TL;DR

The recent downturn in optical stocks appears to be triggered by cooling sentiment around CPO narratives, but fundamentally reflects a market reassessment of a more sensitive question: Is the volume ramp-up in 2027–2028 a period of actual revenue realization, or still merely an early-stage validation phase?

CPO (Co-Packaged Optics) itself is not being rejected. The bandwidth, power, and switching density pressures in AI data centers continue to grow, and the physical limitations of copper cables and traditional pluggable optics remain intact. The issue lies in the fact that prior stock price appreciation already priced in an aggressive timeline: following Nvidia’s push toward commercialization windows for CPO, key components such as optical engines, lasers, silicon photonics, and switching chips would rapidly scale into mass production during 2027–2028.

The SemiAnalysis report dated June 9 hit precisely this pricing assumption. According to public summaries, the report suggests that Nvidia’s 800V DC and large-scale CPO mass production may be delayed until around 2028–2029, while 400V DC remains on track for ramp-up in 2026, with some NPO (Near-Packaged Optics) projects potentially accelerating. After the report sparked market volatility, optical and related chain stocks including AAOI, LITE, COHR, GLW, and MRVL experienced single-digit to double-digit corrections. What markets are trading is not whether CPO is valid, but how quickly CPO can turn into concrete orders.

Yet this is not a one-sided bearish narrative. “White Hair Warren Buffett,” AI supply chain analyst Serenity (@aleabitoreddit), responded swiftly to SemiAnalysis, arguing that their analysis overly relies on conservative engineering models and underestimates Nvidia’s ability to compress hardware cycles. Based on her interpretation of signals from Nvidia, Lumentum, Foxconn, and others, she emphasized that CPO remains on track for ramp-up in the second half of 2026, the second half of 2027, and throughout 2028.

The value of this debate lies not in determining who “won,” but in shifting the valuation anchor of the optical supply chain back from end-state narratives to timeline validation: CPO will arrive—but its slope determines value allocation among NPO, pluggable modules, light sources, and switching chips.

Timeline Reassessment Behind the Optical Stock Decline

Over the past few months, investors bought into the optical supply chain not for current revenues, but for capital expenditure shifts tied to next-generation network architectures in AI data centers.

As model training and inference clusters expand, communication pressure between GPUs, across racks, and within data centers continues to intensify. Networking is no longer just a peripheral system outside servers—it increasingly resembles a bottleneck in the AI factory. Higher bandwidth density and lower power consumption directly enable greater scalability per compute cluster, which explains why CPO has moved to center stage.

The theoretical appeal of CPO is straightforward: place optical engines as close as possible to ASICs (application-specific switching chips), shorten high-speed electrical signal paths, reduce power, loss, and signal integrity stress caused by serial-parallel conversion circuits and copper traces. Compared to traditional pluggable modules, CPO offers superior power and density potential in the high-bandwidth era.

The market’s flaw lies in prematurely trading “correct direction” into “certain volume ramp-up.” Nvidia’s official press release states that the Vera Rubin platform will introduce Spectrum-X Ethernet Photonics, with CPO switches already entering production for horizontal scaling and cross-cluster deployment in AI factories. According to a June 3 Taiwan media report, Nvidia network executives confirmed that Spectrum-X CPO switches have been shipped to select partners, with capacity expected to scale in the second half of 2026.

These signals confirm CPO is advancing, but do not equate to risk-free, production-grade mass orders already locked in. For capital markets, there’s a significant valuation gap between “entering production,” “shipped to select partners,” “customer evaluation,” and “mass production.” The recent correction was essentially the market beginning to recalibrate these distinctions.

SemiAnalysis’ Conservative Model: CPO’s Challenge Lies in Systems Engineering

SemiAnalysis isn’t denying CPO’s future. Its core thesis is more nuanced: while CPO’s theoretical advantages are clear, scaling it into real-world deployment is slower than the market imagines.

The reason goes beyond isolated component readiness. CPO concentrates complexity previously distributed across modules, line cards, and systems into a deeply integrated, tightly coupled architecture. Higher integration brings better single-point performance, but also amplifies manufacturing, testing, repair, and supply chain resilience challenges.

The advantage of traditional pluggable modules lies in modularity—when a module fails, it can be replaced, and suppliers are relatively easy to switch. CPO differs: optical engines are placed closer to ASICs, sometimes even within the same package. While this proximity enables superior power and density gains, it also expands the repair radius. If a single optical component fails, the impact extends beyond a simple plug-in module—it could affect higher-value switching chips and entire system units.

SemiAnalysis’s previous CPO Book repeatedly stressed serviceability, reliability, yield, and supply chain maturity. Especially in hyperscale cloud environments, performance is not the only metric. Large customers demand high reliability and maintainability. If failure rates, repair processes, and replacement costs remain uncontrollable in production settings, even optimal power models may face delayed adoption.

InP lasers are another point of contention. Laboratory-level port longevity data proves technical feasibility, but does not guarantee long-term operation, batch manufacturing, field maintenance, or supply chain redundancy at scale. For investors, this distinction is critical: lab validation confirms direction, field reliability determines ramp-up timing.

Within SemiAnalysis’s framework, NPO and pluggable modules aren’t outdated paths—they’re pragmatic intermediates before full engineering risks are resolved. CPO is theoretically superior, but if full-scale deployment requires more time, the market must reprice these “less terminal, yet easier to manufacture and maintain” solutions accordingly.

