Six months on a Lightning Network node, by someone who is not a Bitcoiner
Six months, 59 payments, and $112 in fees on a self-hosted Lightning node. Notes from someone who came to Bitcoin's L2 with no ideological priors and an Ethereum background.
I bought a Start9 Server One on 14 December 2025 because a friend who runs an Ethereum infrastructure company told me that everyone in crypto should run at least one node they cannot easily turn off. Six months and roughly 220 GB of disk later, my node has routed 59 payments — 47 outgoing, 12 incoming — held an average of 4.2 million satoshis across nine channels, paid $112 in on-chain channel-open fees, earned $0.43 in routing fees, and given me an unreasonable amount of opinion about how the Lightning Network actually works. I am not a Bitcoiner. I came to Bitcoin’s L2 with no ideological priors and an Ethereum-shaped intuition about what L2s are for. This is what the experience taught me.
What is at stake in this kind of personal account is whether Lightning is a real consumer payments network, a developer-and-power-user curiosity, or a third thing that is neither but that I had to spend six months running a node to understand. The answer, I think, is the third thing, and the third thing is more interesting than the first two. Lightning is a settlement medium for a particular kind of small, low-trust, internet-native transaction that has not had a good payment rail before. It is not a replacement for Stripe. It is also not a replacement for nothing. The shape of that “third thing” is what this piece is about.
The hardware decision and what surprised me
I considered three options before buying: a Start9 Server One, an Umbrel Home, and a self-built RaspiBlitz on a Raspberry Pi 5 with a 2 TB NVMe SSD. The Start9 won on three criteria: the StartOS package manager is genuinely well-engineered, the company’s stance on user sovereignty is uncompromising, and I did not want to spend a weekend flashing SD cards. The Umbrel was a close second; its UX is arguably better but its app-store model felt closer to Apple than I wanted from a self-custody device. The RaspiBlitz would have been cheapest and most educational, and I regret not building one as a second node alongside the Start9.
What surprised me about the hardware was how aggressively the initial block download (IBD) of Bitcoin Core dominated the first three days. The node downloaded, verified, and indexed roughly 580 GB of historical Bitcoin data on a residential fibre connection over 71 hours. The CPU was pinned at 60–90% for the entire period. The fan, on a fanless Start9, did not exist; the device simply got hot. Once IBD completed, the resource profile fell to almost nothing — 4–7% CPU, 1.4 GB of memory — and the device has been silent since. I have not touched it physically in five and a half months.
Opening the first channel was not pleasant
Channel-open transactions are regular on-chain Bitcoin transactions that establish a 2-of-2 multisig output. They cost what on-chain transactions cost, which on the day I opened my first channel was 28 sat/vB at the bottom of the mempool.space fee curve. My first channel cost roughly $4.20 to open. My second cost $11, my third $7, my fourth $3.10, and so on. Over six months I have paid $112 in on-chain channel-open fees, which is a meaningful onboarding cost for a network that markets itself on cheap payments. The fees are not pathological — they are the price of having a non-custodial position on a public network — but they are the largest line item in my experience by an order of magnitude.
| Channel | Peer | Capacity (sats) | Open fee (USD) | Lifetime payments |
|---|---|---|---|---|
| 1 | ACINQ | 500,000 | $4.20 | 14 |
| 2 | LNBig.com | 1,000,000 | $11.00 | 9 |
| 3 | Bitrefill | 750,000 | $7.00 | 21 |
| 4 | WalletOfSatoshi | 400,000 | $3.10 | 6 |
| 5 | Voltage routing node | 1,500,000 | $14.50 | 3 |
The peer selection was deliberate. I opened my first channel to ACINQ because their node is among the best-connected on the network and because BOLT spec implementations matter; ACINQ’s Eclair was one of the three reference clients. I opened to Bitrefill because I actually wanted to spend sats on gift cards and a direct channel guaranteed routing. I opened to Voltage because I wanted to see what a routing-focused peer looked like. Channel selection is the part of Lightning that most resembles building a small network rather than using a payment app, and it is the part that consumer-facing wallets correctly try to abstract away.
What I actually spent sats on
Of the 47 outgoing payments over six months, 18 were Bitrefill gift cards (Amazon, Steam, an Uber credit, three coffee-shop top-ups). Eight were Stacker News tips. Six were sub payments to Nostr accounts I follow. Five were domain renewals through Njalla. Four were Voltage Cloud account top-ups for a hosted node experiment that I eventually abandoned. Two were to a Wikipedia donation page that accepts Lightning. The remaining four were experiments: paying for a single API call, sending a tip to a podcast, settling a coffee bet with a friend who also runs a node, and once paying for a Mullvad VPN renewal. The amounts ranged from 21 sats (a tip) to 250,000 sats (a Bitrefill card).
