, rework, scrap, and potentially lower gross margins 17. Initial adoption is likely to concentrate in the highest-performance applications, where power and bandwidth benefits justify the premium 17. The measured base case for NVIDIA is therefore that optical integration becomes strategically important first in premium AI systems, rather than immediately replacing every copper or pluggable connection.
This creates a tension between rapid market-growth forecasts and gradual qualification cycles. TrendForce forecasts the combined CPO/NPO market growing from $100 million in 2025 to $39 billion in 2030—a 390-fold increase 5. CPO is already progressing into early commercial deployment 17, and Viavi reportedly held purchase orders with initial CPO revenue beginning in the current quarter 17. Yet network-switch CPO remains in validation and selective deployment 19, CPO standards and architectures remain unsettled 19, and adoption remains uncertain 10.
The appropriate interpretation is not that the forecast is necessarily invalid, but that the revenue curve may be highly nonlinear and back-end loaded. Qualification delays, constrained high-power lasers, and system-level integration requirements could defer volume even if switch roadmaps remain aggressive.
Supply Constraints and Competitive Positioning
Through the prism of supply-chain analysis, optical shortages are both a near-term bottleneck and a potential source of supplier power. Optical-component shortages have delayed hyperscaler cluster deployment 2, while high-power laser capacity carries lead times of 21–24 months 18. Applied Optoelectronics expects limited external laser-source production in 2026, a larger ramp through 2027, and approximately 400,000 units per month in 2028 18.
The company has also received more than $200 million in 1.6T orders from a major hyperscale customer 18. However, early 1.6T yields and product mix may keep margins below mature-product levels 18. These data points suggest that NVIDIA’s system shipments may be constrained not only by GPU availability, but also by optical engines, lasers, memory, packaging, and other adjacent components.
The supplier landscape is bifurcating around vertical integration and the ability to deliver qualified systems at scale. Coherent combines existing finished-module scale—estimated at roughly 17% of the market—with vertical integration across indium-phosphide lasers, silicon photonics, VCSEL arrays, photodiodes, couplers, optical engines, and transceivers 21. Vertical integration can reduce dependence on scarce external components and enable greater gross-profit capture when module prices rise 21.
Lumentum retains considerable upstream value in high-speed EMLs, continuous-wave lasers, and related components, while its Cloud Light-derived module business provides a route toward downstream value capture 21. Fabrinet benefits from complex optical manufacturing and assembly 9,17. Credo spans copper, optical, silicon-photonic, retimer, memory, and software products 12. For NVIDIA, this breadth matters because supplier selection may increasingly depend on the ability to provide complete, qualified, high-volume interconnect solutions rather than isolated components.
Where Value Moves Across the Optical Stack
Optical integration can increase total system value while reducing content for selected incumbent components. Low-loss photonic integrated circuits and integrated lasers may reduce discrete-laser content per link 14, potentially pressuring Lumentum and parts of Coherent even if total optical volumes continue to rise 14. LPO and CPO can similarly reduce DSP content in selected links 18.
There is no contradiction between rising optical demand and declining discrete-component content. The market must be viewed at several layers. Value may migrate from discrete lasers, DSPs, and module assembly toward photonic integrated circuits, optical engines, advanced substrates, switch silicon, packaging, testing, and system design. NVIDIA is positioned closer to the system and switch layers, but its economics will depend on how much of this value it captures rather than passes through to suppliers.
Standards, Regulation, and Emerging Alternatives
Optical Chiplet Integration standardizes an optical physical layer and uses NRZ signaling and WDM 19. It does not, however, define the complete protocol, coherence, congestion-control, collective-communication, or software stack 19. This leaves room for NVIDIA to differentiate through system architecture, networking software, collective communications, and workload-level optimization rather than relying on a physical-layer standard alone.
Policy introduces a separate source of uncertainty. Potential FCC oversight or restrictions covering CPO and NPO received the highest corroboration in the cluster, with five sources 3. A broader rule could create compliance costs, restrict market access, or make optical-product bookings and lead times less reliable indicators of true consumption 3,21. The proposed policy could also accelerate domestic and allied optical manufacturing 21, benefiting U.S.-oriented suppliers such as Coherent, Lumentum, and Applied Optoelectronics while increasing risk for Chinese vendors 18,21. For NVIDIA, export and ownership rules could affect the availability, qualification, and geographic deployment of optical infrastructure even where accelerator demand remains strong.
