External broker reports(aggregated public sources), CMB International

CMB International

External broker reports(aggregated public sources)

Incorporated in Hong Kong, supported by the mainland and targeting in global market, CMB International Capital Corporation Limited (“CMBI”) is an integrated financial institution providing comprehensive and professional services. As a wholly-owned subsidiary of China Merchants Bank ("CMB"), by leveraging on the Group’s synergy and coordination in domestic and overseas markets, as well as CMB’s strong resources and outstanding social reputation, CMBI and its subsidiaries have been actively implementing diversified business strategies and set up an overall layout of main business segments including Corporate Finance, Asset Management, Wealth Management, Global Markets and Structured Finance, etc. Adhering to the philosophy of “being clients oriented and increasing our own values by the way of creating value for clients”, CMBI endeavors to provide quality and integrated financial services to institutional, corporate and individual customers both domestic and overseas including Hong Kong IPO sponsorship and underwriting, share placement and rights issue for listed companies, financial advisors, bonds issuance services, asset management, direct investment, wealth management, equity sales and trading, etc.

Thematic (Sector/Industry)

<ul><li>We noticed SemiAnalysis published an article on Jun 9 discussing potential deployment delays in 1) co-packaged optics (CPO) and 2) Nvidia’s native single-ended 800VDC architecture, which has triggered market debate and put pressure on AI infrastructure-related names, particularly optical, power semiconductor and datacom interconnect suppliers.</li><li>We think the note is better interpreted as a timing reset rather than a demand reset.</li><li>On optics, its conclusion is broadly consistent with our prior view that mass CPO adoption is unlikely before late-2027 at the earliest (see our prev. note), while pluggable optical modules and NPO should remain key deployment paths for the foreseeable future, supporting leading domestic optical module suppliers including Innolight (300308 CH, BUY) and Eoptolink (300502 CH, NR).</li></ul>Content is external broker report sourced from online content aggregator through publicly available sources and is displayed below for general informational purposes only. Refer full disclaimer below.

We noticed SemiAnalysis published an article on Jun 9 discussing potential deployment delays in 1) co-packaged optics (CPO) and 2) Nvidia’s native single-ended 800VDC architecture, which has triggered market debate and put pressure on AI infrastructure-related names, particularly optical, power semiconductor and datacom interconnect suppliers.We think the note is better interpreted as a timing reset rather than a demand reset.On optics, its conclusion is broadly consistent with our prior view that mass CPO adoption is unlikely before late-2027 at the earliest (see our prev. note), while pluggable optical modules and NPO should remain key deployment paths for the foreseeable future, supporting leading domestic optical module suppliers including Innolight (300308 CH, BUY) and Eoptolink (300502 CH, NR).Content is external broker report sourced from online content aggregator through publicly available sources and is displayed below for general informational purposes only. Refer full disclaimer below.

Semiconductors - A quick read on recently reported CPO and 800VDC deployment delay

Equity & Credit Research

Duncan Chan

<ul>
<li>AI investment has shifted to infrastructure buildout, where growth is driven by rising content per server rather than unit volumes, supporting more stable system-level demand.</li>
<li>Value is concentrated at the server board level, with interconnect offering standardised, architecture-linked exposure, while PCB provides direct but execution-dependent leverage to system complexity.</li>
<li>Earnings visibility depends on alignment between hyperscaler capex and order momentum; divergence signals a shift from expansion to a more volatile digestion phase.</li>
</ul>


AI investment has shifted to infrastructure buildout, where growth is driven by rising content per server rather than unit volumes, supporting more stable system-level demand.
Value is concentrated at the server board level, with interconnect offering standardised, architecture-linked exposure, while PCB provides direct but execution-dependent leverage to system complexity.
Earnings visibility depends on alignment between hyperscaler capex and order momentum; divergence signals a shift from expansion to a more volatile digestion phase.


AI Infrastructure: Value Shifts to the Server Board Level

Asia Pac ECM and Index Rebal, Shenton Research

Ke Yan, CFA, FRM

Asia Pac ECM and Index Rebal

Asia Pac IPO & Placement Research, China Biotech

Shenton Research

Shenton Research is an independent Asia Pacific equity specialist. Our quantitative research covers index rebalancing flows, alongside exchange data analytics accross Asia Pacific. On the equity capital markets side, we focus on Asia Pacific ECM, especially biotech and TMT. Our lead analyst is frequently cited by international media.

Index Rebalance

<ul>
<li>HSIL adopted all six proposed HS TECH changes, five unchanged and one with modifications, and the new rules take effect on 7 Dec 2026.</li>
<li>Selection by market value and sales growth was adopted with modifications, adding a turnover floor for non-constituents and a minimum sales test for sales growth candidates.</li>
<li>The velocity test, innovation screening, weighting, caps and quarterly review are retained from the current rules.</li>
</ul>


HSIL adopted all six proposed HS TECH changes, five unchanged and one with modifications, and the new rules take effect on 7 Dec 2026.
Selection by market value and sales growth was adopted with modifications, adding a turnover floor for non-constituents and a minimum sales test for sales growth candidates.
The velocity test, innovation screening, weighting, caps and quarterly review are retained from the current rules.


