Answer
Do not need it on every request: serve the last good value from a cache, fail over to a second provider, or accept the work and queue it, so it finishes when the provider is back. Without such a fallback, your availability can never exceed the provider's.