We see customers aggressively developing scale-up and scale-out CPO, with Asia semis names including TSMC, ASE, KYEC, FOCI, AllRing, and TFC helping realize these goals. We remain optimistic about CPO, although it is still very early.

2026 marks the beginning of CPO development: We published a CPO deep-dive in January 2026 (see report) highlighting that we might see scale-out and scale-up development in the coming years. After a year of continuous supply chain effort, we have started seeing more CPO/NPO and even OIO developments from NVIDIA, AMD, Broadcom, and Marvell (all covered by Joe Moore).

Scale-out CPO – NVIDIA as a pioneer, with Broadcom and others all catching up: At GTC 2025, NVIDIA launched its Quantum and Spectrum CPO switches. It then pushed Spectrum as the standard spec for Vera Rubin NVL72 racks at January's CES. For 2026, we see optical engine shipments at 600k-1mn units and switch shipments at ~23k units due to assembly lead time, with Spectrum as the key product (see Andy Meng's recent Insight report). Looking to 2027, we expect the next-generation Quantum switch to gain momentum. Elsewhere, Broadcom launched its first 102.4T CPO switch, the "Davisson" platform for Tomahawk 6, with shipments starting 2H26.

Scale-up CPO – NVIDIA likely to launch a CPO version for Rubin Ultra as the Kyber alternative; AMD also actively participating: Our checks suggest TSMC targets 10kwpm PIC capacity in 1Q27 to support strong CPO demand. We think NVIDIA could consume the most. Besides scale-out, we see it likely launching a CPO version of the Rubin Ultra NVL144 solution, with Quantum supporting GPU-to-GPU communication as the NVLink switch interconnect. An Optical I/O solution is also under research for the next-generation GPU. On the other hand, AMD targets a scale-up CPO version for its MI550 or MI650 rack system in 2027e/2028e, based on its development progress. We also see ASIC vendors exploring CPO on GPU.

How to think about Transceiver vs. CPO? Share competition could follow technology innovation. However, we believe the demand outlook is positive for both thanks to ongoing AI infrastructure investment. We think the fundamental picture could be similar to NVIDIA's GPU and CSPs' ASICs – both have a positive outlook despite some share competition.

Stock implications: We remain optimistic about CPO development, although we are still at the very early stages. We are OW on key Asia CPO enablers, including TSMC (COUPE platform), ASE (packaging/testing), KYEC (final test), FOCI (FAU), AllRing (AOI/coupling equipment), and Chroma (OE and CPO switch ASIC testing equipment). Andy Meng is EW on TFC after a strong rally in recent weeks, as potential positive impact from CPO appears largely priced in. See page 2 for an ASIC update.

Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley Research as only a single factor in making their investment decision.

• Our latest supply chain checks suggest that MediaTek's Google TPU's initial chip test result is okay, and that during mid-March we could have better visibility for mass production timing and volume for 2026.

• OpenAI and AWS has established a strategic partnership on Feb 27 (see announcement). AWS will invest US$50bn in OpenAI, and OpenAI will consume 2 gigawatts of Trainium capacity through AWS infrastructure to support demand for Stateful Runtime Environment, Frontier, and other advanced workloads.

• Following the OpenAI/AWS news, our checks indicated that Trainium 3 system performance didn't seem very competitive. Hence, we expect a pull-in of Trainium 4. All told, we expect limited upside for Alchip in 2026-27 from Trainium 3, but more upside from 2028 onward from Trainium 4 (2nm without CPO) and Trainium 5 (14nm with potential CPO adoption). We reiterate OW on Alchip, and believe the company might comment on this demand implication on its March 6 results call.

Our global colleagues Andy Meng and Meta Marshall recently published two reports to further explore networking datacenter market dynamics, including discussions regarding CPO vs. Transceivers.

• Global Technology: AI Transceivers: Growth Dominates Disruption (26 Feb 2026)

• Telecom & Networking Equipment: Into the Spotlight: Optical Market Opportunities (23 Feb 2026)

We believe the AI transceiver market is entering an exponential growth phase, driven by scale-out, scale-up, and scale-across of architectures in AI data centers. We project that global AI transceiver demand will rise from ~41mn units in 2025 to ~95mn units by 2028, while industry TAM expands from ~US$18bn (2025) to ~US$50bn (2028) – almost 3x growth in three years. We think these three forces together explain why absolute transceiver volumes continue to surge – even with CPO penetration.

Some share competition between CPO and transceivers could follow the technology innovation. However, we believe the demand outlook is positive for both thanks to ongoing AI infrastructure investment. We think the fundamental picture could be similar to NVIDIA's GPU and CSPs' ASICs – both have a positive outlook despite some share competition.

In 2025's GTC, we saw NVIDIA launch its Quantum X800 and Spectrum-X CPO switch. In the 2026 CES, we saw it elaborate more on Spectrum's product plan, which is considered to be the recommended Ethernet switch product together with the new GPU Vera Rubin platform.

In the upcoming GTC event on March 16, we expect NVIDIA to talk about more details regarding the Spectrum CPO. It is likely to launch 1-2 new versions of Quantum switches, for which the launch time will mainly be in 2027. We believe the Quantum switch's development is highly correlated with NVIDIA's plans for further introducing scale-up architecture for the in-rack communication solution.

For 2026, we expect total CPO switch shipments to be 23k units, with the majority being 100T switches with NVIDIA's Spectrum switch taking the lion's share.

Looking into 2027, we expect total CPO switch shipments to reach 59k units, growing further to 200k units in total in 2030e, which implies a 144% CAGR during 2024-2030e.

Our supply chain checks suggest that currently TSMC has ~500 wafers per month of PIC capacity, and we are seeing that its Silicon Photonics platform (COUPE) customers are requesting a very aggressive expansion schedule starting from 1Q27 of 10kwpm of capacity, which is partially to prepare for the scale-up or Optical I/O introduction. This points to our bull case scenario – in which CPO switch shipments are likely to reach 80k in 2026, 180k in 2027, and 250k in 2028.

