tags: #publish links: [[Business Strategy and Competition]], [[Vendor Lock-in]], [[Tech Companies]], [[Software Architecture]] created: 2022-02-01 Tue --- # Vendor Lock-in and Public Cloud Platforms See [[Business Strategy and Competition]] and [[Vendor Lock-in]] for more... ## Lock-in and Public Cloud Platforms Here's an interesting take from Tim Bray: [Lock-in and Multi-Cloud](https://www.tbray.org/ongoing/When/202x/2022/01/30/Cloud-Lock-In) **Highlights:** * Acquisitions force large companies to end up multi-cloud, even if they don't want to. * Public cloud vendors brutally exploiting vendor lock-in isn't actually that bad yet, because they're competing hard and fairly customer centric * ...but we might expect vendor lock-in and price-raising to *generally* get worse in the long term, if direct competition lessens or if early customer-centric founders are replaced by value-extracting investors * Data gravity probably isn't *that* bad a lock-in - how much of the data is actually live and needs moving? * Strategies - which makes sense for which cloud consumers? * **All-in (accept lock-in).** Makes sense if you need velocity and can worry about lock-in later. * **Host-your-own (no lock-in).** Makes sense for... no-one really. You're spending your resources on hosting instead of whatever you were actually trying to do. * **Only use services that are plausibly portable to another platform (have a path out of lock-in)** (e.g. amazon-hosted offerings of standard OSS things). Makes sense for (according to Tim) non-tech companies. (Implied: doesn't make sense for tech companies. See "All-in".) * I'm not sure I agree with this even for non-tech companies? At all? He rather neglects the value-add, even for non-tech companies, from vendor-specific integrations with the rest of their offering. You'd have to scrupulously avoid them all to avoid accidentally defeating your theoretical ability to port. * But, possibly overriding all this: Tech hiring is a problem and will get worse. Multi-cloud experience is rare. This suggests using the highest-level abstractions wherever possible, regardless of lock-in, because staffing needs are lower. This might outweigh all other factors.