The 39-Basis-Point Week: What Just Happened to Staking Yield

1. The number

The network that pays Ethereum's stakers paid them 2.18% this week — on the same weekly series that read 2.62%, 2.57% and 2.57% across the prior three prints. A 39bp single-step fall, between consecutive prints, is the largest move this site has recorded in nine weeks of tracking the series. The same print's companion series — the stETH 7-day average APR — fell 8bp, from 2.60% to 2.52%. Two series from one source, moving 39bp and 8bp in the same week. This report is about what can honestly be said about that asymmetry, because the tempting conclusions — the yield regime has broken, or the print is wrong — are both ahead of the evidence.

Source: Ebunker Ethereum staking weeklies (the last four editions), the same series this site has tracked since edition three.

2. The series, and what one source is worth

The record first. This site's network APR series changed source in mid-September — from validatorqueue (2.61% on Sep 2, 2.59% on Sep 10) to the Ebunker weekly (2.62%, 2.57%, 2.57%, and now 2.18%) — a change we disclosed at the time with the note that a step between providers is not a move in the market. That caution now runs in both directions. The 2.18% is one provider's weekly print, assembled from that provider's own validator fleet performance and network accounting, and this site cannot independently re-derive it: the beacon-chain explorers that publish the underlying quantities were not reachable this week. What we can do is exactly what this section does — bound the print with arithmetic that does not depend on trusting it.

Source: series as cited in section 1; the source change was disclosed in edition five's rate board; underlying data endpoints (beaconcha.in, validatorqueue) returned access errors on Sep 29.

3. What stake growth explains: about 2bp of the 39

The first candidate cause is the mechanical one, and it is testable with two numbers already on this page. Network issuance is effectively fixed per epoch; the yield per staked ETH therefore falls roughly in proportion to how much ETH is staked. Between the last two weekly prints, total staked ETH rose from 43,349,255 to 43,681,037 — +0.77%. Mechanical dilution alone would take a 2.57% yield to about 2.55%: a 2bp move (DERIVED: 2.57 ÷ 1.0077). The observed step is 39bp — roughly twenty times what the week's stake growth can produce. The same arithmetic run over the whole series makes the point starker: staked ETH has grown about 1.8% across the four prints in which the yield went 2.62 → 2.57 → 2.57 → 2.18. Supply growth is the quiet background of this series, not its plot.

Source: weekly staked totals per the Ebunker weeklies cited in section 1; issuance per Ethereum protocol parameters; dilution arithmetic our own, shown so any reader can redo it.

4. What the rest could be — and what the evidence will not yet say

Three candidates survive the dilution test, and the evidence ranks rather than resolves them. Execution income. Staking rewards are issuance plus tips and MEV; the issuance half is stable, the execution half is a market and can fall sharply without any change in stake. A week with thinner block-level tip activity would print exactly this shape — a network APR down sharply while a smoothed 7-day average falls far less. Methodology. The series changed source fourteen days before the step; if the new provider's sampling window or fleet weighting shifted, a 39bp step could be partly an artifact wearing a market's clothes. Both. The honest position is that the 8bp move in the same print's smoothed series is consistent with a real but smaller underlying fall plus smoothing, and equally consistent with a noisy print. What would settle it: the next weekly print. A second read near 2.18% makes a regime; a reversion toward 2.5% makes this a footnote with a lesson about single sources.

Source: reward-mechanics per Ethereum protocol documentation; the two series and their asymmetry per the Ebunker weekly (Sep 28); the ranking of candidates is our assessment, labeled as such.

5. The carry consequence: every record, reset by one print

Last week this column printed a set of record cash-versus-staking gaps and attributed them to the Fed. This week the records were reset — by the yield leg, with the Fed on hold:

All four moves are DERIVED and all four are the same number, because one leg moved and three did not. The pre-committed translation table this column has used since the FOMC — every 25bp of policy change moves the gap 25bp — now runs in reverse: a single weekly print moved every gap on this page by 39bp without a committee meeting. If the print is real, the squeeze on staked ETH is two-sided for the first time in the series: the reference rates rose in September and the yield is falling. If it is not, all four records retrace together. There is no third arithmetic.

Source: rates as cited in sections 1–2 and the data report's corridor section; FOMC decision and SEP (Sep 16, 2026); all gaps DERIVED.

6. The price paradox: capital arrived while the coupon shrank

The same window that produced the 2.18% print produced: ETH +5.1% to $2,709.26, ETF inflows of +$689.8M across five sessions, Lido TVL +2.9% to $26.47B, and staked ETH +0.77% on the weekly print. Every demand indicator this site tracks moved up while the advertised return on the asset's productive use moved down. Two readings fit, and we hold both loosely. The first: the marginal buyer of ETH exposure — an ETF allocator, a treasury company adding to 5.98M ETH — is buying scarce supply with a coupon attached, and the coupon's level is a second-order term at current sizes. The second: the yield print is wrong or overstated, and the market is efficiently refusing to price an artifact. The week's data cannot choose between these. The next print can, which is why section 8 turns both readings into dated markers rather than opinions.

