If you want to compete with YT you need to basically build AWS S3 in your own data centers. You'd have to find a way to make your service run cheaper than google can if you wanted to survive. You'd have to get very scrappy and risky. I'd start with questions like: how many 9s of durability do we actually need here? Could we risk it until the model is proven? What are the consequences for losing cat videos and any% speed runs of mario64? That first robotic tape library would be a big stepwise capex event. You'd want to make sure the whole thing makes sense before you call IBM or whoever for a quote.