I strongly recommend keeping a local mirror of your dependencies... however, I've spent years maintaining such mirrors for fairly large projects (incl. Artifactory, Nexus, and one-off OSS setups like Docker registry server), and I think it's easy to underestimate how much work it is.
Whether you use expensive 'turnkey' solutions like Artifactory or keep things simple, there's just a surprising number of ways for a local mirror to go wrong, especially if you depend on it for any kind of third-party dependency compliance control.
Some repository mirrors will also become very large, which means that if you're e.g. running them in a cloud provider the bill can add up. Not really a problem on local hardware but the up-front cost of hardware can be substantial and a lot of startups have little to no in-house IT capability (e.g. the org I work with right now has reached hundreds of employees without having a single system administrator on staff, so as devops person I end up having to do the care and feeding of our recently purchased local hardware as well).
In general I think this is an important and often overlooked issue in modern tech businesses - it is amazing how many technology-centric firms like software startups get to appreciable size relying entirely on outside SaaS/PaaS providers with no real in-house IT operation. This reduces up-front and staffing cost but has a way of coming back to bite you when you hit a certain point. A conversation I've been in before, in reasonably large software outfits, is "we want actual real office phones now, but telephony-as-a-service is real expensive and the on-prem products use scary words like VLAN and QoS in their setup documentation". As someone with an IT rather than software background it's a little baffling to me how this happens, I feel like a combo sysadmin/network engineer would be an early hire. But here I am working for a company instead of running one...