Google Cloud released a report this week concluding that AI startups now buy cloud services as full stacks rather than point solutions. We note that this report is essentially marketing—but dissected, the selection trends it reveals are worth breaking down.
What this is
Google Cloud published an "AI Startup Cloud Selection Trends" report on its official blog, with three core observations:
1. Gemini Enterprise—a complete package of models, Agent building tools, and cluster management—is growing strongly
2. Startups use both "frontier models" and "workhorse models" simultaneously—expensive and cheap models deployed together
3. Buying compute (GPU/TPU) is just the entry point; what retains customers are the second and third layer products (data, storage, Agent platforms)
Google also listed recently signed customers: Artificial Agency (game AI behavior) and Arya Health (medical administrative Agents), among others.
Industry view
The report's tone is a shout—"look, everyone's using us." We note several points worth flagging:
Opaque sample: Google didn't disclose "how many" or "what share"—everything is qualitative description. The so-called "trend" reads more like a marketing narrative.
Homogenization pressure: When all cloud vendors start selling "full stacks," switching costs for startups keep rising, potentially locking them into one cloud—which may not benefit customers.
Agent tools are the new battleground: The report repeatedly mentions Agent building platforms, indicating this layer is becoming cloud vendors' new moat. AWS has Bedrock Agents, Microsoft has Azure AI Foundry—this fight is just beginning.
Another neglected voice: many startups still choose to build in-house or use open-source solutions (vLLM, Ollama, local deployment), not spending big on cloud. Google didn't mention this side.
Impact on regular people
For enterprise IT: Companies spending on AI will find it increasingly hard to "order à la carte"—cloud vendors will push bundled procurement, so remember to break out and compare prices when negotiating.
For individual careers: The "workhorse model" concept is worth remembering—AI tool procurement going forward will be a combination of "expensive + cheap," not a one-size-fits-all approach using the priciest.
For consumer markets: Cases like Arya Health's "AI handling administrative work for doctors" mean that next time you call to book a hospital appointment, the one answering might be AI. If you care about human contact, ask proactively.