Morgan Stanley Report: CPO Growth Overestimated by 30x, TSMC Capacity the Real Constraint

Written by: Rita

Tide Guide

Morgan Stanley’s CPO supply chain report has shattered the market’s rosy illusions. The current market narrative around CPO is: “AI chip traffic explosion → NVIDIA/Broadcom require co-packaged optics → Spectrum switch orders surge → suppliers benefit.” But the supply chain data in the report tells a different story: global CPO switch shipments are projected at only 23k units in 2026 and 59k in 2027—orders of magnitude below market expectations. The real bottleneck is not technological innovation (GlassBridge is already mature), but rather TSMC’s PIC production ramp-up (only 10k wpm in 2026, 25k wpm in 2027) and optical engine yield (currently 20–50%). This implies that for investors holding NVDA and AVGO, CPO demand may again fall below guidance from Q4 2026 through mid-2027, pressuring stock prices. The recommendation is clear: avoid buying into the “CPO theme” in the first half of 2026. Instead, wait until after the disappointment in Q2 2027, when TSMC’s yield data will be clearer and true demand will begin to emerge.

NVIDIA and Broadcom's Spectrum switch shipments in 2026–2027 are far below market expectations. FOCI’s report shows that NVIDIA contributed only 18% to FOCI’s 2026 revenue, indicating that actual CPO switch shipments are significantly smaller than the macro narrative suggests. If you see NVIDIA or AVGO lower their guidance in mid-2026 due to weaker-than-expected CPO demand, this disappointment stems from delays in production capacity and customer validation—not from the technology itself. Morgan Stanley’s data, though cold, is undeniable: global CPO switch shipments were 5k units in 2024, approximately 15k in 2025, and only 23k in 2026. Compared to market expectations in 2024—where many analysts anticipated CPO shipments exceeding 100k in 2026—the actual figures are less than 22% of the forecast.

The real constraints come from two stages. The first bottleneck is PIC (photonic interconnect key intermediate) capacity. The report notes that TSMC currently has a PIC capacity of 500 wafers per month, with a target to scale from 10k wafers per month in 2026 to at least 25k wafers per month in 2027. By comparison, TSMC’s typical capacity expansion cycle is 12–18 months, meaning PIC capacity may remain a bottleneck until mid-2027. This is only the intermediate PIC capacity; downstream factors must also be considered, including the complete optical engine insertion process, packaging yield (currently only 20–50%), and final system integration validation. Given last year’s actual shipment of 0.39 million optical engines, the target for 2027 is 7.78 million units, rising to 48.60 million units in 2028.

The optical engine yield challenge is the second bottleneck. The report highlights a key data point: the EPIC wafer test for Insertion 2 has been reduced from "one wafer per day" to "one wafer every six hours." While this sounds positive, it also indicates how severe the previous yield issues were, necessitating a significant acceleration in the testing process. Current installed yield remains at only 20–50%, with a target of 50% by 2028. If yield remains stuck at 30–40% by the end of 2027, it means the actual usable output of optical engines will be far below nominal capacity.

For investors holding TSMC, the contribution of the CPO business to profits over the next two years is likely to be far smaller than management’s optimistic statements suggest. While TSMC typically highlights CPO capacity expansion in its earnings reports, actual capacity utilization and yield pressures may result in lower-than-expected gross margins for this segment. TSMC’s 2026–2027 financial guidance is expected to include cautious language regarding CPO (e.g., “uncertainty in yield improvements”), which could become a downward pressure on the stock price. NVDA’s Spectrum switch shipment curve (23k → 59k → 100k) implies that this business will contribute minimally to revenue in 2026–2027 and cannot serve as a “new growth engine” narrative. If NVDA lowers its data center growth guidance in its 2027 earnings report due to underwhelming Spectrum shipments, it should not come as a surprise. AVGO faces even higher risk, as Broadcom’s base is smaller than NVDA’s; should Broadcom’s CPO orders be revised downward, the stock reaction could be more severe.

The CPO opportunities for U.S.-listed fiber optic and optical equipment companies (such as Lumentum and Coherent) have been significantly overestimated. The Morgan Stanley report identifies FAU and optical engine suppliers primarily as Taiwanese and Japanese firms (FOCI, TFC, Senko), but U.S. investors naturally ask: What about these optical equipment companies? The report’s answer is straightforward: CPO orders are still too small to be game-changers. For example, optical engine shipments are projected at just 0.39 million units in 2026, 7.78 million in 2027, and only 48.60 million in 2028. In comparison, Lumentum’s annual revenue typically ranges between $1.5–2 billion; even with high gross margins, CPO business is unlikely to generate more than tens of millions of dollars in revenue through 2026–2027. More critically, there is uncertainty around the technology roadmap: traditional FAU pathways face long-term challenges from Corning’s GlassBridge. GlassBridge eliminates precision manufacturing through passive alignment using glass waveguides, potentially displacing labor-intensive traditional FAU processes over time. Although GlassBridge has not yet been deployed at scale (the report explicitly states it is “far from mass production”), the ultimate winner of this technological competition remains unclear. If GlassBridge prevails, the long-term profit potential for companies reliant on high-precision FAU manufacturing—including U.S. optical component firms—could be severely constrained.

