I don't know. I'm just curious. I do know that channel selection for optimizing routing is sometimes viewed as an art more than a science. LNDg makes suggestions, or the autopilot like Rene Pickhardt's, so I was wondering whether any tools utilizing AI might be in the works.
From Amboss:
using AI-driven insights derived from over five years of network data.
I'm not discounting the possibility that this is just marketing drivel.
Even if they have a massive tensor file with five years of captured data, past performance isn't a guarantee for the future, especially not in a network like LN that has been rapidly evolving over these same five years. Are patterns from early LN still valid? I don't think so, but happy to be proven wrong.
tools utilizing AI might be in the works.
I think that there could be useful scenarios, but remember that LN still has known weaknesses in the channel closure process - which is why you run watchtowers - so you need to make a judgement call every time you open a channel. I'd not leave using judgement to something that literally has no capacity for judgement 1
Footnotes
unless there's nothing at stake for you and losing a couple M sats is a drop in the bucket - a luxury I used to have on testnet3, but not anymore since I gave my testnet sats to a colleague in 2018. ↩
Thank you for the information! I'm getting interested in LQwD. It seems that LQwD has not yet settled on a stable business model. According to the news on its website, in recent years, LQwD has explored routing nodes, AI-driven channel management optimization, LSP, and Bitcoin reserves. However, when looking at its financials, the revenue for the 2024 and 2025 fiscal years is only 506 CAD and 5553 CAD, respectively, which suggests that it might not be fully prepared for market entry. Additionally, there have been considerable fluctuations in its stock recently, which has also sparked my curiosity.
Well, from my point of view, their node management is pretty bad so I was more incline to see these as a classic speech to pump the bags "We have more channels than anyone in LN", "We reach every corner of the planet with our nodes"; in case you heard something like that... run away from the stock, or better, from fiat ;).
Plot twist: It's only a facade to sneak into LN and get information for some big brother out there.
Totally — AI could be a game changer for routing nodes.
Think: smarter fee setting, auto-rebalancing, ditching deadweight peers. Most of us still do this by hand or run basic scripts. A model trained on your own node data could easily outperform that.
Still early, but huge potential. Someone just needs to build it.
In my view, this is a living thing, you can "process" and "crunch" the data all you want, but in the end, it's a 0 sum game, and most of the time you will be working with old photo of the network.
afaik machine learning (aka AI) is part of how CashApp is making 10% yield. They have a dedicated team of engineers working on optimizing their node. None of that is open source afaik.
Also, I won't name names, but there's a lightning company that released a lightning wallet in the last few years (targeted at latam) that they released for the sole purpose of gathering data that they could use for machine learning purposes.
Some nodes in non-replicable positions, based on their affiliation or proximity to a service and scale do so, but those are the exception. Routing isn't a business, its an extension of a business.
Routing fees in Lightning are simply to make the system work, it is 0 sum... your fees earned offset fees paid.
As a permissionless system that anyone can enter, your dumb capital has no advantage... you'll be blindly hoping to arb something opaque.
So instead of buying a product claiming to give you magic data to direct your dumb and entirely un-special capital, OFFER a product which attracts throughput on your node, the routing fees can then extend your margins on that. Maybe that product is a loss leader in and of itself designed just to generate routing traffic.
Footnotes