What this list does not include is anything resembling traditional commerce. I did not buy groceries. I did not pay rent. I did not split a restaurant bill. The reason is structural: the merchants and venues I interact with in the physical world do not accept Lightning, and even those that do (a small number of cafes in Berlin and Lisbon) treat it as a curiosity rather than as a real payment method. Lightning’s product-market fit, in my six months of use, is unambiguously internet-native and almost always low-value: tips, micropayments, gift-card arbitrage, and the small economy of crypto-adjacent services that have chosen to integrate the BOLT-11 invoice format.
The routing-fee economy is barely an economy
I left my node configured to route payments for other users from day one because the marketing materials suggested this was a meaningful source of yield. Over six months my node earned 43 cents of routing fees, all of it from a single corridor between my Voltage channel and my Bitrefill channel. The corridor was active for about eleven days in late February when something I do not understand caused a brief surge of routed volume. Otherwise my node has been a payment endpoint, not a routing hub. The routing-fee economy on Lightning is dominated by a small number of professionalized operators — the LNBig nodes, the LightningNetwork+ swap circles, the institutional liquidity providers — and a casual home node will not meaningfully participate.
- Channel reserve requirements (1% of capacity) sit on the channel for its lifetime, which means even small channels permanently lock up a few thousand sats.
- Force-closure events are rare but expensive: I have not had one, but the worst-case fee is a sweep transaction at whatever the prevailing on-chain rate is.
- Backup discipline matters: a missed channel backup (the static channel backup file, “SCB”) can mean losing the funds in that channel if the node disk dies.
- The Tor onion service that the Start9 exposes by default makes the node addressable from anywhere without port-forwarding, but Tor latency adds noticeable delay to invoice generation.
What it felt like, six months in
Running a Lightning node is, in the most accurate description I can give, like running a very small ISP for yourself. There is a setup cost that is not trivial. There is a discipline cost that is small but real: I check on the node maybe once a week, glance at the channel balances, occasionally rebalance using a circular payment to redistribute liquidity from outbound-heavy channels to inbound-heavy ones. There is almost no per-payment cost; once the channels are open and balanced, sending a payment takes about 800 milliseconds and feels instantaneous. The thing that feels different from any other payment rail I have used is the absence of any third party in the transaction record. There is no merchant processor. There is no card network. There is no acquiring bank. The payment goes from my node to the recipient’s node and the only record is the HTLC on the channels in between.
That feels meaningful in a way I did not expect. Most of the time it does not matter; tipping someone 21 sats is not an activity that requires constitutional protection. But the architecture of payment without intermediary is interesting precisely because it is rarely available in real life. The closest analogue is cash, and cash does not work online. Lightning works online. That is the thing it does that no other payment rail does, and after six months of using it casually I find I value that property more than I expected to. The macro context for how this fits into Bitcoin’s broader cycle is on our market dashboard, and our halving tracker covers the underlying on-chain economics that determine when channel-open fees become punitive.
Would I recommend it
To a developer, an infrastructure operator, or anyone with professional curiosity about how Bitcoin’s L2 actually behaves, unambiguously yes. The Start9 plus a 2 TB SSD is about $600 of hardware, the time investment is real but bounded, and you will learn things about payment systems that you cannot learn any other way. To a normal consumer who wants to pay for coffee with Bitcoin, no — Wallet of Satoshi or a similar custodial Lightning wallet will give you 95% of the practical utility for 0% of the operational overhead. The middle case — the person who wants to receive payments for their writing or their software or their podcast without giving a third party custody of the resulting funds — is the one for whom a self-hosted node makes the most sense, and that case is rare enough that the network’s growth depends on someone deciding the rarity is worth solving.
My node is still running. It will keep running. I will probably open another channel or two before the end of the year, mostly to see what happens when on-chain fees spike during the next halving cycle and channel opens become genuinely expensive. The thing I most want to test in the next six months is whether splices (BOLT-7 dynamic capacity changes) materially improve the experience of long-lived channels, and whether the Taproot Asset Protocol meaningfully changes what Lightning can carry. Neither of those experiments will result in 47 outgoing payments; they will result, if I am lucky, in a small further increment of understanding. For tracking the next halving’s on-chain dynamics in real time, our halving tracker updates block-by-block, and the next Lightning developer summit is on our events calendar.