The protocols Lemmy uses rely on additional instances and federation to scale horizontally.
That’s kind of the point of the fediverse. There shouldn’t be big instances.

The issue is a lot of users want to be on the busy instances.
Whilst it shouldn’t matter which instance you actually join and use, some instances might have community/moderation that aligns with how you want to experience Lemmy (eg beehaw)

Lemmy hasn’t been developed for a single instance to scale horizontally. Throwing bigger hardware at it is the correct way to implement scaling when a project is this size/maturity.

Having stateless middleware, running caches, sharding databases, database replication, read/write load balancing on databases, having the actual front end load balancers…
It’s a difficult problem. Companies have entire teams that work on this, and it requires a lot of skill and attention to keep all the parts working correctly, and ensure things are fault tolerant.
Most instances are run by volunteers and community funding.

Like I said, hopefully Lemmy will move to a format that allows for easier scaling. But it’s a lot of work.
There is probably more value in squashing bugs, improving user experience, adding some well-needed features, and any optimizations they find along the way - than there is in rebuilding the stack to support horizontal scaling.

Remember, Lemmy has a core team of 2 developers.
And this massive influx of users became apparent at the start of this month.
It’s going to take some time, and things will be rough round the edges.