Choosing Between GUILD_MEMBERS Intent Caching and API Fetching for Large-Scale Discord Bots

Context

When deploying a discord.js bot across hundreds of guilds, the GUILD_MEMBERS privileged intent determines whether the library maintains a local member cache. Enabling it provides immediate O(1) access to member objects via guild.members.cache, but memory consumption grows linearly with total member count across all guilds. Disabling the intent forces reliance on guild.members.fetch() or client.users.fetch() for individual lookups, eliminating the cache footprint but introducing network round-trips and exposure to Discord's REST rate limits.

Constraint

The bot must handle frequent member-resolution operations (role assignments, moderation logs, presence checks) in guilds ranging from 50 to 150,000 members. Memory headroom on the host is limited to 1.5 GB for the Node.js process, and the bot must remain responsive during peak traffic without triggering 429 errors.

Open Questions

1. At what approximate member-count threshold does the GUILD_MEMBERS cache typically exceed the 1.5 GB budget under discord.js v14?
2. Can a hybrid approach—enabling the intent but periodically sweeping the cache with guild.members.cache.sweep()—provide a stable middle ground, or does the library rebuild the cache aggressively enough to negate the benefit?
3. Are there documented patterns for pre-warming only high-priority guilds' member caches while leaving others to API fetching?