Race Track Hyper | AI First: Google's Weakness

The time window for the comprehensive implementation of AI LLM on the smartphone side is getting closer.

In addition to the well-known Qualcomm Snapdragon 8 Gen 3, which will support AI LLM at the hardware level, on October 4th, Google also released the software system with the largest support for AI LLM on the smartphone side so far-Android 14, and simultaneously released the intelligent terminal integrated with AI capabilities: the Pixel 8 series.

Google has been releasing smartphones for eight years, but it has not received much market attention. Now, Google wants to reverse this embarrassing fact by implementing AI LLM on the smartphone side. However, from the perspective of hardware configuration, the Pixel 8 series is quite ordinary and has once again been ignored by the market.

Compared with Google's efforts to expand the market space for smartphones, the secondary market is actually more concerned about the integration capabilities of Android 14, which Google has launched for AI. On the day Google officially released Android 14, the stock price rose by 2.23%, which is the largest single-day increase Google has achieved since August 29, 2023.

However, a 2.23% increase is not as exciting as the configuration and performance of the Pixel 8 series, and it is difficult to judge its sustainability.

Tensor G3: The New Source of Terminal AI Capability

This year marks the eighth year since Google released the Pixel brand smartphones. From the release of the first-generation Pixel in 2016 to today, the sales market share of Google smartphones is still classified as "others", without even a chance to show up.

In the early morning of October 4th, Beijing time, Google held its annual Made by Google 2023 event in New York. In addition to the Pixel 8/Pro smartphones, Google also brought the new version of the Android system, Android 14, and also released the Google Assistant with Bard, which is equipped with generative AI.

How amazing is Android 14? Or what are the new features of this new generation of Android operating system?

In addition to security, support for Ultra HDR-style photos and videos, stronger health functions, automatic framing, one-click access to Google Home controls, and other features, the most important highlight is the integration of AI capabilities. This is also Google's latest move to counter Microsoft's impact.

This capability is integrated into the Pixel 8 series smartphones, which were released on the same day. The Pixel 8 series, as the first smartphones to be equipped with the Android 14 system, had the most mentioned word on the stage, which is AI.

According to Brian Rakowski, Vice President of Product Management at Google, the Pixel 8 series phones are "centered around artificial intelligence" and powered by Google's self-developed Tensor G3 chip.The Tensor G3 chip is based on Samsung's 4nm process. The CPU architecture is a 1+4+4 structure, which consists of 1×Cortex-X3 (3.0GHz), 4×Cortex-A715 (2.45GHz), and 4×Cortex-A510 (2.15GHz).

In simple terms, within the CPU modules of the Tensor G3 chip, the Cortex-X3 is the high-performance core, responsible for handling high-load tasks such as gaming and multimedia applications. The Cortex-A715 is the mid-performance core, responsible for handling daily applications. The Cortex-A510 mainly assists in handling background tasks.

Based on the Geekbench 6 scores, the performance of the Tensor G3 chip is average and not as good as the Qualcomm Snapdragon 8 Gen2. This is related to Google's "tradition" - Google's self-developed Tensor G series chips have never had strong core performance, with AI being the core existence that surpasses chip performance. Therefore, the biggest technological highlight of the Tensor G3 chip is its AI capabilities.

Google claims that the Tensor G3 is capable of running more complex machine learning (ML) models, bringing AI-enhanced features to almost all parts of the Pixel 8 and 8 Pro.

Through the Tensor G3, Google further narrows the gap between smartphones and AI, transforming smartphones from mere communication tools to user personalized service assistants.

How does the AI capability of this chip manifest in the Pixel 8 and Pixel 8 Pro?

The number of machine learning (ML) models running on the Pixel 8 is twice that of the Pixel 6 (Tensor first-generation chip). The Pixel 8 Pro is the first smartphone to directly run Google AI models on a personal intelligent terminal, with computational power 150 times that of the largest ML model on the Pixel 7.

In terms of practical applications, the Tensor G3 can translate face-to-face conversations, transcribe messages in real-time, and translate symbols in 49 languages. This feature is very practical and also demonstrates the powerful AI capabilities of the Tensor G3.