Source: price, flow and TVL data as cited in the week's news review and data report; Bitmine totals per the Sep 21 disclosure via Cryptopolitan.

7. What a real re-rate would mean for the numbers this site carries

If the next print confirms a ~2.2% network regime, the consequences propagate through every layer this column tracks, in order. The user-net band (2.21–2.24%, carried since Sep 2) becomes untenable as a carry basis — a staker cannot keep more than the network pays for long, and the band would follow the network print down by roughly its historical wedge. The frontend APR (2.40%, carried Sep 7) reprices with it. The carry tables re-widen again: against a carried 3.92% bill, a network yield of 2.18% is already 174bp, and a user-net that settles near 1.9% would put the retail gap past 200bp (DERIVED, illustrative). The competitor board re-ranks: a carried 2.19% rETH mid-point would sit above the network's fresh print for the first time, and the "decentralisation premium" argument CMv2 just sharpened becomes harder to sell at a yield deficit. And the queue becomes the tell: a month-long entry line in front of a falling yield is either sticky conviction — treasury companies do not read rate boards — or a lag that resolves on the downside. Each of these is a number this site already publishes, which is the point: a regime change here is observable in our own pages, not in someone's forecast.

Source: carried rates and board reads as published in the rate board and data report of this edition; wedge and illustrative gap arithmetic our own.

8. Three scenarios, each falsifiable

The base case is the middle one for an unglamorous reason: of the three candidates in section 4, a fall in execution income is the one with market precedent and the right size, and smoothing's 8bp response is what a real but partially-buffered drop looks like. We hold it loosely — the whole point of this section is that one more print outranks any argument we can make.

Source: scenarios are our own framework over the series cited in section 1; no external forecast is used.

9. Risks to this reading

We are featuring a single print. The entire report hangs on one weekly observation from one provider, fourteen days after this series changed source. Every headline number in section 5 inherits that fragility, which is why the corridor section of the data report prints pairs — on the carried user-net band, the gaps are 168–171bp, not 174bp.

The two series inside one source moved differently. A 39bp network step beside an 8bp smoothed step is consistent with our base case and with source noise in nearly equal measure; we have ranked the candidates but cannot weight them from outside.

The underlying quantities were not independently checkable this week. The beacon-chain endpoints that would let this site re-derive total rewards returned access errors, so section 3's dilution test — the one part of this report that does not require trusting the print — is the whole of our independent verification.

The demand-side data has its own lag. Queue waits and ETF flows respond to expectations, not prints; using this week's inflows as evidence about a yield print is a correlation observation, not a mechanism.

Source: limitations are our own assessment of the sources cited in sections 1–4.

10. What we watch

Five dated checks. (1) The next Ebunker weekly — the single number that chooses between the scenarios in section 8. (2) The stETH 7-day average's path — if the smoothed series walks down toward 2.2% over successive prints, the network print was real and the pass-through has begun. (3) The first fresh read of the active validator count — frozen this week by an access failure — against CMv2's one-third reduction plan, because a re-rating yield and a re-keying validator set arriving together would compound. (4) The user-net band's first dated refresh — the number a staker actually banks, carried since Sep 2 and now older than every other input on the rate board. (5) The entry queue — four straight prints of shortening waits into a falling yield; a fifth would tell us the demand side of section 6 is conviction, not lag.

11. Conclusion

One number moved this week — 39bp — and it moved everything measured against it: four record gaps, a corridor whose carried rows are suddenly flattering, a competitor board re-ranked by a rounding error, and a question this site cannot yet answer about whether the number is real. The arithmetic in section 3 is the report in miniature: stake growth explains 2bp of the fall, so the other 37bp is either a market (execution income repricing) or a measurement, and the distinction belongs to next week's print, not to this one. What we can already say is what the week proved about the market around the print: $689.8M of ETF inflows, a 5.1% ETH week and a fourth consecutive print of shortening entry queues, all while the advertised return fell. Capital is not waiting for the yield to confirm itself. The next print will tell us whether it should have.

12. Why the step is a question, not an answer

The report's spine is one arithmetic fact from section 3: between the last two prints, staked ETH rose 0.77%, which mechanically dilutes the yield by about 2bp — yet the print fell 39bp. The other 37bp is the entire question, and this closing section states the discipline that follows. A single weekly print from one provider cannot be a regime; the next print decides. Until then, every headline number in section 5 (the four record gaps), section 7 (the propagated re-rate) and section 8 (the three scenarios) inherits the fragility of that one observation. We have said where the print could be real (execution-income repricing, which has market precedent and the right size) and where it could be a measurement artifact (a source change fourteen days before the step, an 8bp move in the same print's smoothed series). The honest position is that both are live, and that the distinction belongs to next week's read, not to advocacy. This site will print that read with its date when it lands.

Source: dilution arithmetic as in section 3; scenarios and gaps as in sections 5, 7 and 8.

FAQ

Related reading

• Lido & stETH Weekly Data Report — Week of September 21–27, 2026
• stETH Yield Update — Week of September 21–27, 2026
• 8.4M ETH on the Move: What Lido's CMv2 Consolidation Actually Changes