The basic idea is to create a “channel” between 2 participants (let’s say Alice and Bob) and place some funds in, for example 1 BTC each. This requires a transaction on the Bitcoin blockchain, that locks the funds in a two-signature multisig address. Here is the situation of the open channel:
Alice <= 1 BTC == 1 BTC => Bob # opening of a new channel
Alice and Bob can now make as many transactions as they want “off-chain”, they just need to exchange an updated “balance-sheet”: that is a transaction signed by the other party (so it’s missing one of the signatures) that is not propagated to the bitcoin mainnet (yet), and that pays the updated funds to the participants.
Because the transactions are stored “off-chain”, Alice and Bob do not need to wait for a confirmation from the main blockchain (instant transaction) and also do not need to pay a fee to the mainnet miners (low fees).
So, some nodes perform the crucial role of routing transactions through multiple hops in order to allow any sender to transfer Bitcoin to any buyer. As running a lightning node incurs a cost (however, not as high as running a Bitcoin node), nodes that route transactions usually charge a fee for performing routing services. The cost of running a node is approx. around $4-5/per month (mostly electricity expenses) and the revenue from fees is only a miniscule fraction of that currently. Anecdotal evidence shows that the average revenue is around 200 sats which is worth less than a dollar. The economics of running a lightning node is unviable in its current state. This is expected to change as LN becomes more widely adopted.
The reddit thread below has some interesting discussions around the economics of lightning network.