Google Cloud's Darren Mowry on startups consuming AI credits faster than ever and sticky infrastructure
Sep 18, 2025 with Darren Mowry
Key Points
- Google Cloud startups are consuming AI credits faster than ever and staying put, as integrated stacks of TPUs, GPUs, and multiple LLMs make switching platforms prohibitively expensive.
- Google Cloud's agent SDK and ADK are production-ready today, not future roadmap items, according to founder feedback shared by Darren Mowry at the company's first Global AI Builders Summit.
- The credit arbitrage model that once let startups hop between cloud providers has collapsed, replaced by deep infrastructure lock-in around bundled compute and AI capabilities.
Summary
Darren Mowry, Managing Director for Global Startups at Google Cloud, spoke at the company's first Global AI Builders Summit in Mountain View, an event drawing several hundred founders in person and tens of thousands digitally. Featured customers on stage included Lovable, Replit, and Fireworks.
Mowry's core observation is that AI has fundamentally changed what startups actually do with cloud credits. In the old model, credits were a commoditized incentive — a startup could burn AWS credits, then Azure credits, then Google credits, hopping between virtual machines with minimal friction or switching cost. That arbitrage no longer works the way it did.
Over the past 18 months, Google Cloud for Startups has seen more participants than at any point in Mowry's five years at the company, and those startups are consuming credits faster than ever. More importantly, they're not leaving. Mowry attributes the stickiness to the integrated depth of what's now on offer: TPUs, GPUs, Gemini, Claude, Llama, and agentic tooling — including an SDK and ADK for building and publishing agents — all within a single stack. When infrastructure, models, and agentic capabilities are tightly bundled, the calculus of switching changes. Startups that came for the credits stay because rebuilding on a different platform would mean unpicking the entire stack.
On agents specifically, Mowry pushes back on any perception that Google Cloud's agent offering is roadmap vapor. He says founder feedback consistently describes the agent capability as production-ready today, not a planned future release.
He also flags Gemma Nano as a recent release out of DeepMind, with further model and agent announcements expected shortly.