Two markets, one direction, two different clocks

CAISO is now a clean series

Modo's benchmark gives $80/kW in 2023 and $51 in 2024, and its 2026 review states that full-year 2025 "landed near $40/kW". The $38 in your model sits inside that, so it can stay. Against a fleet that went from about 0.5 GW in 2020 to 15.5 GW, this is a fall of roughly three quarters in three years.

The 2026 bar is a pace, and January was the wrong month to take it from

January 2026 annualised to $21/kW-year, but it was a winter-storm month. April, May and June each landed in a $2.5–2.7/kW-month band, and June came in at $2.55/kW, 16.9% below June 2025's $3.07. That band annualises to about $31/kW-year, which is what the bar now shows.

Modo has not published February or March 2026 separately, so this is a three-month pace rather than a half-year outturn.

The source mixing turned out to be harmless

Your series takes $103 for 2022 from CAISO's DMM and the later years from Modo's benchmark. The two are constructed differently, but where they overlap they nearly coincide: DMM has $78 and $53 for 2023 and 2024 against Modo's $80 and $51. The mixing does not distort the trend, though the basis is worth a footnote.

The A$148k is Modo's own figure, and the objection to it was wrong

This note previously said the A$148k for 2024 could not be reconciled, and advised against publishing the NEM series. That was a mistake. Modo's ME BESS AUS NEM Index gives A$148k/MW for 2024, a 45% rise on 2023, which implies about A$102k for 2023.

The basis is stated: all merchant revenue streams, energy from five-minute SCADA plus all ten FCAS services, with marginal loss factors applied, costing an average A$3.2k/MW/year in 2024. The A$50–70k figure elsewhere is a different basis, not a contradiction.

Why the NEM panel still is not annual

Modo has published no full-year 2025 for the NEM. What exists after October 2025 is the six-month window at A$73k/MW/year, down 38%, and annualised months: A$54k in February, A$35k in June, A$44k in July.

The July rebound was two evenings

July 2026 rose 25% to A$44k/MW/year, and price events on 8 and 30 July contributed A$9k of it — a fifth of the month from two evenings. The 8 July event was winter demand meeting wind at a tenth of its Victorian average, with prices reaching A$19,070/MWh in Victoria and A$16,971 in South Australia.

Victoria earned A$53k/MW/year, up 80%, and led the states for the first time since May 2025; South Australia fell 17% to A$33k. The month was still 72% below July 2025 and just over half the twelve-month average of A$85k.

What is coming at the supply side

Both markets fall because fleets grow and compete the spread away, so what happens to cell costs matters. On 17 July 2026 China’s finance ministry ended eleven years of exemption: lithium-ion carries a 2% consumption tax from 1 September 2026 and 4% from 1 September 2027, while sodium-ion, solid-state and fuel cells stay exempt until the end of 2028. That is a 28-month window in which one chemistry is taxed and its substitute is not, in the market that builds most of the world’s storage.

Two cost claims travel with it and they are not the same claim. On price, Chinese 314 Ah LFP storage cells averaged CNY 0.365/Wh in April 2026 (range 0.335–0.395) against sodium quoted at 0.32–0.50 — overlapping ranges, and CATL says it expects cost parity with LFP by the end of 2026, which is a statement that the cells are not cheaper yet. The separate “~12% below LFP” is a levelized cost of storage: lifetime cost over lifetime throughput, which a dearer cell can still win on.

But the mechanism for that has closed. A levelized win at an equal or higher cell price needs more cycles, better round-trip efficiency, or cheaper balance of system. Sodium’s headline is 10 000 cycles and LFP storage cells are now quoted at 6 000–10 000, so the cycle term is parity. Lower energy density means more steel, land and wiring per stored kWh, pushing the other way. No assumptions are published with the 12%.

Worth noting what the LFP price is doing underneath: those cells were CNY 0.300/Wh in late October 2025 and rose about a fifth in six months, so any sodium comparison drawn now is drawn against a rising baseline.

Structural and cyclical, kept apart

Both markets are falling for the structural reason: fleets grew and the arbitrageurs compete away the spread they live on. The NEM fleet more than doubled from 2 GW to 4.6 GW during 2025 alone, and Modo attributes the June and July 2026 level — 15% of last winter's average — to fleet saturation as much as to weather.

The CAISO mechanism is visible in the prices. In June 2026 the 2pm day-ahead trough averaged $7/MWh against a 7pm peak of $35, where a year earlier the pair was $16.55 and $52. Negative-price hours tripled to 66, and spring solar curtailment averaged a record 4.8 GW. The spread the battery lives on is being squeezed from the top, not widened from the bottom.

AEMO now measures the spread itself

Everything above infers cannibalization from revenue. AEMO’s Quarterly Energy Dynamics for Q2 2026 measures the mechanism directly: the NEM-wide battery price spread fell from A$342/MWh to A$51/MWh in a year, −85%. That is the operator’s own number for the quantity this page argues about, and it moves the NEM case from inferred to observed.

The fleet grew into it. Capacity rose 4,640 MW/12,353 MWh between Q2 2025 and Q2 2026, more than doubling to over 9,000 MW, while average discharge went from 162 MW to a record 476 MW — almost triple. Charging rose 371 MW. AEMO’s Q2 2025 base was therefore about 4.4 GW, which is not the 4.6 GW quoted above for end-2025: different counts from different sources at different dates, not one series. More assets, all arbitraging the same hours, and the hours stopped paying.

The revenue arithmetic behind the spread

Net battery revenue fell to A$57.5 million, down A$73 million on Q2 2025 — so the year-earlier figure was A$130.5m. Almost all of it is arbitrage: net energy arbitrage revenue fell A$68.1m (−56%) to A$52.8m.

The interesting part is that it did not fall because they sold less. Gross energy revenue fell A$44.5m to A$106.2m while charging costs rose 79%, from A$29.8m to A$53.4m. Discharge VWAP fell A$326/MWh, to A$101/MWh. Volumes went up and the margin on them collapsed from both ends at once.

Two FCAS numbers that are not the same number

Battery FCAS revenue fell A$5.0m (−51%) to A$4.8m — yet FCAS rose to 8.3% of battery revenue, up 0.8 pp, because arbitrage fell faster. Separately, total NEM FCAS costs across the whole market were A$9.2m, down 61% year-on-year and about 0.3% of the cost of consumed energy.

These get conflated in secondary coverage. The first is what batteries earned; the second is what the market paid for frequency control in total. Both fell, for the same underlying reason, but a 51% and a 61% decline are answers to different questions.

What is queued behind it

Of 75.4 GW progressing through the NEM connection process, 39.6 GW is battery — more than half the pipeline, and 32.9 GW of that is grid-forming. A record 6.9 GW of generation and storage connections were approved in the quarter alone.

Nothing in the spread data suggests that queue will find the spreads it was underwritten against. That is the honest reading of an 85% compression in twelve months: the returns are being competed away faster than the assets arrive, and the next margin has to come from dispatching better rather than from owning more.