CPO Volume Ramp Pushed to 2028 — and That's Not Bad News: Pluggable Optics Just Got a Longer Shelf Life (W25)

2 days ago

7 min read

This week the market read a SemiAnalysis report as the death knell for optical components. It read it backwards: if CPO ramps a year later, pluggables — and the laser upstream of them, already short enough to command price hikes — get an extra year of earnings.

Introduction

On June 9, SemiAnalysis released a report that cut the market's optimistic expectations for large-scale CPO (Co-Packaged Optics) deployment in 2027, arguing that meaningful volume will likely slip to 2028–2029 (source: KAD8). Optical component stocks swung on the news, with traders selling it as "the CPO story is broken."

But that reaction gets cause and effect upside down. A delayed CPO ramp doesn't hurt the optical supply chain as a whole — it hurts only the small group betting that CPO would displace pluggable optical modules early. For the much larger group of players — pluggable makers Coherent and Lumentum, China's big three module makers, and the lasers upstream of them that are short through 2027 — the delay is a lifeline. This weekly review makes one point: while the market panics about "CPO arriving a year late," money is flowing to the pluggable camp whose shelf life just got extended.

1. CPO a Year Late Means Pluggables Earn a Year Longer

Starting point: one report, two readings

SemiAnalysis's argument isn't complicated: CPO solders the optical engine right next to the switch ASIC, saving power and space, but yield, serviceability, and supply-chain maturity aren't yet ready to support mass deployment. So "full ramp in 2027" is wishful thinking, and the real inflection waits until 2028–2029 (source: KAD8).

The market's knee-jerk reading was "CPO is dead." Change the subject of the sentence, though, and it means something entirely different: if next-generation data centers can't switch to CPO until 2028, then the 800G and 1.6T upgrade wave of 2026–2027 has to be carried by pluggable optical modules. Every quarter CPO slips is another quarter of orders pluggables get to keep.

Evidence: demand numbers aren't shrinking — they're being revised up

If CPO really threatened pluggables, their demand curve should be bending down. The opposite is happening. This week, overseas customers raised their 2026 1.6T procurement from 10 million units to 15 million and then to 20 million, driven by accelerating GB300 and Rubin deployments; analysts estimate total industry 1.6T shipments in 2026 could reach 25 million units, with 800G above 55 million (sources: Sina Finance, Cailian Press).

The most direct beneficiaries are China's big three module makers. Innolight (300308) was first to mass-produce 1.6T silicon photonics modules, is tied to NVIDIA and Google, leads 1.6T market share, and hit an intraday all-time high this week; Accelink Technologies (002281) has 1.6T volume delivery capability and is in customer qualification; Eoptolink (300502) is using LPO (linear pluggable optics) to stake out the short-reach, low-power niche, with its 800G LPO already certified on NVIDIA's H100 platform (source: Cailian Press). On the U.S. side, Fabrinet (FN) posted fiscal Q3 optical communications revenue of $833 million, up 29% year over year, and broke ground on a new 2-million-square-foot facility that adds 50% capacity (source: optics.org). A contract manufacturer willing to commit 50% more capacity is betting that pluggables have several more years of volume growth.

Interpretation: CPO vs. pluggables isn't zero-sum — the market did the math wrong

Put SemiAnalysis's timeline, the upward 1.6T revisions, and the big three's shipments side by side, and this week's conclusion is clean: in the near term, CPO and pluggables are not substitutes but a relay. CPO arriving late pushes out the pluggable lifecycle, and extends the life of short-reach AECs (active electrical cables) along with it — which is why Credo (CRDO), focused on AECs and expected to triple FY26 revenue to over $1.3 billion, was actually grouped with the winners this week (source: Yahoo Finance).

CPO's timeline is for 2028 bets; pluggable orders are for 2026 income statements. This week the market made the latter's shareholders pay for the former's delay.

The practical takeaway for STT readers: don't treat "CPO progress" as the only thermometer for optical communications. A slower CPO actually extends the good times of shortages for vendors making 1.6T pluggables and their components by another year. Follow the "who can actually ship" thread upstream, and you hit the hardest bottleneck in the entire chain.

2. The Real Bottleneck Isn't the CPO Timeline — It's the Laser

Starting point: shortages through 2027, at the very top of the chain

While downstream module makers race to expand capacity, the tightest link in the supply chain has long since retreated to the light source at the very top. This week's signals were blunt: shortages of InP (indium phosphide) substrates and 200G/lane EMLs (electro-absorption modulated lasers) remain unresolved, with no second source in sight. Lumentum is currently the only supplier able to mass-produce 200G/lane EMLs — the critical component for 1.6T modules — and analysts expect double-digit price increases for 200G EMLs in 2026, with ASPs roughly twice those of 100G (sources: Ben Pouladian, TechTimes).

What makes this bottleneck so damaging is where it sits. Whether a module can be built or a CPO engine can be lit ultimately comes back to "is there a laser to serve as the light source." You can add another foundry for DSPs or open another line for PCBs, but the learning curves for InP crystal growth, epitaxy, and EML processing are measured in years. However much capacity downstream adds, it just idles when upstream materials run short.

