Google's Diverging AI Hardware Strategy: Is Independent Processing the Best Solution?

Google's Diverging AI Hardware Strategy: Is Independent Processing the Best Solution?On August 13th, Google officially held its 2024 annual "MadebyGoogle" event, showcasing its latest advancements and innovations in hardware and software, with AI at its core. New products including the Pixel 9 series, Pixel 9 Pro Fold foldable phone, Pixel Watch 3, and Pixel Buds Pro 2 were unveiled, showcasing a significant increase in new products compared to the previous year

Google's Diverging AI Hardware Strategy: Is Independent Processing the Best Solution?

On August 13th, Google officially held its 2024 annual "MadebyGoogle" event, showcasing its latest advancements and innovations in hardware and software, with AI at its core. New products including the Pixel 9 series, Pixel 9 Pro Fold foldable phone, Pixel Watch 3, and Pixel Buds Pro 2 were unveiled, showcasing a significant increase in new products compared to the previous year. Typically, September and October are peak months for phone market competition, with brands like Google, Apple, and Huawei releasing their annual flagship products. While Google's annual event was expected to take place in October, the company likely chose to hold the event earlier to gain a competitive advantage. During the event, Google introduced its new Gemini AI model while subtly criticizing Apple's Apple Intelligence, highlighting the intense competition between these tech giants.

The Pixel 9 Leads Google's Product Line-Up with AI as the Main Highlight

The most eye-catching hardware product at the event was undoubtedly the Pixel 9 series, featuring the Pixel 9, Pixel 9 Pro, and Pixel 9 Pro XL, along with the second-generation foldable phone Pixel 9 Pro Fold, which was integrated into the Pixel 9 series for the first time.

Among the three straight-screen models, the Pixel 9 standard model has a 6.3-inch screen, while the Pixel 9 Pro and Pixel 9 Pro XL have 6.3-inch and 6.8-inch screens respectively. The entire Pixel 9 series is powered by Google's latest Tensor G4 processor, and the main camera uses Samsung's 50MP GNK sensor. The ultra-wide-angle lens uses Sony's IMX858 sensor. The Pixel 9 Pro also utilizes the IMX858 sensor for its telephoto lens and front camera.

To better support on-device AI, the Pixel 9 series has upgraded storage, starting at 12GB for the standard version and reaching 16GB for the Pro version, keeping pace with mainstream Chinese smartphone brands.

Compared to the previous generation, the Pixel 9 Pro Fold foldable phone has noticeably improved in terms of thinness and lightness, with a thickness of 10.5mm when folded and 5.1mm when unfolded. However, with a 6.3-inch external screen and an 8-inch internal screen when unfolded, the phone weighs 257g, making it heavier than almost all other new generation foldable phones. In other aspects, the Pixel 9 Pro Fold is equipped with 16GB of memory and a camera combination of 48MP wide-angle + 10.8MP ultra-wide-angle + 10.8MP telephoto, offering a decent but not exceptional overall performance.

Gemini: The Future of Google's On-Device AI

While the Pixel Buds Pro 2 and Pixel Watch 3 are also noteworthy, the biggest highlight of the event is undoubtedly the deeply integrated Gemini large language model. As described, Gemini is deeply integrated with all Google applications and the Android system, capable of independently handling complex queries. Gemini also boasts multimodal processing capabilities, able to analyze image, voice, and text information.

Specifically, Google's on-device AI is driven by the lightweight multimodal model Gemini Nano and has incorporated the more flexible Gemini 1.5 Flash. This enables new Pixel devices to have innovative AI features, such as the Pixel Studio image generator. This tool uses local diffusion models to generate stickers and images based on text prompts, similar to Apple's Image Playground. Additionally, AI can edit photos by adding objects that didn't originally exist and generate customized weather reports. The camera app has a new feature that merges two photos to include the photographer in group pictures. It can also categorize and memorize information from screenshots. The "Call Recording" function can record and summarize phone calls. The GPT-4-equivalent voice assistant Gemini Live allows users to have free-flowing conversations with Gemini.

Similar to Apple's Apple Intelligence, Gemini also utilizes local processing. However, Google emphasizes that Gemini doesn't rely on third-party AI services, providing users with greater peace of mind regarding privacy and security. While this statement is intended to criticize Apple's decision to leverage ChatGPT for text generation, image creation, and other features, it also contrasts the way AI integrates with third-party large models. This has sparked curiosity among many: which is the "correct" path for AI hardwarerunning self-developed large models or integrating various large models?

