So you’re not using any third party vendor for anything and you’re doing everything in house?
Cloud is only an interchangeable commodity if you’re treating it like an overpriced colo and not using it to save costs on staff, maintenance, and helping deliver product faster.
That's why it's important to use k8s (and various abstraction layers, standards APIs, etc.), which helps with portability, and thus you can risk load balance between cloud vendors, and you can even throw in on-prem into the mix.
And why wouldn't you just do a colo then? From my experience, cloud infrastructure is always more expensive than the equivalent on prem/colo infrastructure unless you depend on hosted solutions and you're willing and able to operate with fewer infrastructure people and it doesn't help you move faster.
The last time I did the math a reasonable highly available setup was $1-3m CapEx per data center, I’d want no less than three. That’s 30kw worth of gear per for a total of 90kw at ~$180/kw MRC if I’m lucky plus transit fees, so $20k a month.
Doable, but it’s a hell of a lot of hassle and that CapEx is huge for a startup.
I’d go bare metal in a second for any kind of cost conscious business that needed scale and had established revenue.
In the case of a startup, the question is just the opposite. It’s about moving fast and being nimble more than it is about worrying about a distant future where lock-in could possibly be an issue. I would be all in on leveraging as many of the services that the cloud provider offers and where they could take care of the “undifferentiated heavy lifting”.
Cloud is only an interchangeable commodity if you’re treating it like an overpriced colo and not using it to save costs on staff, maintenance, and helping deliver product faster.