"Cloud" is not only for storage; it's also for compute. Doing compute directly on encrypted data (homomorphic encryption) is very slow and very complicated, so when using a cloud, the data is usually either unencrypted, or encrypted but the key is elsewhere in the same cloud.
I get that FHE is not realistic today, but can’t I use ( if it’s really critical) a combination of confidential vms and an external hsm ? I understand I’ll be limited to traditional workloads , and not managed services though.
I asked the wrong question, what I really meant was ‘if I run in a less trusted environment, am I not supposed to use all possible crypto mechanisms available to make that environment more trustworthy , so that I can’t be deceived by my cloud operator sending my data to the us government’
Absolutely do what you can, but understand that it's futile to defend against your own cloud provider.
The only issue left would be managed services though, which then I wouldn’t use, but I’d be able to run my own postgre safely on infra I’m renting.
Intel's SGX has been broken a number of times and that should be harder to break than TDX. Like I said in my original comment though, do all the things. But if you find yourself relying on TDX to protect live(s), please pay a computer security professional to audit your security and do a threat assessment.
I’ll do all the things if ever needed, but I get that if a cloud act request happens , your cloud provider will be able to get your stuff.
I’m specialized enough in another field to know that I’m not a security person in spite of my interest in it ( I used to enjoy reverse engineering back in the days ) - I wouldn’t make that kind of decision without consulting a professional first.