Exhibit 1: We expect scale-out CPO switch to grow rapidly, from 5k units in 2025 to 200k units in 2030e, implying 144% CAGR over 2023-2030e

Source: Yole, Morgan Stanley Research (e) estimates

Exhibit 2: CPO helps to reach scale-out networking and scale-up computing goal for AI/HPC applications

CPO supply chain comments during 4QCY25 earnings calls

Exhibit 3: TSMC's COUPE technology: Across SiPh pluggables to 3DIC

Exhibit 4: Broadcom's CPO platform

At ISSCC 2026, NVIDIA (covered by Joe Moore) presented a new silicon photonics technology aimed at improving high-speed data connectivity for AI systems. In the paper by Song et al., titled "A 32Gb/s/ 256Gb/s/Fiber Half-Rate Bandpass-Filtered Clock-Forwarding DWDM Optical Link in a 3D-Stacked 7nm EIC/65nm PIC Technology," the company demonstrated a 256 Gb/s per-fiber optical link using a wavelength-division multiplexing (DWDM) approach.

As the industry moves toward 224G electrical signaling, NVIDIA highlights the growing technical trade-offs. When data rates double, receiver sensitivity drops by roughly 4 dB, and bandwidth limitations become more severe. At 224G, traditional electrical I/O can have a relatively high raw bit error rate (around 1E-4 to 1E-6), which requires complex forward error correction (FEC) to meet system requirements. That added correction introduces roughly 10 nanoseconds of latency.

Instead of pushing a single electrical channel to ever-higher speeds, NVIDIA uses DWDM (Dense Wavelength Division Multiplexing) to combine eight 32G data channels onto a single fiber, delivering 256 Gb/s in total. This approach keeps latency below 1 nanosecond and avoids the need for complex FEC or machine learning-based signal processing.

A key component in this design is the Microring resonator, a very small optical device that can precisely select and manage different wavelengths of light. NVIDIA uses nine Microrings—eight for data and one for the clock signal—within the 1310 nm wavelength range, spaced about 200 GHz apart. While Microrings are sensitive to manufacturing and temperature variations, they allow multiple optical functions to be integrated into a compact area.

The system also uses a clock-forwarding architecture with bandpass filtering. Rather than recovering the clock signal at the receiver in the traditional way, NVIDIA sends a clean reference clock along with the data. The receiver then filters noise and stabilizes the signal using injection-locked oscillators, helping achieve low latency and good energy efficiency.

Finally, NVIDIA packages this technology using a 3D-stacked design called "Optics on Interposer" (by leveraging TSMC's SolC technology and COUPE platform). A 7nm EIC (electronic chip) is vertically bonded to a 65nm PIC (silicon photonics chip) and placed next to the GPU on the interposer. The result is a compact, high-density solution—

reaching 1.33 Tb/mm² — that improves space utilization around the GPU and supports the high-bandwidth interconnects required for large-scale AI datacenter systems.

See Joe's report discussing scale-up network content for more details – Semiconductors: Scaling the AI Opportunity: A Scale-Up Network Primer (29 Aug 2025).

Source: Song, S., Mehta, N., Nedovic, N., Rekhi, A., Kalogerakis, G., Xu, L., ... Gray, C. T. (2026). A 32Gb/s/λ 256Gb/s/Fiber Half-Rate Bandpass-

Filtered Clock-Forwarding DWDM Optical Link in a 3D-Stacked 7nm EIC/65nm PIC Technology.

We believe the CPO sector is still in the very early stages of development – similar to where the GPGPU vs. ASIC debate stood two years ago. Given the early product cycle, we think it is too early to downgrade any CPO stocks. Significant progress and multiple product announcements are still ahead, and each new disclosure is likely to serve as a catalyst for CPO stocks.

• Among Asian semis, we are OW TSMC, KYEC, ASE, Samsung and Aspeed.

• We are also positive on Asian ASIC design service providers Alchip and GUC, and CPO suppliers FOCI and AllRing.

Exhibit 6: P/E multiple trend of AI semis

Alternative AI semis group: AMD, Alchip, Andes, Marvell, Broadcom. AI semi enablers group: TSMC, Synopsys, Cadence, ASML, BESI, Ibiden, KYEC, Advantest. Source: Company data, Morgan Stanley Research.

Exhibit 7: We continue to see AI chip quarterly revenue increasing

Data center/HPC semi revenue: NVIDIA + AMD

Source: Company data, Refinitiv, Morgan Stanley US Research (e) estimates.

Exhibit 8: TSMC AI-related revenue CAGR could reach 60% from 2024e to 2029e

TSMC AI revenue breakdown

☑ General-purpose AI ☐ Custom AI chips (ASICs) ☐ CoWoS/wafer test ☐ AI server CPU

Source: Company data, Morgan Stanley Research (e) estimates

Exhibit 9: AI GPU H100 per GPU per hour as of end-March

Source: Company data, Morgan Stanley Research

Exhibit 10: AI ASIC equivalent computing power – 16x Inferentia 2 per hour

Source: Company data, Morgan Stanley Research

Exhibit 11: NVIDIA 5090 graphic cards pricing rebounded recently, mainly because of price hike expectations in the market and strong AI inference demand China

Source: TaoBao, Morgan Stanley Research, various media (e.g., Vice, Jan. 3, 2026)

Asia-Pacific Technology: AI Supply Chain: TPU/ASIC updates; ICMS NAND demand calculation (27 Jan 2026)

Asia-Pacific Technology: AI Supply Chain: CES implications, ASIC production, China AI chips (6 Jan 2026)

Asia-Pacific Technology: AI Supply Chain: TSMC CoWoS Expansion; ASIC Dynamics; Asia Field Trip (1 Dec 2025)

Greater China Semiconductors: 2026 CoWoS Expansion to Catch Up After TSMC's 3nm Capacity Increase (17 Nov 2025)

Asia-Pacific Technology: AI Supply Chain: TSMC to expand 3nm capacity for major AI customer's growth? (12 Nov 2025)