Serenity’s Counterargument: Nvidia May Compress Hardware Cycles

Serenity’s rebuttal doesn’t deny CPO’s engineering challenges—it argues that SemiAnalysis underestimates Nvidia’s organizational capability in AI hardware cycles.

Her logic is clear: while typical hardware rollouts face delays due to yield, reliability, and customer validation, Nvidia is not a typical customer. It defines GPU cluster architecture, drives network, switch, system integration, and supply chain rhythms. When AI factory expansion hits bottlenecks from power and bandwidth constraints, Nvidia has strong economic incentives and industry leverage to compress traditional rollout timelines.

Serenity’s evidence rests on two layers. First, publicly verifiable company statements: Nvidia’s official confirmation that Spectrum-X Photonics has entered production, along with Lumentum’s Q2 FY26 disclosures mentioning CPO order volumes and delivery schedules. Lumentum reported receiving multi-billion-dollar incremental CPO orders with deliveries scheduled for early 2027, and noted that CPO-related business is expected to enter broader ramp-up in the second half of 2026.

The second layer involves her interpretation of supply chain signals—such as Foxconn delivering optical switches to Nvidia ahead of schedule. However, the exact scale, whether they are test prototypes or production orders, still requires further public confirmation.

This is the crux of the divergence: SemiAnalysis trusts that systemic engineering variables will naturally extend timelines, whereas Serenity believes Nvidia’s supply chain execution capability will steepen the curve.

These views aren’t mutually exclusive. Nvidia can accelerate CPO into production and customer validation, possibly enabling early adoption in certain horizontal scaling scenarios—but this doesn’t automatically imply all AI data center networks will rapidly shift to CPO by 2027. Variations in horizontal vs. single-rack expansion, intra-rack vs. inter-rack, and differing reliability tolerance and cost models across clients mean adoption will likely be layered.

Serenity counters the overly conservative conclusion that “CPO will be significantly delayed”—not to prove that “CPO is already fully risk-free.” For markets, this is enough to support a rebound from oversold levels, but insufficient to rewrite the aggressive 2027–2028 revenue trajectory back into certainty.

Why NPO Suddenly Gained Importance

NPO has gained sudden prominence in this debate because it sits squarely between SemiAnalysis’s and Serenity’s two competing frameworks.

NPO is neither the opposite of CPO nor a simple continuation of traditional pluggable modules. Its core concept is placing the optical engine on a pluggable substrate near the ASIC, shortening electrical signal paths to achieve partial power and density gains, while preserving better testability, replaceability, and supply chain flexibility.

If SemiAnalysis’s conservative model aligns more closely with reality, deep packaging in CPO may slow down due to yield, repair, and reliability issues—making NPO a more sustainable choice over a longer horizon. It allows hyperscalers to gradually accumulate experience in optical interconnect operations without assuming full CPO risk, and gives existing optical module and engine suppliers a longer window.

If Serenity’s assessment of Nvidia’s execution strength proves accurate, NPO won’t disappear. More likely, NPO, CPO, pluggable modules, and copper interconnects will coexist across different network tiers. Nvidia’s own roadmap indicates that horizontal scaling with CPO can proceed first, while certain single-rack expansion scenarios may still rely on copper or hybrid architectures through 2027–2028.

For investors, this means no longer pricing the optical supply chain using a binary “CPO wins, others lose” lens. Different technology paths benefit different segments: CPO favors high-integration optical engines, laser sources, silicon photonics, and switching chip ecosystems. Extended windows for NPO and pluggable modules could sustain orders and gross margins for existing optical module vendors, connectors, materials, and certain light source suppliers.

The market’s prior mistake was prematurely translating technological end-states into single-path revenue trajectories. What’s now being reopened is the valuation space for intermediate pathways.

Production-Grade Data Is the Next Validation Point

This debate won’t conclude with a single report or a series of social posts. SemiAnalysis reminds markets that CPO’s difficulty lies in systems engineering. Serenity reminds markets that Nvidia’s supply chain orchestration might reshape traditional hardware rollout rhythms. The true divergence will be validated by production-grade data between the second half of 2026 and 2028.

The next critical factor isn’t whether shipments exist, but shipment granularity. Shipment to select partners, customer evaluation, initial production, scale ramp-up, and mass deployment represent entirely distinct phases. Subsequent descriptions from Nvidia on Spectrum-X / Quantum-X Photonics mass production, and forward guidance from optical suppliers like Lumentum and Coherent on orders, capacity, and gross margin in earnings reports, will carry far more weight than isolated meeting statements.

Equally important is field reliability and repair data. If CPO demonstrates stable failure rates, manageable replacement workflows, favorable yield curves, and acceptable total cost of ownership in production environments, SemiAnalysis’s conservative model will be revised. Conversely, if such data remains confined to lab tests or small-volume validations, the window for NPO and pluggable modules will continue to expand in market perception.

What the optical supply chain is now pricing isn’t CPO’s survival—it’s the slope of the timeline. The next validation points will lie in whether “entering production” translates into sustainable volume ramp-up, and how quickly that volume ultimately reflects in orders, gross margins, and customer deployment metrics.

Although SemiAnalysis expressed concerns about CPO technology over the next two years, they still identified five semiconductor sub-sectors they view positively: Copper/AEC/ACC; Pluggable Optics/DSP; CPO Test Equipment; Power Gray Space/UPS Continuation; Board-Level VRM/Silicon-Based Power/Passives.

Specific names are integrated in the image below for readers' reference.

Original: BlockBeats