Evidence: NVIDIA locks in capacity with equity stakes, Taiwan epi houses ride the price hikes

To see how scarce this laser is, just watch where capital is moving. Back in March, NVIDIA put roughly $2 billion each into Lumentum and Coherent, with one goal — locking up EML capacity through 2027 (source: TechTimes). When the customer-side leader is willing to stockpile a single component through prepayments and equity stakes, this is no ordinary procurement relationship — laser chips are being fought over as strategic materials.

This shortage directly benefits Taiwan's epitaxy foundry players. WIN Semiconductors (3105) grew Q1 revenue 28% year over year and swung back to profit; all four of its product lines are growing in Q2, with optical and infrastructure expected to grow double digits. It has positioned itself in Lumentum's VCSEL and EML supply chain, with 2026 EPS estimated at about NT$6.5 and 2027 targeting NT$11 (source: StockFeel). VPEC (2455) is expected to more than double shipments of transmit-side lasers (LD) for data center 800G and silicon photonics, with AI-related revenue rising to an estimated 70% of the total. With upstream supply short, prices rising, and second sources slow to arrive, these Taiwanese epi foundries sit where bargaining power in the entire supply chain is most concentrated.

Interpretation: the longer the shortage, the more pricing power concentrates upstream

Put the SemiAnalysis delay, the 1.6T upward revisions, and EML shortages through 2027 together, and you reach a counterintuitive conclusion: the players most certain to profit from this supply-demand imbalance aren't the buzziest CPO names, but the upstream makers of light sources and epitaxy.

Downstream competes by expanding capacity; upstream wins on scarcity. When everyone is short the same laser, pricing power isn't in the buyer's hands.

The practical takeaway is concrete: to judge the health of this optical rally, don't just watch how strongly module makers' revenue is growing year over year — watch upstream light-source lead times and gross margins. Longer lead times and steeper price hikes are hard proof that demand is real and supply can't keep up; the day EML lead times start loosening and upstream margins fall back is when this supply chain needs a fresh checkup. Right now, not a single loosening signal has appeared.

3. Other Signals Worth Tracking This Week

Intel (INTC): 18A foundry momentum is strengthening, with yields improving 7–8% per month. Google is reportedly outsourcing more than 3 million TPUs to Intel (2028), NVIDIA is evaluating 18A for a next-generation multi-die GPU, and Tesla is the first major 14A customer. This is the first real signal of major external customers for Intel Foundry, and TSMC's monopoly on leading-edge nodes is starting to show some noise (source: TrendForce).
Marvell (MRVL): its Computex keynote set the theme "The Future of AI Scaling Depends on Connectivity," with the CEO saying the copper interconnect bottleneck is moving inside the rack and that CPO is the only answer for scale-up. In April it acquired silicon photonics device startup Polariton, integrating upstream from DSPs into silicon photonics and setting up a head-on clash with Broadcom's CPO roadmap (source: DataCenterDynamics).
Broadcom (AVGO): FY26 Q2 AI semiconductor revenue was $10.8 billion, up 143% year over year, with Q3 AI revenue guided to $16 billion, up more than 200%. Tomahawk 6 (a 100T Ethernet switch) has been in volume production for over a year, and Tomahawk 7 (2027) doubles bandwidth again. With twin engines in custom ASICs and AI networking, it remains the biggest connectivity beneficiary outside NVIDIA (source: Motley Fool earnings call transcript).
TSMC (2330): its COUPE silicon photonics platform is set for volume production in 2026, the CoPoS panel-level packaging pilot line completes in June with volume ramp in 2028–2029, and CoWoS/SoIC capacity grows more than 80% annually through 2027. Even with CPO delayed, Taiwan's foundations are being laid on schedule (source: Focus Taiwan).
China's rising supply chain: a long TechNews piece (6/12) warned that U.S.–China decoupling can't stop the rise of China's optical communications supply chain, and that Taiwan risks falling behind. The tension between non-China supply chains and Chinese substitution is a structural theme worth digging into further (source: TechNews).

4. Conclusion

This week's takeaway is clean: SemiAnalysis pushed the CPO ramp to 2028, and the market read it as bearish — backwards. If CPO comes a year late, pluggable optical modules and the upstream EML lasers short through 2027 earn an extra year. Demand hasn't weakened — 1.6T procurement revised up to 20 million units, the big three shipping at record levels, Fabrinet committing 50% more capacity — the only thing that weakened was share prices oversold in the panic.

Two things are worth watching next. First, upstream light-source lead times and margins: whether 200G EMLs see the double-digit price hikes analysts expect, and whether WIN Semiconductors' and VPEC's epitaxy utilization holds up, will be the best thermometer for how long this imbalance can last. Second, the battle for CPO positioning is intensifying — Marvell buying Polariton, Broadcom pushing third-generation CPO, TSMC scheduling COUPE and CoPoS. Volume may arrive in 2028, but the 2026–2027 design-in window is now. What's delayed is the ramp, not the preparation. Until EML lead times loosen, the pluggable-extension story will keep making the bull case for optical communications.

This article is for technology and industry trend analysis only and does not constitute investment advice.

STT Market Insights | June 15, 2026 | Compiled by the STT editorial team