Asking an AI Expert: What Exactly Is a 'Full Stack'?

In the past, in software development, Full-stack meant the ability to handle frontend, backend, and databases all together. Google has expanded this concept into the AI domain, aiming to provide an integrated system without the hassle of directly connecting fragmented technologies.

Google's AI full stack consists of the following core layers:

• Compute infrastructure: custom hardware such as TPUs (Tensor Processing Units)
• AI models: the Gemini family developed by Google DeepMind
• Orchestration platform: Gemini Enterprise Agent Platform
• User interfaces: everyday services such as Maps and Gmail

This vertical integration strategy has been a deliberate move spanning decades. In particular, by investing in its own supply chain such as TPUs starting a decade ago, Google has been able to secure a much higher level of service performance and reliability than would be possible by relying on outside vendors.

Google emphasizes that its platform has an 'Extensible' structure. Users can use everything immediately within Google's ecosystem, but they can also flexibly choose to connect models from other companies as needed.