EP667 | 🌍
- Optical communication and Co-Packaged Optics (CPO) have become an unshakeable industry consensus, with massive long-term growth potential despite early-stage revenue contribution.
- AIPC is undergoing a massive upgrade in silicon and software, enabling offline execution of open-source models and saving substantial cloud bandwidth and token fees.
- Open-source models are trailing proprietary ones by only six months, paving the way for a co-existing model where high-end cloud services and edge computing complement each other.
- AI market leaders like Nvidia and TSMC are trading at relatively attractive valuation multiples compared to speculative high-beta peers, with technical charts showing signs of capital rotation back to quality.
Computex Observation: Optical Communication and CPO Form an Inevitable Industry Consensus
At this year's Computex Taipei, the absolute center of attention was the keynote speeches delivered by global tech CEOs. Nvidia founder Jensen Huang specifically highlighted Marvell as a potential trillion-dollar enterprise, instantly igniting massive market momentum and sending the stock soaring. However, evaluating its fundamentals soberly, while Marvell excels in the digital signal processor (DSP) space, its roadmap and visibility in the Application-Specific Integrated Circuit (ASIC) domain remain less clear compared to leaders like MediaTek and Broadcom.
This highlights the incredible market-moving power of Jensen Huang's endorsements.
A real thinking move here is the separation of endorsement from economics. The host doesn't treat Jensen Huang's mention of Marvell as proof that Marvell's ASIC position is suddenly de-risked; he splits market reaction, narrative power, and actual business visibility. That matters because celebrity signal can move price instantly, while roadmap clarity takes much longer to earn. Listeners can reuse that habit anywhere a famous operator's comment starts doing the valuation work by itself.
From a technical perspective, the core consensus established at the exhibition is the irreversible transition to optical communication. Although Nvidia continues to extract physical performance limits from copper wiring for short-range rack configurations, the exponential growth of data transmission means that inter-rack and data center connections are inevitably shifting toward optical transceivers. From historical pluggable optical modules to the co-packaged optics (CPO) technology currently under heavy discussion, while material revenue contribution remains in its infancy, long-term technical consensus is solidifying, and capital will likely rotate back to this sector continuously.
The Second AIPC Revolution: Rise of Local Open-Source Models and Hardware Material Upgrades
Looking back at the early iterations of AIPC over the past two years, limited silicon performance and insufficient software optimization rendered the user experience subpar, leading the market to dismiss it as short-lived hype. However, after two years of architectural iterations and optimization, the new generation of AIPCs is bringing real utility. Edge devices now possess the capability to run various open-source large language models smoothly in offline environments. This implies that enterprises and individual users can perform local inference with ultra-low latency, drastically reducing cloud API token and data transfer costs.
This transition will directly trigger a comprehensive hardware specification upgrade across the PC supply chain. To support heavy local computing, specialized materials surrounding key chips have seen significant upgrades, including advanced PCB substrates, high-performance passive components, and more complex power management ICs. Consequently, the average selling price (ASP) of these components has increased. Even if overall consumer PC shipments remain sluggish this year, the high-ASP and high-margin component demand driven by AIPC will likely offset volume declines, injecting vital momentum into hardware brands and component suppliers.
Cloud and Edge Co-existence: Model Evolution from the Perspective of a Snowflake Employee
Based on first-hand communication with an insider at database giant Snowflake, current enterprise-level AI workflow shows a complementary cloud-edge collaborative framework. When a user submits a query, initial natural language understanding is outsourced to proprietary cloud LLMs, while subsequent data retrieval, processing, and sensitive computations are handled locally via internal data lakes and local models. Crucially, open-source models are trailing proprietary models by only about six months.
Sign up to keep reading
10 more sections await — finish reading on Podket.
Sign up freeFree to read. No credit card required.
Tickers are shown only because the company was mentioned in this episode, for your reference. Not investment advice, not a recommendation to buy or sell.
Disclaimer: The above is an AI-generated summary of a third-party programme, may contain errors or omissions, and is provided for personal, non-commercial reference only. Podket accepts no responsibility for its content and makes no representation or warranty as to its accuracy, completeness, quality, timeliness or reliability, and expressly disclaims any liability for any loss or damage arising from all or part of it. The content may reflect the personal opinions and views of the original programme's author and does not represent the position of Podket. Nothing in it constitutes any solicitation, offer, opinion or recommendation by Podket for any investment, nor legal, tax, accounting or investment advice or services regarding the returns or suitability of any security or investment. Investors must make their own investment decisions in light of their own investment objectives and financial circumstances.