The batteries that don't follow the grid

A diesel generator does not need permission to make electricity. Spin the shaft, and the machine produces a voltage whether or not anything else in the world is running. That is so obvious it is almost never stated, and because it is never stated, most people assume a battery works the same way.

It usually does not. The overwhelming majority of grid batteries installed today cannot make a voltage on their own. They measure the voltage the grid is already making, lock onto its phase, and push current in step with it. Take the reference away and they have nothing to be in step with.

That distinction has a name, and it is the single most consequential thing on a storage spec sheet that nobody puts on a storage spec sheet.

The inverter is a measurement device first

Almost every inverter-based resource on the grid today, meaning solar plants, wind plants and batteries, runs what the industry calls grid-following controls. NERC describes the architecture plainly:

"GFL typically includes a fast current control loop and a phase-locked loop (PLL). The control objective of the current loop is to achieve fast regulation of the IBR output current, and the control objective of the PLL is to synchronize the IBR output current with the grid voltage and to provide the phase information for the internal current control loops."

A phase-locked loop is where I want to stop, because I spent a career on the other side of one. In electronic warfare a PLL is the thing you attack. It is a circuit that watches an incoming waveform, estimates its phase, and steers a local oscillator to match. It is exquisite at tracking a clean, strong signal. It degrades in a very specific way when the signal it is tracking gets weak, noisy or fast-moving, because it can only ever report a phase it has already measured and processed. There is always a lag between what the world did and what the loop believes.

Put that circuit in charge of a 200 MW power plant and the consequence follows directly. NERC states it: "Due to this control structure, there is a delay of the PLL phase information and voltage information calculation. Therefore, GFL controls cannot instantaneously change IBR's active and reactive current output."

The modeling convention gives away the rest. Grid-following converters, NERC writes, "are typically represented as controlled current sources." A current source needs somewhere to push. It is defined by what it injects into a voltage that already exists, which means it is, in the most literal circuit-theory sense, a device that presumes a grid.

Everyone is calibrating, nobody is the standard

This is a measurement problem wearing a power-systems costume, and I recognize it from a lab.

Imagine a room full of instruments where every one of them calibrates against a neighbor. As long as one real primary standard sits in the corner, the whole room is traceable and the arrangement works beautifully. Retire the primary standard, and nothing announces the change. Every instrument still reports a number, every calibration still completes, and the readings still agree with each other. What has quietly disappeared is any reason to believe them.

The grid ran on spinning masses for a century, and a synchronous generator is a primary standard for phase. Its rotor angle is the reference; the waveform is a physical consequence of a heavy object turning. Everything else could afford to be a follower because something was leading.

As those machines retire, the count moves in both directions at once. Fewer things set the reference, more things depend on it. NERC's language for the resulting condition is "system strength," and the finding is blunt: "It has been shown that GFL controls become unstable under certain low system-strength conditions." Not less efficient. Unstable. And this is not a tuning failure, it is structural: "GFL controls fundamentally require some minimum system strength to maintain stable operation."

The sentence I keep coming back to is this one: "Every IBR control system topology and parameter value set has a finite region of stability in terms of system strength and system inertial strength." Every inverter on the grid has an edge. Nobody prints it on the box.

What grid-forming actually changes

A grid-forming inverter inverts the dependency. NERC's definition:

"Grid Forming Control for BPS-Connected Inverter-Based Resources are controls with the primary objective of maintaining an internal voltage phasor that is constant or nearly constant in the sub-transient to transient time frame. This allows the IBR to immediately respond to changes in the external system and maintain IBR control stability during challenging network conditions."

Read the priority ordering, because that is the whole design. In the first few cycles after a disturbance, roughly zero to five cycles, a grid-forming inverter holds its voltage magnitude and angle and lets power fall where it falls. A grid-following inverter does the opposite: it holds its current and lets the voltage be whatever the grid says it is. Same hardware, same cabinet, same cells. Different answer to the question of which quantity is sacred.

And critically, the grid-forming machine is not waiting on a measurement to act. NERC: "GFM most commonly uses an instantaneously measured voltage signal rather than a processed signal from a PLL in a GFL inverter. GFM response and support to the grid are instantaneous in the transient time frame."

What that buys, per NERC's own list, includes operation in low system strength, frequency and voltage stabilization, damping of power oscillations, resynchronization after separation, fault ride-through, and, for some units, blackstart. The 2021 paper notes field projects have already demonstrated "islanded operation capability without synchronous generation, blackstart capability, and operation in parallel with grid-following resources and synchronous machines."