This is the latest terminal for implementing AI models on the smartphone side. Before the release of the Qualcomm Snapdragon 8 Gen3, only Huawei had mass-produced intelligent terminals that implemented on-device AI models - Huawei Mate 60 Pro/Pro+ and Mate X5 (foldable screen). However, Huawei's on-device AI integration is based on the Pangolin large model, which is a cloud-side model rather than a locally deployed on-device model.

The Qualcomm Snapdragon 8 Gen3 can empower smartphone manufacturers to implement on-device AI models, and the Google Tensor G3 can do the same.

Google must have confidence in the AI capabilities integrated into Pixel. Rick Osterloh, Google's hardware chief, emphasized that "Pixel combines hardware and software, centered around AI, to provide users with a simpler, faster, and smarter experience."Although Osterloh's statement is suspected of "copying" Apple's "software-hardware integration" strategy, his emphasis on "AI-centric" can also be seen as a partial innovation.

So besides the AI power support from Tensor G3, what AI capabilities does Google's updated Android 14 have?

AI First: Hardware Constraints

To understand the role of AI in Google's future business, it is necessary to understand Google's business landscape.

According to Google's annual report, Google's business can be divided into two categories: Google Services and Google Cloud.

Among them, Google Services can be further divided into three major sectors: software, hardware, and network services. Google software includes Chrome browser and Android operating system, etc.; Google network services are represented by search engines; Google hardware includes Pixel terminals and Fitbit wearable devices, etc.; Google Cloud includes Google Cloud Platform and Google Workspace, providing cloud services and office service assistance.

Looking back at Google's history, Google's connection strategy with AI did not start today.

In 2014, Google acquired the AI company DeepMind. In 2015, the company became a wholly-owned subsidiary of Google's parent company, Alphabet Inc. In 2016, Google launched the first-generation Pixel smartphone and announced a shift from "Mobile First" to "AI First" strategy, with AI gradually becoming the most important module in Google's strategic layout.

The reason why Google launched the Pixel terminal brand is also because Google believes that hardware can support the implementation of AI. In 2023, DeepMind and Google Brain merged to explore the development boundaries of AI together.

Currently, Google has made layouts in various aspects such as deep learning frameworks, algorithm models, and computing power, and has built a high technological barrier.

Guided by the AI First strategic goal, Google also wants to build an integrated ecosystem of software and hardware, just like Apple.

Unfortunately, although Google smartphones have been promoted for 8 years, the progress is not worth mentioning. As software, which is the other extreme of promoting "AI-first," it is Google's traditional strength.

Therefore, Android 14 becomes the key for Google to embrace AI.

Although Google emphasized AI multiple times at the conference, there is not much new description of AI in the official version of Android 14.

Therefore, it is necessary to look back at the Google I/O 2023 developer conference on May 12 this year to find more relevant information. Although Android 14 was only mentioned once at that conference, it revealed three key generative AI features, including Cinematic Wallpapers, Magic Compose, and Generative AI Wallpapers.This is the focus of AI applications for Android 14.

However, these AI capabilities are somewhat superficial and not impressive. The reason for this may be that Android, as the most widely used intelligent terminal operating system globally, may be overwhelmed by a massive amount of deep AI applications.

Therefore, even if Google wants to add advanced generative AI functions to the Android system, it may be unable to do so until higher performance and larger-scale server clusters are put into operation.

Wall Street News has noticed that Google no longer sees the Android system as its core. This is not only reflected in the fact that Android was only mentioned once at the I/O 2023 Developer Conference, but also in the fact that the frequency of mentioning the term AI far exceeds that of Android at the latest Google event on October 4th.

Therefore, Google's determination to embrace AI seems irreversible. Compared to the technically mature and ecologically solidified Android ecosystem, AI is the source of Google's imagination for future performance.

Currently, Google's AI applications still focus on its core search business.

Google CEO Sundar Pichai said during the conference call for the release of Google's second-quarter earnings report on July 25th that the progress of the new search engine "Search Generated Experience (SGE)" exceeded expectations, and the product is currently undergoing small-scale testing. Google will continue to improve ad formats in the new search engine and provide advertisers with tools to utilize generative AI.

Based on the current situation, the prospects of Google's AI layout are still unclear: the hardware, which is considered indispensable for the implementation of AI, such as intelligent terminals, is difficult to make a breakthrough. Whether it can successfully break through with edge AI capabilities is still unknown. The role that the software Android system can play in this counterattack is also uncertain at the moment.