Asia-Pacific Technology: AI Supply Chain: Reconciling the strong AI infrastructure plan with TSMC's prudent CoWoS expansion (29 Oct 2025)

Global Technology: Supply-chain Reorientation (24 Jul 2025)

Global Technology: China – AI: The Sleeping Giant Awakens (13 May 2025)

Global Technology: AI Cloud Capex in the Spotlight (26 Feb 2025)

Global Semiconductors: AI ASIC 2.0: Potential winners (15 Dec 2024)

Global Technology: Global Technology – Dawn of the AI Smartphone Era: Edge AI – Apple Intelligence Fuels Innovation – More Charts, Fewer Words (17 Jul 2024)

Global Technology: AI PCs To Usher In The Next Leg Of PC Market Growth (21 May, 2024)

Global Semiconductors: Windows on Arm AI PC – This Time is for Real (7 May 2024)

Key Upstream AI supply chain companies

Global Technology: AI Transceivers: Growth Dominates Disruption (26 Feb 2026)

Global Technology: AI boosts NAND demand, but Asian module maker margins may soon compress (24 Feb 2026)

Himax Technologies Inc: Management Expects 1Q to Mark the Trough for 2026; EW (12 Feb 2026)

WIN Semiconductors Corp: AI revenue below expectations; reiterate UW (11 Feb 2026)

SMIC: Structural AI Positioning Unchanged; OW (11 Feb 2026)

Greater China Semiconductors: Cloud Semis: Further upside ahead from shipment and margin growth (10 Feb 2026)

We expect CoWoS-related revenue to account for 76% of AllRing's total revenue in 2026, with AllRing focusing on oS equipment supply to TSMC and ASE/SPIL. However, we see more non-TSMC 2.5 packaging solution vendors qualifying more Taiwanese CoWoS equipment vendors, creating a new TAM for AllRing. For 2027, we expect AllRing's CoWoS revenue to be up 15% Y/Y, as TSMC's CoWoS expansion in 2027e would be at least 160-170kwpm, and also expect it to start preparing equipment for CoPoS (if the trial round goes smoothly in 2026) and its AP fab in Arizona (MP likely starting in 2028).

For CPO, besides FOCI, we see AllRing penetrating another CPO supply chain, with the equipment also focusing on coupling and AOI. The company is working to expand its capabilities into front-end (EIC/PIC) and back-end (MPO/fiber) coupling, and is also seeking to broaden its customer base. We raised our coupling equipment assumption from NT $10mn to NT$15mn, as our previous estimate was too conservative. Thus, we think CPO related equipment will contribute 13% of AllRing's total revenue in 2026, accelerating to 29% in 2027, driven by increased CPO switch shipments and opportunities tied to the Rubin Ultra and switch CPO variant.

Exhibit 12: Bottom-up analysis for AllRing's revenue

Source: Company data, Morgan Stanley Research (e) estimates

Exhibit 13: CoWoS supply capacity breakdown (average)

Source: Company data, Morgan Stanley Research (e) estimates. Note: estimates are compiled using our supply chain checks.

Exhibit 14: CoWoS supply capacity breakdown (by year-end)

CoWoS Supply Capacity Breakdown (By Year End)

Source: Company data, Morgan Stanley Research (e) estimates. Note: estimates are compiled using our supply chain checks.

Exhibit 15: Detailed CoWoS capacity expansion by year end and by vendor

Source: Company data, Morgan Stanley Research (e) estimates. Note: estimates are compiled using our supply chain checks.

Exhibit 16: CoWoS demand breakdown by customer

NVIDIA Broadcom AMD XEnx AWS/Annapurna AWS/Achip Marvell GUC MediaTek Others

Source: Company data, Morgan Stanley Research (e) estimates. Note: estimates are compiled using our supply chain checks.

TSMC's advanced packaging fab planning

Source: Company data, Morgan Stanley Research

We factor in actual 2025 results, lift our EPS estimates by 18% for 2026 and 47% for 2027, and introduce 2028 forecasts: We factor in actual 4Q results, with the earnings increase mainly from strong non-op gains from stock investments. We lift our 2026 and 2027 revenue estimates to reflect ongoing advanced packaging capacity opportunities, including CoPoS and CoWoS, and also CPO. We expect AllRing to win more advanced packaging opportunities from the non-TSMC camp. Gross margin is likely to be better than many fear, driven by a better product mix, as CPO equipment's GM is margin accretive at the 55-60% level.

Exhibit 18: AllRing: Estimate revisions

Source: Morgan Stanley Research (e) estimates

Exhibit 19: AllRing: Quarterly financials

Source: Company data, Morgan Stanley Research (E) estimates

We raise our RI-based price target (base case scenario value) from NT$450 to NT$688: We factor in our estimate changes for 2026 and 2027 and newly introduced estimates for 2028, and also roll our RI base forward to 2026, from 2025. Our other key assumptions are unchanged, including cost of equity (CoE) of 8%, derived from a risk-free rate of 2.0%, beta of 1.0 and market risk premium of 6.0%; an intermediate growth rate of 12.4%; and terminal value of 3%.

Our bull and bear case values rise from NT$630 to NT$970 and NT$180 to NT$280, respectively, implying 45x and 13x 2026e EPS.

Source: Company data, Morgan Stanley Research (E) estimates

CoWoS Capacity Expansion and CPO Fuel Growth; OW

Base case, residual income model. Our key assumptions are:

• Cost of equity (CoE) of 8%, derived from a risk-free rate of 2.0%, beta of 1.0 and a market risk premium of 6.0%.

• Intermediate growth rate of 15.3%.

• Terminal growth rate of 4.5%.

Key: — Historical Stock Performance ● Current Stock Price ◆ Price Target

Source: Refinitiv, Morgan Stanley Research

We assume: 1) 35% revenue CAGR over 2024-27, thanks to stronger-than-expected capacity expansion plans for TSMC's CoWoS; 2) gross margin expands to more than 55% in 2024-27, due to stronger-than-expected operating leverage resulting from the equipment production outsourcing strategy; 3) new product launches exceed expectations and AllRing increases its market share in other processes such as "CoW".

