An integrated, full-stack AI platform (chips to models to applications) eliminates the integration challenges that are enterprises' number one issue, cutting implementation and adoption costs, while chip-optimized model training (e.g., on TPUs) lowers token costs that get passed on to customers.
Riyaz argues Google's full-stack integration (hardware through apps) removes the top pain point for enterprises—tool integration—while also lowering the cost of AI intelligence via chip-optimized model training. ✦ AI generated
Riyaz Habibbhai · The TWIML AI Podcast · 2026-04-24 · original ↗
starts at this moment · 3:26
“Talk about how that full stack really manifests itself into customer value.”
What's the number one issue enterprises face is integration challenges, right? Having an integrated stack means out of the box it all works together. So, you're saving on implementation cost, you're saving on adoption costs, right? All of that is just savings for an organization.
verbatim transcript · starts at 3:26
3:26one issue enterprises face is integration challenges, right? >> Yeah. Is or I have multiple different toys in my tool set and I want to all these toys to work together, right? Having an integrated stack means out of the box it all works together. So, you're saving on implementation cost, you're saving on adoption costs, right? All of that is just savings for an organization. Additionally, for our research and models,
3:47building those models and training them on the hardware like TPU specifically allows us to really run things efficiently, right? And that allows us to really reduce the token cost in even in our models. And that gets transferred to our customers as well. So, it's not just a integration the saving cost, but also how you get the cutting-edge value of AI through these cutting-edge frontier models at a at a cost benefit advantage
4:12because it's trained on a specific chip for that particular advantage, right? So, that integration across different layers helps you to deliver something that is a very forward-leaning at the right price point that the customers can quickly adopt, right? And we've seen that across the our model evolution as well, right? We give better and better intelligence and the the cost for intelligence keeps going down from every iteration of the
4:38generation of the model that comes through. I've written extensively about Google's approach to agents and the evolution of Google Cloud's platforms to support agents. And on the one hand, Google's been very early in recognizing the vision of agents and what it agents make possible. But that has led to um let's say, you know, turn and rapid evolution from the platform perspective. Talk a little bit about the evolution of
5:09Vertex AI into Gemini, what's now called Gemini enterprise agent platform, and some of the standout uh features and capabilities of the shift. Yeah, absolutely. So, Vertex AI for the last few years has been our de facto developer platform for you to build uh agents and applications and enroll it out within your organization, right? Right. Um so, last October, we launched Gemini enterprise, the app. And the promise there was to