That is the sentence to sit with if you care about resilience. A grid-forming battery can be the thing the others follow.

The number that should be embarrassing

So how much of the fleet needs this capability? NERC's 2023 paper on functional specifications for battery storage answers the question and then delivers the punchline in the same breath. Studies to date, it says, "indicate these numbers may be upwards of 30%." Then:

"Since the current percentage of GFM resources is near zero in nearly all large, interconnected power systems, it is recommended to start requiring and enabling GFM in all future Battery Energy Storage System (BESS) projects."

Near zero, against a possible requirement of thirty percent. And the pipeline was already enormous when they wrote it: 427 GW of battery storage in United States interconnection queues at the end of 2021, rising to 680 GW by the end of 2022. NERC's observation about that pipeline is the part that should sting.

"In the absence of any requirements or incentives for GFM capability, all of these resources are being planned with GFL controls."

All of them. Not because grid-forming is expensive or unavailable, but because nothing in the process asked.

Why this is a procurement problem, not a technology problem

Here is what makes this worth an owner's attention rather than a modeler's. The hardware delta is small. NERC: "New BESS can be equipped with GFM technology at a relatively low incremental controller and hardware cost," and for existing plants, "implementing GFM controls at existing grid following (GFL) BESS projects may only require controls changes."

A firmware-scale change, in many cases. But the paper immediately closes the trap door:

"However, these changes to an existing plant, as a material modification, will require additional studies to determine any impacts to BPS reliability. Due to the potential costs, time delays and complexities of this retroactive process, it is recommended that all new BESS projects are commissioned with the ability to perform GFM control."

The expensive part is not the capability. It is going back. Changing a commissioned plant's fundamental control mode is a material modification, which means new studies, new queue position risk, and new time. So a decision that costs almost nothing at the specification stage costs a great deal eighteen months later, and the industry is currently making it by default, in the direction of the cheaper-now option, at hundreds of gigawatts of scale.

NERC's fallback recommendation is the pragmatic one and worth quoting for anyone writing a contract: where entities are not ready to commit, they "should consider specifying IBRs that can be configured for both GFL controls and GFM controls." Buy the capability, enable it after it has been studied. That is the same logic as running conduit you do not need yet.

This is the layer of the work I sit closest to. The control mode, the state-of-charge policy and the dispatch logic are one system, not three, and the question of how much headroom a pack keeps in reserve is the same question whether you are asking it for revenue or for stability. I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I do not own that company and earn nothing from this link; I flag it because it is a field I build in and not only write about. Full policy here.

What grid-forming is not

I want to be careful here, because the enthusiastic version of this article would oversell a real technology.

Grid-forming is a control mode, not an energy source. NERC is explicit that in all cases "the inverter controls could be restricted by the inverter and primary energy source capability limits (e.g., available energy, current limits, voltages)." A battery holding up a voltage is still spending charge to do it. An empty pack in grid-forming mode is an empty pack. State of charge decides how long the heroics last, which is why the accreditation question I wrote about in capacity credit and the control question here are two halves of the same problem.

It is also not finished. NERC says field experience "is being accumulated," that grid-forming inverters "also require tuning," and that "careful testing and validation of GFM performance by responsible entities is still needed before broad deployment of this technology in their system." The open-research list includes something that should give anyone pause: control interactions between multiple grid-forming units. We understand one of these on a system reasonably well. We understand a hundred of them interacting rather less well, and pretending otherwise would be exactly the kind of overclaim this publication exists to flag.

One more scope note. Both documents I am citing are NERC white papers, which carry recommendations. They are not Reliability Standards, and nothing in them compels anybody to do anything. That gap between "the reliability organization recommends this" and "the rule requires this" is precisely the gap I flagged in the data-center disconnection story, where an alert that made headlines as a mandate said in its own text that it created no obligation at all. Read the instrument, not the coverage.

The signal

Every grid-following inverter is a device that presumes somebody else is generating the reference it needs. That was a completely sound assumption for a hundred years, because heavy spinning objects were doing it and nobody could imagine them leaving.

The interesting failure mode of a system built entirely from followers is not that it collapses dramatically. It is that it works perfectly right up until the last leader steps out, and there is no instrument on the system that reports how close you are, because "system strength" is a property of the whole network and not of any device in it. Each inverter has a finite region of stability that nobody publishes. The grid finds the edge experimentally.

So the useful question to ask about a storage project is not how many megawatts or how many hours. It is: can this thing hold a voltage up by itself, and if not, was that a decision somebody made, or just the default that nobody was asked about? Right now, at hundreds of gigawatts of scale, it is overwhelmingly the second one.