We expect: 1) 23% revenue CAGR in 2024-27 after strong CoWoS capacity expansion for TSMC in 2024-25; 2) gross margin to be maintained at 48-53% from 2024-27 vs. 48% in 2022 thanks to high operating leverage resulting from the equipment production outsourcing strategy; and 3) with new equipment launches, we expect AllRing to expand its market share in other processes. CPO and other applications continue to bear fruit.

■ NVIDIA's next-generation GPU Rubin will stick with graphene for heat sink attachment, which reduces concerns on competition for Allring.

■ We expect AllRing to be a key beneficiary of the multi-year advanced packaging trend, including SoW-X, CoPoS and SolC, but large-scale revenue contributions will likely take time.

■ With bad news already priced in and upward revision in CoWoS capacity, we expect AllRing to sustain growth in 2026-27.

■ We expect the stock to trade at 32x our 2026 EPS estimate, similar to 2 S.D. above its historical average, as we expect new advanced packaging opportunities including CPO, CoWoS and CoPoS to benefit the company in the long run.

Source: Refinitiv, Morgan Stanley Research

Secular Growth: Positive Technology Diffusion: Positive

View descriptions of Risk Rewards Themes here

We assume: 1) -30% revenue CAGR in 2024-27, owing to slower-than-expected CoWoS capacity expansion plans by TSMC; 2) gross margin falls below 40% owing to the high bargaining power of its customers; 3) market share loss in "WoS" process amid rising peer competition.

• New technology migration such as SolC and silicon photonics

• Strong operating leverage, given equipment production outsourcing strategy

100% APAC, ex Japan, Mainland China and India

Source: Morgan Stanley Research Estimate View explanation of regional hierarchies here

• Stronger-than-expected CoWoS capacity expansion

• Faster-than-expected adoption of silicon photonics

• Market share gains from peers, led by continuous R&D and China's localization trend

• Slowdown in CoWoS capacity expansion

• Allring's "WoS" market share declining

• Slowdown in technology migration such as SoIC and silicon photonics adoption

Source: Refinitiv, Morgan Stanley Research

Note: There are not sufficient brokers supplying consensus data for this metric

Source: Refinitiv, Morgan Stanley Research

E = Morgan Stanley Research Estimates

Source: Morgan Stanley Research, Company Data

As indicated in Looking into the AI Datacenter Networking Market; key charts for CPO developments section, NVIDIA is likely to launch new scale-out models in GTC 2026. For 2026, we expect Spectrum-X to be the mainstream scale-out CPO switch product, with the optical engine production number being 600k-1mn units, and total switch shipments being ~20k units given production lead time. We expect TFC, FOCI and Senko to be the major FAU suppliers, with FOCI taking around 40% of total market shares. Thus, we expect NVIDIA's contribution to FOCI's total revenue to be around 18% in 2026.

Looking into 2027, our industry checks suggest that NVIDIA could introduce a CPO version of the Kyber alternative rack, leveraging two NVL72 architecture, with the launch time likely in 2H27. As a result, we assume 5k and 28k units of CPO Rubin Ultra rack shipments in 2027 and 2028, respectively, with FOCI having around 40% market share. Thus, we expect NVIDIA's total revenue to contribute around 42% and 80% of FOCI's total revenue in 2027 and 2028.

However, there are more than 10 customers that are doing prototypes and could recognize NRE revenue in 2026, with mass production starting in 2027 and 2028, according to our supply chain checks. We have not fully baked this into our model yet, and view it as only a bull case scenario as we await additional product launch and shipment signs in the supply chain.

Exhibit 22: Bottom-up analysis: NVIDIA's revenue contribution to FOCI in 2026e-2028e

Source: Company data, Morgan Stanley Research (e) estimates

Exhibit 23: Spectrum-X CPO Switch

Exhibit 24: Quantum-X CPO Switch

For Optical I/O or optical engine on interposer adoption plan, we see NVIDIA exploring multiple solutions for the next generation Feyman and Feyman Ultra's platform, but whether there would be introduction or not still takes time. For other solutions, we see AMD targeting a scale-up CPO version for its MI550 or MI650 rack system in 2027/2028, based on its development progress. We also see ASIC vendors exploring CPO on GPU.

Our supply chain checks suggest that FOCI would see over 10 customers contributing to its Silicon Photonic/CPO revenue in 2026, besides some mass production revenue starting from 1Q26.

In its recent public offering disclosure for the SEO, FOCI disclosed its projected revenue and profitability for the NT$1.538bn capex spent (Exhibit 25). The company completed its SEO plan in early February, with total acquired capital of NT$3.1bn. It also announced a new private equity plan on Feb 26, with maximum new share issuance of 10mn units. We expect it to utilize all of the capital for R&D purposes (12.8T research) and further building and expanding its FAU capacity.

The company plans an on-site earnings call on March 11 at 2pm HKT. We expect management to share more details regarding SiPh/CPO developments.

Exhibit 25: FOCI's projected revenue and profitability from NT$1.538bn capex spent

Source: Company data, Morgan Stanley Research

Exhibit 27: FOCI's solutions for CPO

Exhibit 28: ReLFACon enables direct transmission of external photonic signals with MCM modules to achieve reliable signal transmission

We factor in actual 2025 results, cut 2026 and 2027 earnings by 74% and 40%, and introduce 2028e estimates, respectively: The weaker than expected 4Q25 results were mainly due to the ongoing capacity transition from the Hsinchu fab to the new Thailand fab, as the company prioritizes its SiPh/CPO development. We expect this impact to persist into 1H26. We also slightly lower our 2027 CPO revenue assumptions from NVIDIA, as NVIDIA has modestly postponed its CPO on GPU plan and introduced a NVLink switch/NIC CPO version instead, resulting in lower content than previously assumed. These revised assumptions are now reflected in our model, including the new 2028 estimates.

