The "new king" of AI ascends the throne, and the A-share Google chain takes off collectively!

Google's AI large model upgrade ignites the A-share Google chain, and the optical communication sector takes off collectively.

On November 25th, the A-share optical communication sector surged, with Guangku Technology hitting the 20% daily limit, and Tengjing Technology rising over 15%, both reaching historical highs. Dekeli, Changguang Huaxin, Changfei Fiber, and Fojing Technology also hit the daily limit, while Changxin Bochuang, Shijia Photon, Juguang Technology, and Taicheng Light all rose over 10%. Analysts pointed out that the market's enthusiasm for the optical communication sector stems from Google's recent release of the next-generation AI large model Gemini 3 and the breakthrough performance of the image generation model Nano Banana Pro.

(Source: Wind)

The Gemini 3 model released by Google this month is primarily trained on TPU chips, with performance approaching or even surpassing OpenAI's ChatGPT. This technological breakthrough has led investors to reassess the AI chip market landscape. Google's stock price soared 6.3% in a single day to a historic high of $318.58, with a cumulative increase of 68% this year. More importantly, Google is leveraging its latest breakthroughs in AI models to promote the deployment of TPU chips in its own data centers to large clients like Meta.

Fangzheng Securities and Huatai Securities pointed out in their latest research reports that Google has introduced optical switching (OCS) technology in TPU interconnection, constructing TPU supernodes composed of multiple cabinets. At the Hot Chips 2025 conference, Google detailed the next-generation TPU chip Ironwood. Cignal AI expects that by 2029, the total potential market size for optical switching OCS will exceed $1.6 billion. Therefore, the domestic optical communication industry chain is expected to benefit significantly.

Comprehensive Upgrade of Google's AI Large Model

Google's AI capabilities have achieved a comprehensive leap. Gemini 3 has significantly enhanced its capabilities in programming, application development, and image generation, supporting a context window of 1 million tokens, suitable for functions such as Agent, advanced programming, long context, and multimodal understanding.

According to Huatai Securities' research report, Gemini 3 Pro topped LMArena with an Elo of 1501 and achieved a score of 37.5% in the "PhD-level reasoning" test without tools in Humanity's Last Exam. The "deep thinking mode" further raised HLE to 41.0%, and it reached 93.8% in GPQA Diamond. In the multimodal field, MMMU-Pro scored 81.0%, Video-MMMU reached 87.6%, and ScreenSpot-Pro for screen understanding obtained 72.7%, significantly leading competitors

More importantly, this model is trained based on Google's TPU chips, supporting a context window of 1 million tokens, suitable for functions such as Agent, advanced programming, long context, multimodal understanding, and algorithm development. It scored 72.1% in the SimpleQA Verified benchmark test, significantly outperforming competitors.

The Nano Banana Pro large model released simultaneously by Google has achieved significant improvements in text rendering accuracy, resolution, and professional control capabilities, claiming to provide "studio-level" precision and control. This model upgrades image generation from "diffusion direct painting" to "understand first/generate later," and is being fully promoted in Google's existing AI tools.

According to a research report by Founder Securities, the scale of tokens in Google Cloud has shown explosive growth. In a conference call in the third quarter of 2025, Google stated that the amount of tokens processed monthly by the company surged from 98 trillion in July to over 130 trillion, growing more than 20 times within a year. The monthly active users of the Gemini application exceeded 650 million, with query volume tripling compared to the second quarter. Huatai Securities pointed out that the leap in model capabilities and ecological integration is shortening the time gap between "input-validation-revenue," combined with the increase in the proportion of inference demand and the scaled growth of cloud tokens.

TPU Strategic Upgrade Promotes OCS Industry Chain Development

More importantly, according to an article from Wall Street Watch, media reports citing informed sources indicate that Meta is negotiating with Google to use TPU chips worth billions of dollars in its data centers by 2027, while planning to rent chips from Google Cloud next year. This potential deal could allow Google to capture 10% of Nvidia's annual revenue, bringing in billions of dollars in new income, and some leaders in Google's cloud division have expressed this goal internally.

For years, Google has been renting its self-developed Tensor Processing Unit (TPU) chips to cloud customers, but it has now begun marketing a new solution to clients, including Meta and large financial institutions: allowing them to use TPUs in their own data centers. This new program, named "TPU@Premises," is supported by software developed by Google called "TPU command center" to facilitate easier use of these chips for customers.

Founder Securities pointed out that at the Hot Chips 2025 conference, Google detailed the next-generation TPU chip Ironwood. The new generation Ironwood supernode achieves cabinet-level connectivity through OCS optical switching with 9,216 Ironwood chips, further strengthening the important role of OCS optical switching in Google's network connectivity Optical Switching OCS is a technology that enables direct switching of optical signals between fiber ports without the need for optoelectronic/electro-optical conversion. Google has introduced Optical Switching OCS technology in TPU interconnects, constructing TPU supernodes composed of multiple cabinets that support a 3D Torus topology.

According to Founder Securities, Cignal AI predicts that by 2029, the total potential market size for OCS optical switching will exceed $1.6 billion. With the advancement of Google's large models, the OCS industry chain is expected to benefit from the ongoing progress of Google's AI initiatives. Huatai Securities strongly recommends core links in the Google chain, including optical modules/OCS, liquid cooling, and other fields.

Huatai Securities analysts point out that amid rising controversies over the overseas AI "bubble" and short-term adjustments in stock prices, Gemini 3 has achieved generational leaps in reasoning, multimodal, and Agent platform three main lines, with the model's capabilities crossing and ecological integration shortening the "input-validation-revenue" time gap. Coupled with the increase in the proportion of reasoning demand and the large-scale growth of cloud tokens, the AI computing power chain market is expected to continue to evolve.

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk