This note follows NEM price spikes from $300/MWh to the price cap: how many escalate, how fast they get there, how long high prices last, and what bounds them.
- Escalation is a cliff, not a ramp. 7.5% of spike episodes reach $1,000/MWh, but 48% of those go on to $5,000. Episodes reaching $5,000 are 3.6% of all episodes and carry 77% of the dollars paid above $300.
- The price often gives no warning. 36% of episodes reaching $5,000 are already there in their first interval, and 63% go from below $1,000 to $5,000 or more in a single 5-minute step.
- Two kinds of episode reach $5,000. Most that start at $5,000 last one interval (101 of 164). Those that get there later build and persist: a $5,000 interval in a continuing episode is followed by another 71% of the time.
- The range between ordinary prices and the cap is nearly empty. Of prices at or above 10% of the cap, 64% are at or above half of it. In South Australia a median of only 90 MW is offered between $300 and $1,000/MWh, against standing offers near the cap from every dispatchable unit.
The plate above shows one example. On 12 June 2025 the price in all five regions was between $959 and $1,185/MWh at 17:05 and between $9,487 and $11,017/MWh at 17:15. Every region stayed above $3,300/MWh until 20:00, and Victoria cleared at the $17,500/MWh price cap twice.
The data are AEMO 5-minute regional prices from 1 October 2021 to 31 May 2026, 490,751 intervals per region. A spike episode is a run of consecutive intervals with price above $300/MWh. There are 12,794 of them.
of spike episodes reach ≥$1,000/MWh (959 of 12,794).
of episodes reaching $1,000 also reach ≥$5,000/MWh (460 of 959).
of episodes reaching $5,000 are already ≥$5,000 in their onset interval (164 of 460).
Only 7.5% of spike episodes reach $1,000, but 48% of those go on to $5,000
Finding. Of 12,794 episodes, 959 (7.5%) reached $1,000/MWh in at least one interval. Of those 959, 460 (48.0%) reached $5,000/MWh. Of those 460, 153 (33.3%) reached at least 98% of the market price cap in force. The largest proportional drop is between $300 and $1,000. Above $1,000 the count halves at $5,000 and falls by two-thirds again at the cap.
Derivation. We took each episode's peak 5-minute price from the episode catalogue and counted episodes at or above each threshold. The pattern holds in every region. The share reaching $1,000 ranges from 2.7% (Victoria) to 12.1% (New South Wales). The share of those that go on to $5,000 ranges from 37% (Victoria) to 51% (New South Wales).
Episodes reaching $5,000 carry most of the money. They are 3.6% of episodes and 16.4% of all intervals above $300. Yet they account for of all dollars paid above $300 on a $300 cap contract, summed over the five regions. By region the share ranges from 66% (Victoria) to 81% (South Australia).
Interpretation. A steep drop at $1,000 followed by shallower drops is consistent with offer stacks that hold little capacity between about $1,000/MWh and the high bands near $10,000/MWh and the cap. Once demand exceeds the capacity offered below $1,000, the clearing price moves to much more expensive offers. Chapter 07 gives the bid-data evidence.
35.7% of episodes reaching $5,000 do so in their onset interval
Finding. For 164 of the 460 episodes that reached $5,000 (35.7%), the onset interval was already at or above $5,000/MWh. The onset interval is the first interval above $300. Another 61 episodes reached $5,000 in their second interval, 5 minutes after onset.
Measured from the first interval at or above $1,000, 288 episodes (62.6%) were already at or above $5,000 in that same interval. The median time from onset to the first interval at or above $5,000 was 10 minutes. 58 episodes (12.6%) took an hour or more.
The two groups differ in duration. Of the 164 episodes already at $5,000 in the onset interval, 101 (61.6%) lasted only that interval. Only 25 (15.2%) were still at or above $5,000 five minutes later. Of the 296 that reached $5,000 later, 163 (55.1%) were still above $300 an hour after onset.
By financial year, the share of $5,000 episodes already at or above $5,000 in their onset interval was 41.7% (FY22), 31.3% (FY23), 27.1% (FY24), 32.3% (FY25) and 52.5% (FY26, July 2025 to May 2026). The share at or above $5,000 in their first interval at or above $1,000 was 57–65% from FY22 to FY25 and 80.0% in FY26.
Interpretation. Two kinds of episode reach $5,000.
- Single-interval episodes. Most episodes already at $5,000 in the onset interval last only that interval. That fits a short supply interruption, or a dispatch outcome reversed in the next 5-minute run, for example after fast-start units or batteries respond, or after a rebid.
- Building episodes. Episodes that reach $5,000 later build over several intervals and often last more than an hour. That fits a sustained shortfall of offered capacity.
When the onset interval is already at or above $5,000, the episode contains no lower-priced interval in which a price-triggered response could act first. Warning for these episodes has to come from forecasts of supply and demand rather than from the price series. The FY26 increase rests on 40 episodes, and its 90% interval is wide (Fig. 4).
71% of intervals at $5,000 or more are followed by another 5 minutes later
Finding. Take a 5-minute interval at or above $5,000/MWh, in a region that is still above $300 at the later interval. The later interval is also at or above $5,000 with probability 70.8% five minutes later, 54.2% after 30 minutes and 43.4% after 60 minutes.
Counting cases where the episode has ended by the later interval, the probabilities are 64.7%, 43.0% and 28.9%. From a spike interval between $300 and $1,000, the probability of reaching $5,000 or more at a later spike interval is 0.4–1.0%. The average across all spike intervals is 2.5%.
Most intervals at or above $5,000 belong to the building episodes of Chapter 02. Each single-interval episode contributes only one.
Derivation. We used all pairs of intervals in the same region 5 to 120 minutes apart. We excluded the June 2022 market suspension and intervals under administered pricing. There are 74,408 pairs 5 minutes apart in which both intervals are above $300. The rank correlation of price between two such intervals is 0.84 at 5 minutes, 0.60 at 60 minutes and 0.46 at 120 minutes.
Interpretation. High prices persist because the high-priced offers remain marginal while the supply shortfall lasts. Rebids, fast-start units and changes in interconnector flow usually need more than one 5-minute interval to change the balance. A risk measured over many intervals, such as a cap contract, should treat a $5,000 interval as the likely start of a block of high prices, not as an isolated value.
Most episodes below $1,000 last one interval; one in nine that reach $5,000 lasts more than four hours
Finding. There are three peak bands:
- Below $1,000: 11,835 episodes, median duration 5 minutes; 53.8% last a single interval.
- $1,000 to $5,000: 499 episodes, median 15 minutes.
- $5,000 or more: 460 episodes, median 35 minutes. 52 of them (11.3%) lasted more than four hours.
Derivation. Duration is the number of consecutive intervals above $300, multiplied by 5 minutes. The peak band is set by the highest interval in the episode.
Interpretation. Single-interval episodes below $1,000 fit brief mismatches, such as ramping or a short constraint, that the next dispatch run clears. Long episodes at high prices need a sustained shortfall of offered capacity. Because duration and price rise together, the few long, high episodes carry most of the exposure, as Chapter 01 measured.
Three in four evenings that reach the cap are otherwise ordinary evenings with a short spike
Finding. We grouped 2,008 region-evenings by the shape of their price path from 16:00 to 21:00. Each was an evening on which a spike episode began in that window. Five of the six clusters have median peaks of $378–426/MWh.
The sixth, 72 evenings, has a median peak of $12,844/MWh and averages 143 minutes above $300. 21% of its evenings reached at least 98% of the cap. Another cluster of 57 evenings, 43 of them in South Australia, starts at a median of −$4/MWh at 16:00 and rises to about $280 by 18:30.
The high-peak cluster holds 15 of the 62 evenings that reached at least 98% of the cap. The other 47 sit in clusters whose median path stays near $300.
Derivation. Prices were transformed as asinh(price/30), and each evening's mean was removed so the clusters reflect shape, not level. We used k-means with six clusters, ten initialisations and a fixed seed. Six is a modelling choice, not a property of the data.
Interpretation. The common shapes follow the evening demand peak as solar output falls. The cluster that starts below zero in South Australia fits afternoon oversupply from rooftop and utility solar, followed by the evening ramp. Cap-reaching evenings in clusters whose median stays near $300 fit short price spikes on otherwise moderate evenings.
The middle of the price range is nearly empty: most prices above 10% of the cap are above half of it
Finding. In each financial year from FY22 to FY26, the maximum 5-minute price in New South Wales, South Australia and Tasmania equalled the market price cap in force. The caps were $15,100, $15,500, $16,600, $17,500 and $20,300/MWh. Queensland's maximum equalled the cap in four of five years; its FY26 maximum was $19,728. Victoria's equalled it in three: its FY23 maximum was $14,633 and its FY26 maximum $3,026.
Intervals at or within 2% of the cap are rare: 0.014% of intervals, about one in 7,100. The range between ordinary prices and the cap is sparsely populated. Of all intervals priced at or above 10% of the cap, were at or above 50% of it (Fig. 8).
An earlier record-statistics analysis looked at South Australian and New South Wales prices since 2019. It found that every record high after January 2019 (SA) and January 2020 (NSW) came from the cap being raised. The cap rose from $14,500 to $20,300/MWh, 40% in all, and is indexed to inflation each July.
Interpretation. The upper tail of NEM prices is set by rule. Peaks collect at the cap because generators keep standing offers near the cap (Chapter 07). Statistical tail models that ignore the cap misstate this tail, and "record price" headlines largely track cap indexation.
In South Australia a median 90 MW is offered between $300 and $1,000
From earlier analysis of AEMO bid data, June 2025 – February 2026; not recomputed here
Finding. In South Australia the capacity offered between $300 and $1,000/MWh had a median width of 90 MW at the daily 17:30 snapshot, flat across nine months. This is the only capacity between the spike threshold and five-figure offers. In New South Wales the median was 833 MW.
Every dispatchable unit kept a band of capacity offered near the cap. On 12 June 2025 all 27 South Australian units did, at a median top-band price of $17,470/MWh.
Before 16 South Australian onsets with no identified physical cause, dispatchable units moved capacity into cheaper bands, a median of +96.5 MW over the hour before onset. Almost none of the offered capacity was undeliverable within 15 minutes. Those onsets happened when the offered stack was nearly fully deliverable, and it was still insufficient.
We also looked for early-warning signs in the two hours before onsets in South Australia and New South Wales. Lag-1 autocorrelation and variance of price did not rise relative to matched calm evenings. These are the usual indicators that a system is approaching a tipping point.
Interpretation. The offer stack has a narrow band between $300 and $1,000 and large standing offers near the cap. A small shortfall of offered capacity can therefore move the clearing price from below $1,000 to the high bands within one dispatch interval. That matches the onset-interval results in Chapter 02 and the sparsely populated middle of the price distribution in Chapter 06.
The rise is not preceded by a gradual change in how prices move. Warning therefore has to come from forecasts of the supply–demand balance: demand, unit availability and interconnector limits.
Summary of measurements
7.5% of spike episodes reach $1,000; 48% of those reach $5,000. Episodes reaching $5,000 are 3.6% of episodes and carry 76.8% of dollars above $300.
35.7% of episodes reaching $5,000 are at that level in their first interval; 52.5% in FY26 (n = 40). 61.6% of those last only that one interval.
An interval at $5,000 or more is followed by another 70.8% of the time while the episode continues. Episodes reaching $5,000 last a median of 35 minutes.
The annual maximum equals the cap in force in 22 of 25 region-years. 0.014% of intervals are within 2% of the cap.