Several emerging alternatives should be treated as exploratory rather than established competitive threats. Olix presents optical processing and light-based interconnects as an alternative or enhancement to copper in AI chips 4,6, but remains dependent on successfully developing optical processors 4 and may face difficulty scaling 6. Aeva is transitioning toward data-center optical connectivity and targets NPO, CPO, and external laser-source modules 16. Sivers supplies indium-phosphide and continuous-wave laser capabilities 16. These developments reinforce the direction of travel toward photonics, but there is insufficient corroboration to conclude that they currently challenge NVIDIA’s platform position.
Implications for NVIDIA
The fundamental optics of this investment thesis are system-level. Interconnect performance increasingly determines the usable scale of an AI system. Accelerator-compute gains can be diluted if electrical reach, power consumption, latency, or optical supply prevents GPUs from operating as a coherent cluster. Optical connectivity therefore supports NVIDIA’s broader system strategy: higher-bandwidth networking, switch integration, advanced packaging, and eventually optical links more tightly coupled to compute. CPO can improve system-level efficiency and bandwidth density 2, while optical chiplet integration may extend scale-up domains beyond the practical limits of electrical signaling.
The principal opportunity is platform pull-through. NVIDIA can benefit indirectly from rising demand for 1.6T and future optical links, optical engines, switch silicon, advanced packaging, validation, and power-management infrastructure. MPS’s communications revenue grew approximately 80% year over year, with contributions from both optical modules and switches 11. This illustrates how adjacent component demand is already being funded from overlapping AI data-center budgets. The same dynamic should support NVIDIA’s networking and infrastructure ecosystem, provided that its system roadmaps convert optical capability into higher cluster utilization and customer willingness to pay.
The principal risk lies at the architecture boundary. CPO requires high-performance switch silicon, photonic engines, advanced substrates, precise assembly, thermal management, known-good-die processes, and multistage optical validation 17. Compared with conventional pluggable architectures, this creates more potential bottlenecks. High-power laser scarcity could slow adoption relative to switch-roadmap announcements 18, and larger CPO laser dies reportedly have materially lower yields 18. NVIDIA may therefore face a mismatch between announced system capability and the availability of qualified, production-ready optical subsystems. Maintaining multiple architectures—copper, LPO, pluggable, NPO, and CPO—should be viewed as strategically valuable during the transition.
The investment implication is to distinguish durable AI demand from the timing and distribution of optical economics. A fully optical-connectivity forecast may overstate the near-term revenue ramp 8, while rapid technology shifts can make current transceiver capacity obsolete 18. Conversely, optical demand may remain resilient because 800G has a large installed base 17 and faster components carry greater complexity and dollar content. NVIDIA’s fundamental exposure is therefore more robust than that of any single optical-component supplier: it is tied to the need to scale AI compute, not necessarily to one transceiver form factor.
That exposure does not eliminate risk. NVIDIA’s margins, delivery cadence, and system-level competitiveness could be affected by optical shortages, policy restrictions, qualification delays, and the eventual allocation of value among GPU systems, networking silicon, optical engines, and external suppliers.
What to Monitor
Following the light of market data, the most useful monitoring framework is operational rather than purely market-size based. Investors should track:
- The pace of CPO and NPO qualification and production.
- Evidence of optical integration in major AI-system deployments.
- High-power laser capacity.
- Adoption of 1.6T and subsequent lane speeds.
- The evolution of Optical Chiplet Integration standards.
- FCC or export-control action.
Supplier diversification and customer-backed capacity commitments may improve availability. However, larger customers financing optical capacity in exchange for purchase commitments can also increase concentration and reduce supplier pricing autonomy 18,21. The cluster therefore supports a constructive long-term view of optical connectivity as an enabler of NVIDIA’s scale-up strategy, while arguing against the assumption that the projected $39 billion CPO/NPO market will translate into immediate or uniformly distributed earnings.
Conclusion
An integrated system perspective reveals a transition governed by two opposing forces. The physical case for optics is strengthening as bandwidth, latency, power, and electrical-reach constraints push connectivity closer to compute 19,20. The industrial case is more conditional: CPO still faces high costs, low yields, difficult qualification, unsettled standards, and laser constraints 15,17,19.
For NVIDIA, optical connectivity is becoming a strategic enabler of AI scale-up rather than a discrete component category. Rising system-level optical content may coexist with declining discrete-laser or DSP content per link 8,14,18. The principal opportunity is greater system-level value capture as networking and photonics become central to cluster performance; the principal risks are supply-chain bottlenecks, policy restrictions, and delays in qualifying complete optical systems 3,17. The prudent conclusion is neither to dismiss CPO as premature nor to treat market forecasts as immediate earnings guidance. The transition is real, strategically important, and likely to unfold through successive layers of integration rather than a single abrupt replacement cycle.