At Tech Week Asia 2025, Moe Abdula, VP of Customer Engineering for Google Cloud APAC, shared thought-provoking insights on the growing pervasiveness of AI — and how the right infrastructure can determine its success.
He began with relatable examples of AI in action: from autonomous vehicles that can decide when to open their trunks, to movie recreations transformed into one-minute clips, and even the viral AI-generated cat videos that continue to flood social media.
These everyday encounters illustrate how deeply embedded AI has become in both business and entertainment — capable of reasoning, decision-making, and creativity all at once.
The Challenge: Building Infrastructure That Scales
While the world celebrates AI’s capabilities, Moe emphasized a crucial point — the unseen complexity behind making it all work efficiently. “AI doesn’t just need data and algorithms; it needs an infrastructure smart enough and affordable enough to sustain it,” he noted.
Google’s approach focuses on optimizing hardware chips and designing systems that place AI agents in the right operational context. This balance of efficiency and scalability ensures that AI remains accessible to organizations of all sizes.
But perhaps the most important takeaway was quality. As Moe pointed out, “The first token generated might be fine — but what about the millionth?” Poor data or inefficient systems translate into opportunity costs and wasted time. Hence, the emphasis on high-quality, well-structured data as the backbone of every AI model.
From Banking to Insurance: Real-World Use Cases
AI agents are already reshaping industries. In banking, for example, AI-powered workflows are helping streamline loan applications — automating verification, assessment, and documentation. Moe shared that by 2027, over 50% of enterprise workflows are expected to include AI assistance through multimodal systems, integrating text, voice, and visual data to enhance productivity.
In the insurance sector, Moe illustrated how Gemini Lite, Google’s multimodal AI, could simplify claims processing. Traditionally, insurance claims — especially after incidents like fires — involved piles of paperwork and confusion over inventory lists. Now, with Gemini Lite, users can simply walk around their home with a camera while the AI automatically compiles an accurate inventory of visible items. This allows for faster, more reliable claims processing — a real example of data in context.
Data in Context: The Future of AI Utility
The concept of “data in context” was a recurring theme throughout Moe’s talk. It’s not just about having massive amounts of information, but about using relevant, well-structured, and contextualized data that enables AI to make targeted and reliable decisions.
When combined with multimodal capabilities — the ability to interpret text, visuals, and audio simultaneously — AI becomes not only powerful but deliberate in its reasoning. This precision can even help mitigate data injection risks and improve model reliability in real-world applications.
Takeaway!
Moe Abdula’s session ultimately spotlighted a key truth: AI’s success isn’t just about algorithms — it’s about infrastructure, data quality, and context. As we move toward an AI-assisted future across industries, building smarter, scalable, and secure systems will be the cornerstone of innovation.
