Cloud Choice
One of the choices you need to make as a startup is the cloud provider you are going to use. Picking the right provider early can save months of churn.
TLDR: I picked Google Cloud. But here's how I made that choice.
Familiar and Reliable. I explored specialists like Cloudflare Workers, Tailscale Tunnels, and CoreWeave GPUs. They’re great at specific jobs but not full‑stack or globally ubiquitous. Most of all, as I am working with different engineers and development teams, using a cloud service provider that all engineers have experience with is probably my most important criterion. For most dev teams, experience usually clusters around AWS first, then GCP, and then Azure.
First-party models. Hobbes uses LLMs as a core layer. What's interesting in the current cloud landscape is that each hyperscaler now anchors around its ‘house’ model—Gemini on GCP, GPT‑4o on Azure, Claude on AWS. So, picking a model sort of locks you into a cloud service provider because making cross-cloud calls make aspects such as security, compliance, and performance more complicated that I want starting out.
After testing the main models for performance and price, I have found that each model excels in different areas, and for my use case, GPT-4.x & Gemini 2.x work better than Claude 3.7. As an added bonus, both GPT-4.x and Gemini 2.x models are between 3x to 20x cheaper than Claude 3.7 - that is a significant factor for a startup.
So, this narrowed down my choice to Azure and GCP.
Credits. Google provides generous credits for startups, including for AI startups using Gemini. I estimate that I can run Hobbes at least for 6-9 months for "free" with the credits that Google Cloud provides. Interestingly, OpenAI does not provide credits to startups (or you have to be in circles that I am not). While Azure is good, I have found that Google and AWS have tie-ins with startup accelerators for credits, whereas Azure is not that prevalent.
So, that's how I arrived at Google Cloud as my choice - familiarity, model preference, and generosity of credits.
If you've navigated the world of cloud providers and have insights to share, specially any great startup credit programs, let me know!