Interview
Google Cloud's AI builder summit signals seismic shift: startups consuming credits faster than ever and staying
Sep 18, 2025 with Darren Mowry
Key Points
- Google Cloud is positioning itself as a full-stack AI platform spanning chips, first-party models like Gemini, third-party options including Claude and Llama, and a new agentic layer that lets startups build and distribute agents.
- Startups in Google Cloud's program are consuming credits faster than ever and staying after depletion, a reversal of prior cloud eras when interchangeable compute allowed easy migration.
- The retention shift stems from stack integration rather than switching costs alone, making the current pace of infrastructure change unlike anything observed during earlier cloud buildout cycles.
Summary
{
"long_summary": "Darren Mowry, Managing Director for North America at Google Cloud, sat down at the company's first Global AI Builders Summit, held at Google's Mountain View campus in a cafeteria named after the company's 57th employee. The event brought together hundreds of founders on-site and tens of thousands of builders joining digitally, with customers including Lovable, Fireworks AI, and Reflect Orbital presenting alongside Google Cloud and DeepMind speakers.\n\n**The infrastructure pitch**\n\nMowry frames Google Cloud's startup proposition as a full-stack AI platform rather than a commodity compute offering. The pitch spans chips (both Nvidia GPUs and Google TPUs), first-party models (Gemini, Veo), and third-party models including Anthropic's Claude and Meta's Llama, all accessible through a single integrated environment. The most recent addition is an agentic layer built around an SDK and ADK that lets startups build, publish, and distribute agents using any of those models. Mowry says founders describe this as a real, shipping capability rather than a product roadmap.\n\n**Credit consumption**\n\nMowry says the Google Cloud for Startups program has seen a dramatic increase in both enrollment and credit consumption over the past 18 months. The more significant change is retention. In earlier cloud generations, startups would collect credits from one provider and migrate elsewhere once they ran out, a pattern that worked when the underlying compute was interchangeable virtual machines. Mowry argues that dynamic has broken down. Startups building on GPU, TPU, Gemini, and the agentic stack are consuming credits faster than ever and, critically, staying on Google Cloud after they're gone. The stickiness, he says, comes from the integrated nature of the stack rather than from switching costs alone.\n\nMowry has been at Google Cloud for nearly five years and previously worked at another major hyperscaler during the early cloud buildout. He describes the current pace of change as unlike anything he observed in that era."
}