There's no perfect amount, its all contextual.

Things to consider are:

ROI, if you're incurring chain costs to open the channel then that cost is more spread out the bigger the channel is.

Fee environments are a factor here too, earlier on in Lightning 1 or even sub-sat fees were not the norm.

Purchasing power has grown vs fiat, but also the average size of payments on Lightning have increased too, depending who's data you base that on.

If you're getting the channel from an LSP, you're paying to lease that capacity, so over-sizing it vs. your needs is a waste. If you're funding it yourself and initializing the open, bigger is better per point 1 but worse per point 5 below:

Lightning is inherently a hot wallet so there's added risk vs. cold storage, putting more in than you can take advantage of is tempting fate.

You should not expect an ROI, you may get some routing fees back over time to help with cost, but Lightning isn't a yield farm. Fees simply make the network function and are meant to offset your fees incurred. Unless you are offering a service as a business on Lightning that has its own network effect of customers to drive routing your goal should be to break even whilest getting the benefits of instant payments and relatively free small payments that aren't practical on-chain.

If you're a merchant, you probably want channels as a multiple of your revenue over a period of time, say monthly or quarterly so that all the smaller payments you receive can be consolidated once in a re-balance/swap.

If you're a spender, do the inverse.

If you want to actually earn money just on routing, you need patronage network effect where you're earning fees off other peoples liquidity because they open channels to you. Some nodes have done this purely on routing alone by bootstrapping network effect with large sums or simply being around and reliable for a long time, but that's the exception.