AI Hardware Divergence: Google, Apple, and Huawei Each Go Their Own Way

To answer this question, we can look at Google, Apple, and Huawei, companies that share both hardware and software expertise. According to the latest data released by Counterpoint, in the first quarter of 2024, Android, iOS, and HarmonyOS held the top three spots in the global mobile operating system market share, with respective proportions of 77%, 19%, and 4%. These three players essentially dominate the global smartphone market, and to foster ecosystem unity, their other hardware products generally run on the same operating system.

In terms of operating system operating models, Google and Huawei are more similar, opening up their systems for collaboration with third-party developers, while Apple maintains tight control over both hardware and software, barring third-party involvement. These different operating models give the operating systems different characteristics, one offering diversity and playability through openness and collaboration, the other promoting ecosystem unity through centralization, often referred to as "tonality".

Looking at specific functions, the AI features on hardware from Google, Apple, Huawei, and most mainstream brands are largely similar, including common functionalities like AI assistants, text-to-image generation, image editing, and real-time translation, without significant distinctions. However, behind these AI functions are individual large models working their magic.

It's clear that compared to operating systems, the strategies of these three players regarding AI large model integration in hardware have changed. Interestingly, the brands are also learning from each other's strengths. Google has shifted from its open Android model to independent processing of on-device AI through Gemini Nano. Huawei, in addition to its PanGu model, has introduced third-party cooperative large models like ERNIE, iFlytek Spark, and Zhipu ChatGLM. Apple has already announced its partnership with OpenAI and is looking for more collaborators. It's rumored that Google Gemini is on the shortlist, but considering Google's stance at the launch event, this is unlikely to materialize.

Google's independent processing capabilities in its AI strategy bring a high degree of unity and privacy to products, minimizing the risk of leaks that could arise from interactions with third-party AI providers. As Apple's ecosystem claims, it boasts a high level of security. Apple and Huawei, on the other hand, favor finding more partners to expand their AI arsenals, sacrificing unity and privacy for greater possibilities due to the additional coordination steps required. For instance, this allows them to support a wider range of hardware, including cars, home appliances, drones, robots, and specific industry hardware combinations, and to provide interfaces to nurture more third-party software apps, creating a richer product ecosystem.

Intensified AI Hardware Competition: "Independent Processing" is Not the Optimal Solution

Currently, nearly all smartphone brands are building end-to-cloud collaborative AI architectures, primarily partnering with Google and OpenAI overseas and Chinese service providers like Baidu and ByteDance domestically. The reason is simple: pure on-device large models have high hardware requirements, leading to the current trend of smartphone brands increasing memory capacity.

On the other hand, from the perspective of smartphone manufacturers' business models, the end-to-cloud integration model creates a better synergy with the manufacturers' operations. From a cost standpoint, offloading all AI-related tasks to cloud processors incurs high cloud reasoning costs and high concurrency processing demands, making it an inefficient solution. Leveraging local resources judiciously can help manufacturers save costs. From a user perspective, many application scenarios require the timely intervention of AI large models. On-device AI models can reduce latency while keeping sensitive data off the cloud, safeguarding user privacy.

Therefore, when end-to-cloud collaboration becomes the norm, except for companies like Google with a comprehensive enough large model ecosystem to independently support on-device AI systems, Apple's approach might be more suitable for most brands.

From the perspective of final presentation, Google, which claims to use independent processing, hasn't demonstrated any particular competitive edge. Even the heavily promoted Pixel Screenshots and AddMe features pale in comparison to the innovative AI applications implemented by Chinese brands such as OPPO, Honor, and Huawei. Google's AI capabilities seem relatively minor.

In the long run, Google's strategy of not relying on third-party AI may not be the optimal solution. As the number of compatible hardware types and software interfaces increases, the limitations of this approach will become increasingly apparent. In such a scenario, brands with more options will gradually gain an advantage in the AI field.

To be fair, Google Pixel began focusing on AI experiences back in 2016, becoming the first brand globally to espouse the "AI First" concept. However, at the time, the dominant AI technology was deep learning, which hadn't branched out into the diverse AI capabilities we see today. Google's early investment is one of the reasons it has become a leader in overseas large models.

The Pixel 9 series and other new products