Google’s Gemini 3 Receives Rave Reviews

Key Concepts

• Gemini Leak: Refers to leaked information about Google's Gemini AI model.
• Multimodality: The ability of an AI model to process and understand information from multiple types of data, such as text, images, and code.
• Visual Reasoning: The capability of an AI model to interpret and draw conclusions from visual information.
• Waymo: Google's self-driving car company, which benefits from AI advancements.
• TPUs (Tensor Processing Units): Google's custom-designed hardware accelerators for machine learning.
• NVIDIA GPUs: Graphics Processing Units manufactured by NVIDIA, widely used for AI training and inference.
• Google Cloud (GCP): Google's cloud computing service.
• Blackwell: A generation of NVIDIA GPUs.
• Training: The process of feeding data to an AI model to teach it.
• Inferencing: The process of using a trained AI model to make predictions or decisions.

Gemini Leak and its Significance

The discussion centers on the implications of a "Gemini leak," suggesting that Google's Gemini AI model represents a significant leap forward. The paper associated with this leak highlighted Gemini's advanced capabilities in multimodality, demonstrating its ability to handle code, image generation, and visual reasoning. This multimodal capability is seen as a key factor in the success of projects like Waymo, Google's self-driving car initiative, which leverages AI derived from such models. The progress from Google's earlier "byte barred" launch to the current "Gemini three model" is described as a well-executed advancement, positioning Gemini as a frontier model.

The Role of TPUs and NVIDIA GPUs

A crucial aspect of the Gemini leak's implications relates to hardware. The success of Gemini is reportedly tied to the use of TPUs, Google's custom chips. This is significant because it suggests that Google's internal reliance on TPUs for AI development might free up their allocation of NVIDIA GPUs.

• Google's NVIDIA Allocation: Google is a major customer of NVIDIA, ranking among their top three.
• Internal vs. External Use: While companies like Meta use their NVIDIA allocations primarily for internal training and inferencing (e.g., for their family of apps, recommendation systems), Alphabet (Google's parent company) utilizes its internal TPUs for many internal workloads.
• Google Cloud (GCP) Opportunity: This shift means that Google Cloud could potentially allocate more of its NVIDIA GPU resources, including the latest Blackwell chips and prior versions, to external customers. This would allow customers to rent these powerful GPUs, thereby generating revenue for Google Cloud.

Impact on Google Cloud Revenue

The availability of more NVIDIA GPUs on Google Cloud is presented as a positive development for their cloud business. By offering these high-demand chips to external clients, Google Cloud can increase its revenue streams, similar to how other cloud providers operate. The analyst suggests that this increased availability could lead to a "lift" in cloud revenue.

Conclusion and Key Takeaways

The "Gemini leak" signifies a substantial advancement in AI, particularly in multimodal capabilities. The success of Gemini, reportedly driven by Google's TPUs, has a direct impact on the allocation of NVIDIA GPUs. This could lead to Google Cloud offering more NVIDIA hardware to its customers, potentially boosting its revenue and competitive position in the cloud market. The progress highlights Google's strong execution in AI development over the past two years.