However, supply chain checks indicate more than 10 customers are currently prototyping, with potential NRE revenue recognition in 2026 and mass production beginning in 2027 and 2028. We have not fully incorporated this upside into our model yet as we look for clearer evidence of product launches and shipment momentum – developments that would push our outlook toward the bull case scenario.

We also reflect the one-off expense from new share issuance impact on gross and operating profit in 2026, as well as the impact on EPS from new share issuance.

Source: Company data, Morgan Stanley Research (E) estimates

Exhibit 30: FOCI: Quarterly financials

Source: Company data, Morgan Stanley Research (e) estimates

We increase our price target (base case scenario) from NT$450 to NT$708. This reflects our 2025–27 estimate revisions, the introduction of 2028 forecasts, and rolling our RI base forward to 2026 from 2025 — the primary driver of our higher price target.

We slightly lower our 2027 CPO revenue assumptions from NVIDIA, as NVIDIA has modestly postponed its CPO on GPU plan and introduced a NVLink switch/NIC CPO version instead, resulting in lower content vs. prior expectations. However, we incorporate updated 2028 assumptions, including contribution from NVIDIA's scale out switch and scale up consumption. Supply chain checks indicate more than 10 customers are currently prototyping and could recognize NRE revenue in 2026, with mass production starting in 2027 and 2028. We have not fully reflected this potential upside in our model yet, as we look for clearer signs of product launches and shipment momentum – developments that would push us toward our bull case price target of NT$920.

We also raise our mid term growth rate from 12% to 16%, in line with Asia optics peers (e.g. Landmark) at 15-16%, to reflect rapid development in CPO and silicon photonics technology. We maintain our other assumptions: cost of equity of 12.9% (risk free rate 2.0%, risk premium 6.9%), payout ratio of 61%, and terminal growth rate of 4.0%.

We raise our bull and bear case values from NT$780 to NT$920 and NT$280 to NT$330, implying 55x and 20x 2027e EPS, respectively.

Source: Company data, Morgan Stanley Research (E) estimates

Source: Company data, Factset, Morgan Stanley Research estimates

Source: Company data, Factset, Morgan Stanley Research estimates

Base case, residual income model. Key assumptions: cost of equity of 12.9% (risk-free rate 2.0% and risk premium 6.9%), a payout ratio of 61%, a medium-term growth rate of 16%, and a terminal growth rate of 4.0%.

Key: — Historical Stock Performance ● Current Stock Price ◆ Price Target

Source: Refinitiv, Morgan Stanley Research

☑ Co-packaged optics (CPO) is an advanced silicon photonics (SiPh) method that integrates optics and silicon on a single packaged substrate, aiming to reduce signal loss, power consumption, and cost.

■ Given geopolitical benefits and close partnership with TSMC, we think FOCI is well positioned to provide CPO's fiber array units (FAU) and related services, leveraging its ReLFACon(TM) technology.

■ SiPh/CPO technology is expected to be FOCI's main growth driver in the future. We see its sales contribution expanding from 7% in 2024 to 80% in 2028e.

The company is still growing at a fast pace and in transition. Thus, we expect FOCI stock to trade to 43x 2027e P/E vs. an EPS CAGR of 233%, 2023-27e.

View descriptions of Risk Rewards Themes here

We assume: 1) 200% revenue CAGR, 2023-27e, from stronger-than-expected CPO development; 2) faster-than-expected CPO introduction; 3) gross margin improving to over 50% in 2027e thanks to scale effects and the company's technology competitiveness; and 4) >400% EPS CAGR, 2023-27e.

We project: 1) 89% revenue CAGR, 2023-27e, on fast CPO development; 2) gross margin improves to over 36% in 2027e thanks to scale effects and the company's technology competitiveness; 3) 233% EPS CAGR, 2023-27e.

We assume: 1) 50% revenue CAGR, 2023-27e, given slower-than-expected CPO development; 2) slower-than-expected CPO introduction; and 3) gross margin falling below 10% in 2027 owing to the high bargaining power of customers.

• Silicon photonics and CPO

Source: Morgan Stanley Research Estimate View explanation of regional hierarchies here

Note: There are not sufficient brokers supplying consensus data for this metric

• Slower-than-expected CPO introduction slower than expected

Note: There are not sufficient brokers supplying consensus data for this metric

Note: There are not sufficient brokers supplying consensus data for this metric

Note: There are not sufficient brokers supplying consensus data for this metric

(%) Note: There are not sufficient brokers supplying consensus data for this metric

Source: Refinitiv, Morgan Stanley Research

E = Morgan Stanley Research Estimates

Source: Morgan Stanley Research, Company Data

• View explanation of Options Probabilities methodology - Options_Probabilities_Exhibit_Link.pdf

• View descriptions of Risk Rewards Themes - RR_Themes_Exhibit_Link.pdf

• View explanation of regional hierarchies - GEG_Exhibit_Link.pdf

• View explanation of Theme/Exposure methodology - ESG_Sustainable_Solutions_External_Link.pdf

• View explanation of HERS methodology - ESG_HERS_External_Link.pdf

For important disclosures, stock price charts and equity rating histories regarding companies that are the subject of this report, please see the Morgan Stanley Research Disclosure Website at www.morganstanley.com/eqr/disclosures/webapp/generalresearch, or contact your investment representative or Morgan Stanley Research at 1585 Broadway, (Attention: Research Management), New York, NY 10036 USA.

The following analysts hereby certify that their views about the companies and their securities discussed in this report are accurately expressed and that they have not received and will not receive direct or indirect compensation in exchange for expressing specific recommendations or views in this report: Charlie Chan; Daisy Dai, CFA; Andy Meng, CFA; Derrick Yang; Tiffany Yeh; Daniel Yen, CFA.

Morgan Stanley Research has been published in accordance with our conflict management policy, which is available at www.morganstanley.com/institutional/research/conflictpolicies. A Portuguese version of the policy can be found at www.morganstanley.com.br

The analyst or strategist (or a household member) identified below owns the following securities (or related derivatives): Andy Meng, CFA - Alphabet Inc.(common or preferred stock).

