Original video in Chinese.
Key Takeaway
- By restraining its desire to make money, Google launched AI Mode for Search, showing its strategic resolve to be AI First.
- Google has achieved a “king’s return” in AI, and its full-stack AI product system covers the foundation layer (AI search, Gemini models, TPU compute), the enabling layer (developer tools), and the experience layer (multimodal content creation, Agent platforms).
- The Gemini model family continues to iterate, improving multimodal and reasoning capabilities, such as Gemini 2.5 Pro’s Deep Think and Flash’s speed optimizations.
- Google keeps innovating in AI product features, such as Veo 3 for video generation, Flow for video editing, and Project Mariner as an Agent platform.
- Google also updated Deep Research with support for uploading documents and generating infographics, making it more practical.
- The article predicts that 2025 will be the year Google fully erupts in AI, and its full-stack AI product system will go into all-out battle.
Let me ask you a question: what kind of giant is the most terrifying?
When a giant can restrain its desire to make money. It would rather earn less in order to complete a strategic transformation and keep up with the times. Never underestimate this point—so many giants in the past died because of it.
Today, Google has entered this state. Even though it is still unclear how the business model of traditional search will combine with AI, it still launched AI Mode for Search.
Now, you can ask Google anything, whether it’s a simple question, a super long question, or even a question that contains ten questions inside it, and it can understand it. You can also ask it to help gather information and help you make shopping decisions. You can speak to it, type to it, or show it something with a photo—anything works.
The capital markets are not very optimistic. Because traditional ad revenue will definitely be affected—you already get the answer directly, so naturally there is no need to click those links anymore. In addition, traffic distribution will change dramatically. This will hit the entire content ecosystem and advertising network.
But there’s no choice. As a giant, if you don’t take the initiative to go AI First, then you’re just waiting for someone else to come and take your life!
Hello everyone, welcome back to my channel. Modestly speaking, I’m one of the few bloggers in China who can explain the why and how of AI clearly. What I provide is far more valuable than tutorials. Remember to follow. If you want to connect with me, come to our newtype community. More than 1,000 friends have already joined for a fee.
Back to today’s topic: Google in the AI era.
In a video this March, I made a prediction: in 2025, Gemini will establish Google AI’s dominance in the consumer market.
What followed, such as the launch of Gemini 2.5 Pro, confirmed my judgment. A few days ago, the I/O conference strengthened my confidence even more. And it wasn’t just Gemini—everything announced at this conference showed us a brand-new Google, a giant of the AI era.
Because there was so much new stuff and so much information, I summarized it into three layers according to my own logic. You can use this framework to look at this I/O conference.
The most important is the foundation layer.
What I just mentioned about search belongs to the foundation layer. Because it is not just a product, but also a reflection of Google’s core capabilities and massive data. Through AI Mode, Search is being reshaped into a smarter, more conversational entry point, while providing models with continuous, real-world data and user interactions.
The models centered on the Gemini family are the second key part of the foundation layer. There were many updates at this conference, including Gemini 2.5 Pro adding Deep Think capabilities, Gemini 2.5 Flash’s speed and efficiency optimizations, and the new Gemma 3n, which has low hardware requirements and is very suitable for running on everyday devices. The continuous iteration of these models, as well as the improvement of multimodal capabilities, are the foundation for achieving more advanced intelligence.
Compute infrastructure is the third key part of the foundation layer. Google’s TPU has already reached its seventh generation. These custom chips provide powerful compute support and energy efficiency guarantees for training complex models and large-scale inference. This is the physical cornerstone of Google’s strategy.
Above the foundation layer are two other layers.
One is the enabling layer, mainly aimed at developers. Through an enhanced Gemini API, Vertex AI, and a whole suite of other tools, Google provides developers with a toolkit for building, deploying, and managing AI applications.
The other is the experience layer, mainly aimed at users. The bling-bling stuff at this conference all belongs to the experience layer. For example, Google Beam’s immersive 3D communication, Android XR’s mixed-reality interaction, and the generative content creation of Veo 3 and Flow.
As for the forward-looking Project Mariner, it ties these three layers together. It represents Google’s ambition to achieve a “universal AI assistant.”
Don’t think all the things I just mentioned are just Google drawing a nice picture. In fact, many of them have already been launched, such as Veo 3, Flow, and Project Mariner. A few more will be launched in the next few weeks, such as the deep thinking feature in Gemini 2.5 Pro.
I specifically upgraded to Ultra membership to try it myself.
Veo 3 is really stunning! No wonder it has been going viral on Twitter. In addition to being highly realistic visually, Google also added audio in this release. For example, this clip.
Flow is an editing platform based on Google’s video models. You can generate clips with text on it, and then stitch the clips together. People on Twitter have already used Flow to create very good works, such as this clip.
Project Mariner is an Agent platform with capabilities similar to Computer Use. I tried it out, and it was already fairly smooth at operating the browser and clicking buttons on web pages. If it encounters a situation where login is required, it will stop and hand control back to the user.
Beyond these cool products, Google also updated some practical features this time. The two I like most, I already mentioned in the community:
First, Deep Research can now upload your own documents. Because some materials cannot be found through public search, you need to provide them yourself.
Second, after Deep Research finishes, if you think there’s too much text and your head hurts from reading it, you can click the top right and generate an infographic. Then the reading pressure becomes much lighter.
All of these products and features—those that already exist, those about to launch, and those that will mature in the future—put together make an AI version of Google, a Google that has awakened and is counterattacking on all fronts.
When some companies are still reflecting on the fact that Chatbots are not the ultimate form of AI, Google has already given the answer: a “full-stack AI product system.” Not competing with you on a single product, but going into all-out battle across the board.
If OpenAI’s GPT-5 is not impressive enough, then 2025 will belong to Google.
OK, that’s it for this episode. If you want to understand AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!