Because it is risky. The more esoteric the knowledge gets, the further it moves away from your core business, the more in-demand the skills are. As an example, maintaining your own metrics and timeseries storage. It takes quite a few skilled hands to do this in house and probably only feasible for larger companies anyway. Or you can simply hand this problem over to DataDog. While they are pricey, it is potentially pricier to build your own internal DataDog-like system, especially if you consider the opportunity cost of pulling your most skilled engineers to build it instead of building your product that your customers are paying for. Companies are perfectly willing to pay a premium to not worry about something, and that includes not worrying about your very skilled engineers leaving and then needing to scramble because no one else understands what has been built.
There are three big risks I see in depending on vendors:
1. You aren't average. Market forces might not align with your use case. The pricing model might change, or it might happen that you find out later it doesn't scale well for your business.
2. They might leave you suddenly. For example, all the google "products" (quotes because there is actually only one--ads) that have disappeared over the years. Even when an open source dependency suffers a cataclysmic licensing event, you can still fork it and carry on, provided you've both chosen your dependencies wisely and hired the people capable of maintaining them.
3. By choosing a vendor you're making a commitment to ossifying a part of your stack. The observability example is a good one here. At the companies I've worked for who do all their logging, metrics, alerting, etc in-house, developers aren't afraid to use the tools. The tools adapt to the requirements, whether it's cost efficiency, features, whatever. At the companies where we've used vendors everyone's perpetually afraid of increasing the monthly bill, and nobody has a say in deciding what goes on the product roadmap. To be clear, this might be the right trade.
Because it is risky. The more esoteric the knowledge gets, the further it moves away from your core business, the more in-demand the skills are. As an example, maintaining your own metrics and timeseries storage. It takes quite a few skilled hands to do this in house and probably only feasible for larger companies anyway. Or you can simply hand this problem over to DataDog. While they are pricey, it is potentially pricier to build your own internal DataDog-like system, especially if you consider the opportunity cost of pulling your most skilled engineers to build it instead of building your product that your customers are paying for. Companies are perfectly willing to pay a premium to not worry about something, and that includes not worrying about your very skilled engineers leaving and then needing to scramble because no one else understands what has been built.