Traditional power plants — gas, coal, nuclear — have turbines that spin at exactly 3 000 rpm. That rotating mass is free frequency stabilisation: when demand surges, the turbines slow just enough to buy operators a few seconds to react. The stored kinetic energy is called inertia and is measured in GVA·s.

When wind and solar replace spinning steel, inertia drops. The chart below shows the monthly average since April 2017. The solid line is total outturn inertia; the dashed line is the new contract market — effectively zero before 2020.

From the 251 GVA·s of 2017 we're down to around 151 today — a 40% drop.

Before 2020 the phrase "inertia market" didn't mean anything — the grid had so much inertia-rich gas and coal generation that there was nothing to buy. Starting early 2020, NESO began paying storage and flywheel operators to provide synthetic inertia.

The bars show what NESO actually spent each month on increasing system inertia actions. Flat for three years, then a hockey stick.

Each dot is one day. The x-axis is the average outturn inertia on the network; the y-axis is the going rate £/GVA·s·day NESO paid to shore it up.

The shape is the point. Above ~160 GVA·s the market barely exists. Below that line the rate climbs. Below ~140 GVA·s it explodes.

The peak rate in the dataset was £58,570 on 2025-07-25 — when average inertia fell to 173 GVA·s and NESO paid out £16.2 k that day alone for inertia-related actions.

Every day NESO pays for four kinds of "constraint" actions. Each category has a different physical reason: wires too hot (thermal), voltage drifted (voltage), inertia run low (inertia), or the biggest plant held in reserve (reducing largest loss).

Over the whole series, thermal costs dominate — 80% of every pound. Mostly Scottish wind output meeting an English transmission bottleneck.

The single most expensive balancing day on record is 2022-07-20 — £62.07 M in 24 hours. Inertia's share that day was 0%.

The previous charts average across 48 half-hourly settlement periods per day. What really matters is the minimum: the lowest point the grid touched in any 30-minute window. If that dips below 130 GVA·s, a well-timed generator trip could cascade into a frequency event.

The deepest low ever recorded was 2023-09-30, when the network briefly slipped to 44 GVA·s. NESO paid £90.0 k that day for inertia actions.

Across the whole series there are 609 days below the 130 GVA·s mark. Every one of them is a night someone had to think about.

Two lines against time. The green one — outturn inertia, sloping down. The red one — NESO's 30-day rolling spend on inertia-increasing actions, sloping up.

The vertical marks are the levers that were pulled along the way: the first Pathfinder asset (a spinning-mass flywheel) synchronising in June 2021; all twelve Phase-1 units online by April 2023; operability floor reductions in Feb and Jun 2024; and the final coal plant at Ratcliffe-on-Soar closing September 2024.

Each event is a node in the causal chain. Remove synchronous generation, pay someone else to stand in for it.

Top panel: outturn inertia against NESO's stepped operational floor. Bottom panel: nine years of daily generation, split into synchronous (gas · coal · nuclear · biomass · hydro — everything with a turbine shaft) vs inverter-based (wind · solar · storage · imports).

The stacked area ratio is the physics. When the grey slice grows, the green line drops — inverter-dominated hours are by definition inertia-poor hours.

Brush the bottom bar to zoom any time window.

One bar per UK fiscal year (April to March), stacked by constraint category. This is the Balancing Mechanism slice only — the amounts NESO pays day-to-day to reshape the physical dispatch.

Thermal dominates — mostly Scottish wind running into English transmission bottlenecks. Inertia is the smaller slice you've been reading about, but it has the steepest growth rate.

Footnote. Excludes Stability Pathfinder availability payments (Phase 1 ~£328 M total contract value; Phases 2–3 totals not separately published). Pathfinder spend flows through Network Services / BSUoS, not BM. Adding Pathfinder would roughly double the line from 2023 onwards — but the per-year split isn't publicly disclosed, so we don't fabricate it.

Five years of 1-second frequency measurements — 160 million readings. Aggregate them by month: the y-axis is the standard deviation of (frequency − 50 Hz) in millihertz. The x-axis is the same month's average outturn inertia.

If the physics is right, low-inertia months should have wider frequency excursions. The scatter should slant from bottom-right (high inertia, tight frequency) to top-left (low inertia, noisy).

The colour marks the year — watch the cloud migrate leftward and upward as the grid decarbonises. The physical relationship is visible and consistent.