As of January 30, 2026, Morgan Stanley beneficially owned 1% or more of a class of common equity securities of the following companies covered in Morgan Stanley Research: ACM Research Inc, Advanced Micro-Fabrication Equipment Inc, Advanced Wireless Semiconductor Co, Alchip Technologies Ltd, AllRing Tech Co., Alphabet Inc., AP Memory Technology Corp, ASE Technology Holding Co. Ltd., ASMedia Technology Inc, ASMPT Ltd, Cadence Design Systems Inc, Chroma Ate Inc., Coherent Corp, Giga-Byte Technology Co. Ltd., Global Unichip Corp, Intel Corporation, JCET Group Co Ltd, Keysight Technologies Inc, King Yuan Electronics Co Ltd, LandMark Optoelectronics Corporation, Lumentum Holdings Inc, Macronix International Co Ltd, Marvell Technology Group Ltd, Meta Platforms Inc, Micron Technology Inc., Nanya Technology Corp., Novatek, Parade Technologies Ltd, Phison Electronics Corp, Powerchip Semiconductor Manufacturing Co, Senko Group Holdings, Shanghai Fudan Microelectronics, Siemens, Silergy Corp., Silicon Motion, SK hynix, Suzhou TFC Optical Communication Co Ltd., Teradyne Inc, Tesla Inc, TSMC, UMC, Vanguard International Semiconductor, Visual Photonics Epitaxy Co Ltd, WIN Semiconductors Corp, Winbond Electronics Corp, WPG Holdings, WT Microelectronics Co. Ltd.

Within the last 12 months, Morgan Stanley managed or co-managed a public offering (or 144A offering) of securities of Alphabet Inc., Intel Corporation, Lumentum Holdings Inc, Meta Platforms Inc, Micron Technology Inc., Montage Technology Co Ltd, Senko Group Holdings, Siemens, SK hynix, Sumitomo Corp., Wistron Corporation.

Within the last 12 months, Morgan Stanley has received compensation for investment banking services from Alphabet Inc., Amkor Technology Inc., ASMPT Ltd, Coherent Corp, Intel Corporation, Lumentum Holdings Inc, Meta Platforms Inc, Micron Technology Inc., Montage Technology Co Ltd, Senko Group Holdings, Siemens, SK hynix, Sumitomo Corp., Wistron Corporation.

Lumentum Holdings Inc, Meta Platforms Inc, Micron Technology Inc., Montage Technology Co Ltd, Senko Group Holdings, Siemens, SK hynix, Sumitomo Corp., Wistron Corporation.

In the next 3 months, Morgan Stanley expects to receive or intends to seek compensation for investment banking services from Alchip Technologies Ltd, Alphabet Inc., Amkor Technology Inc, Andes Technology Corp, AP Memory Technology Corp, ASE Technology Holding Co. Ltd., ASMedia Technology Inc, ASMPT Ltd, Cadence Design Systems Inc, Coherent Corp, Espressif Systems, Giga-Byte Technology Co. Ltd., GlobalFoundries Inc, GlobalWafers Co Ltd, Gudeng Precision, Himax Technologies Inc, Intel Corporation, Keysight Technologies Inc, King Yuan Electronics Co Ltd, Lumentum Holdings Inc, Macronix International Co Ltd, MediaTek, Meta Platforms Inc, Micron Technology Inc., Montage Technology Co Ltd, Novatek, Phison Electronics Corp, Powerchip Semiconductor Manufacturing Co, Quanta Computer Inc., Realtek Semiconductor, Samsung Electronics, Senko Group Holdings, Siemens, Silergy Corp., Silicon Motion, SK hynix, Sumitomo Corp., Suzhou TFC Optical Communication Co Ltd., Teradyne Inc, Tesla Inc, TSMC, UMC, Universal Scientific Ind. (Shanghai), Vanguard International Semiconductor, Winbond Electronics Corp, Wistron Corporation, WPG Holdings, WT Microelectronics Co. Ltd.

Within the last 12 months, Morgan Stanley has received compensation for products and services other than investment banking services from Alphabet Inc., Amkor Technology Inc., ASE Technology Holding Co. Ltd., Cadence Design Systems Inc., Coherent Corp., Foxconn Technology, Giga-Byte Technology Co. Ltd., GlobalFoundries Inc, Intel Corporation, King Yuan Electronics Co Ltd, Marvell Technology Group Ltd, MediaTek, Meta Platforms Inc, Micron Technology Inc., Nanya Technology Corp., Novatek, Nuvoton Technology Corporation, Quanta Computer Inc., Realtek Semiconductor, Siemens, Silicon Motion, SMIC, Sumitomo Corp., Tesla Inc, TSMC, UMC, Universal Scientific Ind. (Shanghai), Winbond Electronics Corp, WT Microelectronics Co. Ltd.. Within the last 12 months, Morgan Stanley has provided or is providing investment banking services to, or has an investment banking client relationship with, the following company: Alchip Technologies Ltd, Alphabet Inc., Amkor Technology Inc, Andes Technology Corp, AP Memory Technology Corp, ASE Technology Holding Co. Ltd., ASMedia Technology Inc, ASMPT Ltd, Cadence Design Systems Inc, Coherent Corp, Espressif Systems, Giga-Byte Technology Co. Ltd., GlobalFoundries Inc, GlobalWafers Co Ltd, Gudeng Precision, Himax Technologies Inc, Intel Corporation, Keysight Technologies Inc, King Yuan Electronics Co Ltd, Lumentum Holdings Inc, Macronix International Co Ltd, MediaTek, Meta Platforms Inc, Micron Technology Inc., Montage Technology Co Ltd, Novatek, Phison Electronics Corp, Powerchip Semiconductor Manufacturing Co, Quanta Computer Inc., Realtek Semiconductor, Samsung Electronics, Senko Group Holdings, Siemens, Silergy Corp., Silicon Motion, SK hynix, Sumitomo Corp., Suzhou TFC Optical Communication Co Ltd., Teradyne Inc, Tesla Inc, TSMC, UMC, Universal Scientific Ind. (Shanghai), Vanguard International Semiconductor, Winbond Electronics Corp, Wistron Corporation, WPG Holdings, WT Microelectronics Co. Ltd..

