Bitcoin fee market, September 2026 measured: fees are 0.61% of block rewards over the last 3,024 blocks, 62% of blocks clear at a 1 sat/vB median, and ~690k daily transactions buy miners roughly $280k a day against a $39M subsidy
The number that matters: over the last 3,024 blocks (965,179 to 968,202, roughly three weeks to 22 September), mempool.space's reward stats show 57.68 BTC of fees against 9,507.68 BTC of total block rewards, a 0.61% fee share. The last 144 blocks came in at 0.72%. That sits between the 0.52% April low and the 0.69% August print Glassnode flagged as a ten-year low, so the regime is flat, not recovering. Per block, fees averaged 2.27 million sats this month (min 0.50M, max 21.1M), about $1,960 at $86,351, or roughly $282k a day. The subsidy, 450 BTC a day, is worth about $38.9M. Hashrate is 925 EH/s.
What the headlines got wrong is reading transaction counts as demand. Galaxy Research's 6 September reading of 893,391 transactions was the fourth-highest day in Bitcoin's history, up 102.6% year on year. The same day's total fees were $191,073, an average of $0.19 per transaction, with a median transfer of $34.69. Blocks are physically full (average 3.94 MWU this month, 98% of the 4 MWU cap), but full of transactions paying the floor. Across the last month's 1,421 fee-rate samples on mempool.space, 62% of blocks had a median fee rate at or below 1 sat/vB, 88% at or below 2, the average 90th percentile was 3.4 sat/vB, and 73% of blocks included sub-1 sat/vB transactions. Quantity of block space is saturated; its price is pinned to the relay minimum. The 86,667-transaction, 42.4 MvB mempool on 22 September is not an economic backlog: total pending fees are 0.088 BTC at an average 0.2 sat/vB (ChainQuery), and ChainQuery's 7-day window of 993 blocks scored 100% low fee pressure.
Composition explains the paradox. Glassnode data reported by CoinDesk in June had runestones above 600,000 a day out of 820,000 transactions, with Runes paying about 25% of fees, so roughly three quarters of transaction count contributed a quarter of a very small fee pool. CryptoQuant puts transfers under 0.01 BTC at about 80% of daily transactions, up from 44% in 2023, and OP_RETURN outputs were approaching 700,000 a day by 10 September; mempool research's OP_RETURN report finds the 2024-25 wave to be almost exclusively Runes. They consume weight without bidding. The last full UTXO census (mempool research, April 2025) counted 173.2 million outputs, 29.6% inscription-related and 49.1% under 1,000 sats, so the state that every node carries is growing with data no one will spend, at fee rates that do not pay for it.
The 2028 arithmetic follows directly. Around April 2028 the subsidy drops to 1.5625 BTC. If price and fee rates are unchanged, miner revenue falls about 49% and the fee share merely doubles to around 1.2%. To replace the lost 1.5625 BTC per block with fees on a 985k vB block would require a sustained average of roughly 159 sat/vB, against a September average of 1 to 2. Hashrate has already fallen from a 1.3 ZH/s October 2025 peak; the network is secured by subsidy and price and essentially nothing else.
What to watch: whether any Runes or inscription surge ever holds the median above 5 sat/vB for more than a few blocks; whether the monthly fee share crosses 1% or breaks below 0.5%; the ratio of mempool vsize below 1 sat/vB to above it, which tells you whether block space is being priced at all; and hashrate's path into the 2028 halving, since every EH/s that leaves is paid for by subsidy that is about to halve.
Sources (6)
AI Research
Key Takeaway
In September 2026 Bitcoin block space is saturated by Runes and dust that pay the 1 sat/vB floor, fees are 0.61% of block rewards (~$280k/day vs a $39M subsidy), and replacing the 2028 halving's lost subsidy would need ~159 sat/vB sustained, roughly 100x today's rates.
The number that matters: over the last 3,024 blocks (965,179 to 968,202, roughly three weeks to 22 September), mempool.space's reward stats show 57.68 BTC of fees against 9,507.68 BTC of total block rewards, a 0.61% fee share. The last 144 blocks came in at 0.72%. That sits between the 0.52% April low and the 0.69% August print Glassnode flagged as a ten-year low, so the regime is flat, not recovering. Per block, fees averaged 2.27 million sats this month (min 0.50M, max 21.1M), about $1,960 at $86,351, or roughly $282k a day. The subsidy, 450 BTC a day, is worth about $38.9M. Hashrate is 925 EH/s.
What the headlines got wrong is reading transaction counts as demand. Galaxy Research's 6 September reading of 893,391 transactions was the fourth-highest day in Bitcoin's history, up 102.6% year on year. The same day's total fees were $191,073, an average of $0.19 per transaction, with a median transfer of $34.69. Blocks are physically full (average 3.94 MWU this month, 98% of the 4 MWU cap), but full of transactions paying the floor. Across the last month's 1,421 fee-rate samples on mempool.space, 62% of blocks had a median fee rate at or below 1 sat/vB, 88% at or below 2, the average 90th percentile was 3.4 sat/vB, and 73% of blocks included sub-1 sat/vB transactions. Quantity of block space is saturated; its price is pinned to the relay minimum. The 86,667-transaction, 42.4 MvB mempool on 22 September is not an economic backlog: total pending fees are 0.088 BTC at an average 0.2 sat/vB (ChainQuery), and ChainQuery's 7-day window of 993 blocks scored 100% low fee pressure.
Composition explains the paradox. Glassnode data reported by CoinDesk in June had runestones above 600,000 a day out of 820,000 transactions, with Runes paying about 25% of fees, so roughly three quarters of transaction count contributed a quarter of a very small fee pool. CryptoQuant puts transfers under 0.01 BTC at about 80% of daily transactions, up from 44% in 2023, and OP_RETURN outputs were approaching 700,000 a day by 10 September; mempool research's OP_RETURN report finds the 2024-25 wave to be almost exclusively Runes. They consume weight without bidding. The last full UTXO census (mempool research, April 2025) counted 173.2 million outputs, 29.6% inscription-related and 49.1% under 1,000 sats, so the state that every node carries is growing with data no one will spend, at fee rates that do not pay for it.
The 2028 arithmetic follows directly. Around April 2028 the subsidy drops to 1.5625 BTC. If price and fee rates are unchanged, miner revenue falls about 49% and the fee share merely doubles to around 1.2%. To replace the lost 1.5625 BTC per block with fees on a 985k vB block would require a sustained average of roughly 159 sat/vB, against a September average of 1 to 2. Hashrate has already fallen from a 1.3 ZH/s October 2025 peak; the network is secured by subsidy and price and essentially nothing else.
What to watch: whether any Runes or inscription surge ever holds the median above 5 sat/vB for more than a few blocks; whether the monthly fee share crosses 1% or breaks below 0.5%; the ratio of mempool vsize below 1 sat/vB to above it, which tells you whether block space is being priced at all; and hashrate's path into the 2028 halving, since every EH/s that leaves is paid for by subsidy that is about to halve.