HS TECH Consultation Conclusions: Retained and Refined Rules

<ul><li>On Day 1 of Google Cloud Next 2026, held on April 22 in Las Vegas, delivered three key infrastructure updates: 1) Launched two 8 th gen TPUs (TPU 8t and TPU 8i) with dedicated designs for training and inference, respectively, highlighting a clearer split between compute-intensive training and latency- and memorysensitive inference workloads;</li><li>2) introduced a broader AI Hypercomputer architecture built around Virgo networking, tighter storage integration, and largescale cluster orchestration, reinforcing that AI performance is increasingly defined at the system level rather than by standalone chip capability;</li><li>3) framed agentic AI as the next major demand driver, which raises the importance of memory bandwidth, interconnect scale, utilization, and power efficiency alongside raw silicon performance.</li></ul>Content is external broker report sourced from online content aggregator through publicly available sources and is displayed below for general informational purposes only. Refer full disclaimer below.

On Day 1 of Google Cloud Next 2026, held on April 22 in Las Vegas, delivered three key infrastructure updates: 1) Launched two 8 th gen TPUs (TPU 8t and TPU 8i) with dedicated designs for training and inference, respectively, highlighting a clearer split between compute-intensive training and latency- and memorysensitive inference workloads;2) introduced a broader AI Hypercomputer architecture built around Virgo networking, tighter storage integration, and largescale cluster orchestration, reinforcing that AI performance is increasingly defined at the system level rather than by standalone chip capability;3) framed agentic AI as the next major demand driver, which raises the importance of memory bandwidth, interconnect scale, utilization, and power efficiency alongside raw silicon performance.Content is external broker report sourced from online content aggregator through publicly available sources and is displayed below for general informational purposes only. Refer full disclaimer below.

Semiconductors - Google Cloud Next: TPU v8 and rising of AI Hypercomputer

AI and Energy Infrastructure, Explained Plainly, The Physical Layer

The Physical Layer

AI and Energy Infrastructure, Explained Plainly

A journal on the physical constraints behind the AI buildout: power, grid connection, data centres, cooling and the supply chains that gate them. Written to be understood by anyone.

Equity Bottom-Up

<ul><li>An AI processor needs electricity at less than one volt, but power reaches the cabinet at hundreds of volts.</li><li>Texas Instruments makes many of the small, cheap chips that step it down and keep it safe on the way.</li><li>It is spending tens of billions of dollars on its own American factories to make those chips more cheaply than anyone who rents factory space.</li></ul>

An AI processor needs electricity at less than one volt, but power reaches the cabinet at hundreds of volts.Texas Instruments makes many of the small, cheap chips that step it down and keep it safe on the way.It is spending tens of billions of dollars on its own American factories to make those chips more cheaply than anyone who rents factory space.

The most expensive chip in an AI rack runs on Texas Instruments' cheapest ones

Hong Kong

Innoscience is the largest 8-inch Integrated Device Manufacture (IDM) fully focused on GaN technology in the world.

Price Gauge
52-week Range	0.00 - 0.00
Market Cap (USDm)	0.00
P/E	0.00
Dividend (Yield)	0.00

InnoScience Suzhou Technology (2577 HK)

Semiconductors - A quick read on recently reported CPO and 800VDC deployment delay

We noticed SemiAnalysis published an article on Jun 9 discussing potential deployment delays in 1) co-packaged optics (CPO) and 2) Nvidia’s native...

CMB International

AI Infrastructure: Value Shifts to the Server Board Level

Re-sharing as recent AI infrastructure spending and deployment trends continue to reinforce the discussion around where value capture is becoming...

Duncan Chan

HS TECH Consultation Conclusions: Retained and Refined Rules

HSIL adopted all six HS TECH proposals and added two new screens to constituent selection. We set out which rules are retained and which are refined.

Ke Yan, CFA, FRM

Semiconductors - Google Cloud Next: TPU v8 and rising of AI Hypercomputer

On Day 1 of Google Cloud Next 2026, held on April 22 in Las Vegas, delivered three key infrastructure updates: 1) Launched two 8 th gen TPUs (TPU...

CMB International

The most expensive chip in an AI rack runs on Texas Instruments' cheapest ones

An AI processor needs electricity at less than one volt, but power reaches the cabinet at hundreds of volts. Texas Instruments makes many of the...

The Physical Layer

Price Gauge

52-week Range

Market Cap (USDm)

P/E

Dividend (Yield)