Advice for investors holding optical equipment stocks like LITE and COHR: Do not buy or increase positions solely due to "CPO demand." Revenue contributions from CPO in 2026–2027 will be negligible and insufficient to alter the overall trajectory of these companies. Focus instead on their performance in other high-speed signal applications, such as high-speed interconnects in data centers and 5G base stations, rather than CPO. If you see LITE or COHR raise guidance in 2026 based on "CPO order prospects," proceed with caution—this may reflect management overstating the significance of small orders.

The story of AllRing is more complex. In 2026, its CoWoS-related revenue is expected to account for 79% of total revenue, primarily from equipment supply to TSMC and ASE/SPIL. However, Morgan Stanley’s supply chain review reveals a shift: TSMC’s CoWoS capacity outlook has been lowered from 45kwpm previously to 40kwpm in 2027, and from 75kwpm to 70kwpm in 2028. This suggests weakening momentum in TSMC’s CoWoS expansion, possibly due to non-infinite demand for CoWoS from high-end AI chips, with newer technologies like CoPOS potentially beginning to replace it after 2027. AllRing’s revenue growth forecast for 2027 is 53% (from $7.4bn to $11.3bn), but this projection assumes steady CoWoS expansion, consistent Flip Chip growth, and rapid adoption of new CPO packaging. Any delay in one of these areas—such as TSMC’s CoPOS test yield falling short of expectations or a customer’s CPO adoption being postponed—could negatively impact overall growth.

Regarding A-share participation in CPO, the reality is more complex than "complete absence." SMIC (688981), as China’s largest wafer foundry, is undoubtedly investing in R&D for CPO processes. Leading optical module companies InnoLight (300308) and NeoPhotonics (300502) also have related product lines in high-end optical interconnects. Tech-Fu Communications (300394) offers high-end optical connectors that extend beyond traditional low-frequency applications. Chip design firms such as Huawei are also advancing their own CPO solutions. The real bottleneck lies in the absence of A-share companies in core process equipment (such as PIC processing equipment and optical engine manufacturing) and high-end FAU/GlassBridge components. Primary benefits are concentrated in relatively mature downstream segments like optical modules and chip design. As a result, A-share investors pursuing the "CPO concept" are essentially gaining indirect exposure, not direct exposure to core equipment or processes.

The most valuable contribution in Morgan Stanley’s report: identifying the true bottlenecks in the CPO cycle, rather than explaining how GlassBridge disrupts FAU. The report highlights that the rapid scaling of CPO hinges on three factors: whether TSMC can successfully expand its PIC capacity to 25k wafers per month on schedule, whether optical engine insertion yields can quickly rise from 20–50% to higher levels (particularly the 50% target by 2028), and whether key customers like NVIDIA and Broadcom can advance their system-level validation on time. Any delay in these three conditions will push back the entire CPO cycle. The critical constraints lie in foundry capacity and process node execution—not in the innovation of FAU connectors themselves. In contrast, the technological maturity of GlassBridge is relatively assured; the real risks reside in the depth of the supply chain.

Advice for U.S. stock investors is clear: avoid chasing the "CPO concept" stock prices of NVDA, AVGO, and TSM. If you see these stocks surge in Q2 or Q3 2026 due to strong CPO demand, it may present a short-selling opportunity. Wait for disappointment after Q2 2027, when TSMC’s yield data becomes clearer and actual CPO demand gradually emerges—that will be the real buying opportunity. Exercise caution with optical companies like LITE and COHR; CPO-related revenue will be too small between 2026 and 2027 to justify their high valuations.

Advice for A-share investors: Treat this report as a contrarian indicator. If A-share chip-related stocks surge due to "CPO opportunities," it may signal that the market has overhyped this opportunity—此时应警惕调整风险. True domestic substitution opportunities will only emerge after the CPO supply chain matures and overseas suppliers face trade barriers. Currently, A-share companies have virtually no meaningful participation in high-end chip packaging and optical interconnect technologies. While Tianfu Communications operates in the optical connector space, it primarily produces low-frequency optical connectors for data centers and does not engage in high-speed optical integration for CPO. Although photoresists and specialty gases may benefit from chip capacity expansion, their connection to CPO is less direct than that of U.S. stocks.

Disclaimer

This article is a compilation and interpretation by Chaoxiang Research of third-party brokerage research reports. The ratings, target prices, earnings forecasts, and related judgments cited herein are the views of Morgan Stanley analysts and represent the position of that institution only; they do not reflect the views of Chaoxiang Research nor constitute any investment advice.

The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.