Within the last 12 months, Morgan Stanley has either provided or is providing non-investment banking, securities-related services to and/or in the past has entered into an agreement to provide services or has a client relationship with the following company: Alphabet Inc., Amkor Technology Inc, ASE Technology Holding Co. Ltd., Cadence Design Systems Inc, Coherent Corp, Foxconn Technology, Giga-Byte Technology Co. Ltd., GlobalFoundries Inc, Intel Corporation, King Yuan Electronics Co Ltd, Marvell Technology Group Ltd, MediaTek, Meta Platforms Inc, Micron Technology Inc., Montage Technology Co Ltd, Nanya Technology Corp., Novatek, Nuvoton Technology Corporation, Quanta Computer Inc., Realtek Semiconductor, Siemens, Silicon Motion, SMIC, Sumitomo Corp., Tesla Inc, TSMC, UMC, Universal Scientific Ind. (Shanghai), Winbond Electronics Corp, WT Microelectronics Co. Ltd.

Morgan Stanley & Co. LLC makes a market in the securities of ACM Research Inc, Himax Technologies Inc, Keysight Technologies Inc, Silicon Motion.

Morgan Stanley and its affiliates do business that relates to companies/instruments covered in Morgan Stanley Research, including market making, providing liquidity, fund management, commercial banking, extension of credit, investment services and investment banking. Morgan Stanley sells to and buys from customers the securities/instruments of companies covered in Morgan Stanley Research on a principal basis. Morgan Stanley may have a position in the debt of the Company or instruments discussed in this report. Morgan Stanley trades or may trade as principal in the debt securities (or in related derivatives) that are the subject of the debt research report.

Certain disclosures listed above are also for compliance with applicable regulations in non-US jurisdictions.

Morgan Stanley uses a relative rating system using terms such as Overweight, Equal-weight, Not-Rated or Underweight (see definitions below). Morgan Stanley does not assign ratings of Buy, Hold or Sell to the stocks we cover. Overweight, Equal-weight, Not-Rated and Underweight are not the equivalent of buy, hold and sell. Investors should carefully read the definitions of all ratings used in Morgan Stanley Research. In addition, since Morgan Stanley Research contains more complete information concerning the analyst's views, investors should carefully read Morgan Stanley Research, in its entirety, and not infer the contents from the rating alone. In any case, ratings (or research) should not be used or relied upon as investment advice. An investor's decision to buy or sell a stock should depend on individual circumstances (such as the investor's existing holdings) and other considerations.

(as of February 28, 2026)

ed below apply to Morgan Stanley's Fundamental Equity Research and do not apply to Debt Research produced by the Firm.

For disclosure purposes only (in accordance with FINRA requirements), we include the category headings of Buy, Hold, and Sell alongside our ratings of Overweight, Equal-weight, Not-Rated and Underweight. Morgan Stanley does not assign ratings of Buy, Hold or Sell to the stocks we cover. Overweight, Equal-weight, Not-Rated and Underweight are not the equivalent of buy, hold, and sell but represent recommended relative weightings (see definitions below). To satisfy regulatory requirements, we correspond Overweight, our most positive stock rating, with a buy recommendation: we correspond Equal-weight and Not-Rated to hold and Underweight to sell recommendations, respectively.

Data include common stock and ADRs currently assigned ratings. Investment Banking Clients are companies from whom Morgan Stanley received investment banking compensation in the last 12 months. Due to rounding off of decimals, the percentages provided in the "% of total" column may not add up to exactly 100 percent.

Overweight (O). The stock's total return is expected to exceed the average total return of the analyst's industry (or industry team's) coverage universe, on a risk-adjusted basis, over the next 12-18 months.

Equal-weight (E). The stock's total return is expected to be in line with the average total return of the analyst's industry (or industry team's) coverage universe, on a risk-adjusted basis, over the next 12-18 months.

Not-Rated (NR). Currently the analyst does not have adequate conviction about the stock's total return relative to the average total return of the analyst's industry (or industry team's) coverage universe, on a risk-adjusted basis, over the next 12-18 months.

Underweight (U). The stock's total return is expected to be below the average total return of the analyst's industry (or industry team's) coverage universe, on a risk-adjusted basis, over the next 12-18 months.

Unless otherwise specified, the time frame for price targets included in Morgan Stanley Research is 12 to 18 months.

Attractive (A): The analyst expects the performance of his or her industry coverage universe over the next 12-18 months to be attractive vs. the relevant broad market benchmark, as indicated below.

In-Line (I): The analyst expects the performance of his or her industry coverage universe over the next 12-18 months to be in line with the relevant broad market benchmark, as indicated below.

Cautious (C): The analyst views the performance of his or her industry coverage universe over the next 12-18 months with caution vs. the relevant broad market benchmark, as indicated below.

Benchmarks for each region are as follows: North America - S&P 500; Latin America - relevant MSCI country index or MSCI Latin America Index; Europe - MSCI Europe; Japan - TOPIX; Asia -

relevant MSCI country index or MSCI sub-regional index or MSCI AC Asia Pacific ex Japan Index.

