Google Cloud outlines three key AI stack priorities for startups

Google Cloud identifies three emerging trends shaping how startups build their AI infrastructure, noting that adoption rates on the platform are higher now than at the start of the AI era. The company argues that a complete stack—combining compute, models, and agent management—is the primary driver of this growth.

Key Adoption Trends

The report highlights three specific areas where startups are concentrating their efforts:

• Gemini Enterprise Growth: Startups are increasingly using Gemini Enterprise for managing TPU and GPU clusters, building agents, and accessing first- and third-party models. This adoption often leads to the use of core cloud services like Storage, BigQuery, and GKE.
• Mixed Model Strategies: Companies are deploying a combination of frontier models and "workhorse" models to balance price-performance across diverse tasks such as scientific research, generative media, and financial analysis.
• Compute Flexibility: Access to both GPUs and TPUs is considered critical. Startups rarely use compute in isolation, instead pairing it with additional technologies like agent-building tools and data services.

Recent Customer Implementations

Google Cloud cites deals with over a dozen startups in the past 60 days to illustrate these trends. Notable examples include:

• Arya Health: Uses Gemini 2.5 Pro, 3.1 Flash, and 3.5 Flash Lite for agentic workflows in healthcare administration, selecting models based on reasoning, speed, and cost.
• Casco: A cybersecurity firm using Gemini 3.5 Flash for focused subagent tasks alongside advanced reasoning models for complex attacks.
• CodeRabbit: Uses Cloud Run and Storage to underpin its application, and is beginning to leverage Gemini 3.1 Pro to power features like analyzing code change impacts and generating fixes.
• MicroAGI: A robotics startup accessing NVIDIA Blackwell systems and the Gemini Enterprise Agent Platform for multimodal processing.
• Ineffable Intelligence: Partnering with Google Cloud to access NVIDIA Vera Rubin systems for superintelligence research.

These cases demonstrate how startups are utilizing specific combinations of hardware (NVIDIA GPUs, TPUs) and software (Gemini models, BigQuery) to address sector-specific challenges in gaming, healthcare, cybersecurity, and robotics.