Samsung Electronics (005930.KS) - As of 03/03/26 GMT in KRW Industry: S. Korea Technology

Stock Rating History: 3/1/21 : 0/A; 7/19/21 : 0/I; 8/12/21 : 0/C; 10/4/22 : 0/A; 7/21/24 : 0/I; 9/15/24 : 0/C; 6/13/25 : 0/I; 9/21/25 : 0/A

Price Target History: 2/25/21 : 115000; 5/18/21 : 93000; 6/8/21 : 98000; 8/12/21 : 89000; 9/15/21 : 95000; 12/3/21 : 97000;

3/18/22 : 95000; 4/28/22 : 85000; 6/10/22 : 80000; 7/5/22 : 75000; 7/22/22 : 70000; 9/17/22 : 68000; 3/21/23 : 70000;

5/30/23 : 90000; 7/7/23 : 95000; 3/22/24 : 97000; 4/16/24 : 101000; 6/6/24 : 105000; 9/15/24 : 76000; 12/18/24 : 65000;

3/19/25 : 70000; 8/1/25 : 86000; 9/21/25 : 97000; 10/8/25 : 111000; 10/23/25 : 120000; 10/30/25 : 139000; 11/5/25 : 144000; 1/8/26 : 170000; 1/30/26 : 210000; 2/24/26 : 248000

Source: Morgan Stanley Research Date Format: MM/DD/YY Price Target No Price Target Assigned (NA)

Stock Price (Not Covered by Current Analyst) Stock Price (Covered by Current Analyst)

Stock Ratings: Overweight (0) Equal-weight (E) Underweight (U) Not-Rated (NR) No Rating Available (NA)

Effective January 13, 2014, the stocks covered by Morgan Stanley Asia Pacific will be rated relative to the analyst's industry (or industry team's) coverage.

Effective January 13, 2014, the industry view benchmarks for Morgan Stanley Asia Pacific are as follows: relevant MSCI country index or MSCI sub-regional index or MSCI AC Asia Pacific ex Japan Index.

Samsung Electronics (005935.KS) - As of 03/03/26 GMT in KRW Industry: S. Korea Technology

Stock Rating History: 3/1/21 : 0/A; 7/19/21 : 0/I; 8/12/21 : 0/C; 10/4/22 : 0/A; 7/21/24 : 0/I; 9/15/24 : 0/C; 6/13/25 : 0/I; 9/21/25 : 0/A

Price Target History: 1/12/21 : 90000; 5/18/21 : 80000; 6/8/21 : 84000; 8/12/21 : 77000; 9/15/21 : 82000; 12/3/21 : 83000;

3/18/22 : 82000; 4/28/22 : 73000; 6/10/22 : 69000; 7/5/22 : 65000; 7/22/22 : 60000; 9/17/22 : 58000; 3/21/23 : 60000;

5/30/23 : 77000; 7/7/23 : 82000; 1/20/25 : 52000; 3/19/25 : 60900; 8/1/25 : 74820; 9/21/25 : 84390; 10/6/25 : 96570;

10/23/25 : 102000; 10/30/25 : 120930; 11/5/25 : 125280; 1/8/26 : 147900; 1/30/26 : 182000; 2/24/26 : 210000

Source: Morgan Stanley Research Date Format : MM/DD/YY Price Target = No Price Target Assigned (NA)

Source: Morgan Stanley Research Date Format: MM/DD/YY Price Target No Price Target Assigned (NA)

Stock Price (Not Covered by Current Analyst) Stock Price (Covered by Current Analyst)

Stock and Industry Ratings (abbreviations below) appear as Stock Rating/Industry View

Stock Ratings: Overweight (0) Equal-weight (E) Underweight (U) Not-Rated (NR) No Rating Available (NA)

Industry View: Attractive (A) In-line (I) Cautious (C) No Rating (NR)

Effective January 13, 2014, the stocks covered by Morgan Stanley Asia Pacific will be rated relative to the analyst's industry industry team's coverage.

Effective January 13, 2014, the industry view benchmarks for Morgan Stanley Asia Pacific are as follows: relevant MSCI country index or MSCI sub-regional index or MSCI AC Asia Pacific ex Japan Index.

SK hynix (000660.KS) - As of 03/03/26 GMT in KRW Industry: S. Korea Technology

7/21/24 : 0/1; 9/15/24 : U/C; 3/19/25 : E/C; 6/13/25 : E/I; 9/21/25 : 0/A

Price Target History: 2/25/21 : 174000; 5/18/21 : 146000; 6/8/21 : 156000; 8/12/21 : 80000; 9/15/21 : 88000; 12/3/21 : 110000;

12/23/21：125000；1/24/22：130000；1/28/22：136000；2/11/22：155000；3/18/22：150000；4/27/22：130000；6/10/22：120000；

7/5/22 : 110000; 7/22/22 : 105000; 10/4/22 : 130000; 12/7/22 : 120000; 3/21/23 : 110000; 5/30/23 : 140000; 7/7/23 : 170000;

9/21/23 : 210000; 3/22/24 : 230000; 4/16/24 : 270000; 6/6/24 : 300000; 7/25/24 : 260000; 9/15/24 : 120000; 10/24/24 : 130000;

12/18/24 : 135000; 1/20/25 : 150000; 3/19/25 : 230000; 4/7/25 : 210000; 6/13/25 : 250000; 7/24/25 : 260000; 9/21/25 : 410000;

10/8/25 : 480000 ; 10/23/25 : 570000 ; 10/29/25 : 630000 ; 11/5/25 : 730000 ; 1/2/26 : 840000 ; 1/30/26 : 1100000

Source: Morgan Stanley Research Date Format: MM/DD/YY Price Target = No Price Target Assigned (NA)

Stock Price (Not Covered by Current Analyst) — Stock Price (Covered by Current Analyst)

Stock Price (Not Lovered by Current Analyst) — Stock Price (Lovered by Current Analyst)

Stock and Industry Ratings (abbreviations below) appear as + Stock Rating/Industry View

Stock Ratings: Overweight (0) Equal-weight (E) Underweight (U) Not-Rated (NR) No Rating Available (NA)

Industry View: Attractive (A) In-line (I) Cautious (C) No Rating (NR)

Effective January 13, 2014, the stocks covered by Morgan Stanley Asia Pacific will be rated relative to the analyst's industry (or industry team's) coverage.

Effective January 13, 2014, the industry view benchmarks for Morgan Stanley Asia Pacific are as follows: relevant MSCI country index or MSCI sub-regional index or MSCI AC Asia Pacific ex Japan Index.