The Missing Denominator in US Battery Cancellation Data

Rhodium Group's Q1 2026 clean-manufacturing update reported $1.6 billion in cancelled investment, down from $8.4 billion in Q4 2025. E2's Q1 2026 analysis counted 38 EV and battery projects cancelled, closed, or downsized since early 2025, with the cumulative total still climbing. The quarterly figure falls while the cumulative count rises, and analysts cite the two against each other as competing readings of the same pipeline.

They are not competing readings. The tracked universes differ, the reporting periods differ, and "cancelled" does not mean the same thing in both. Open the methodology and the disagreement resolves. What remains is a harder problem: no publicly available source, including Rhodium, E2, and our own 42-facility registry, provides the structure needed to answer the question everyone is actually asking, which is whether a US battery manufacturing project announced today is more or less likely to reach production than one announced two years ago.

What each source counts

The differences are specific enough to tabulate.

Two rows account for most of the apparent disagreement.

The headline figure type is the obvious one. Rhodium reports a quarterly flow: dollars newly classified as cancelled during a given quarter. E2 reports a cumulative stock: total projects ever cancelled, closed, or downsized since tracking began. A quarterly flow can decline while a cumulative stock rises, because every quarter with any reversals at all adds to E2's running total regardless of what the quarterly pace is doing.

The second row is less visible and more consequential. E2's reversal universe is asymmetric with its announcement universe. E2's year-end methodology states that cancellations occurring after August 2022 can include projects proposed or operating before that date, projects E2 never counted as announcements because they predate the tracking window. A facility announced in 2019 and cancelled in 2025 enters E2's cancellation numerator without its original announcement entering E2's denominator. Any ratio of E2 cancellations to E2 announcements therefore overstates the cancellation rate for the post-IRA cohort, because the numerator draws from a wider population than the denominator. Cancellation counts from advocacy trackers carry this risk structurally: the rule for what qualifies as a reversal can operate over a wider population than the rule for what qualifies as an announcement.

Both Rhodium and E2 are aggregated analytical sources. They classify facility-level events into tracker categories using rules that are partially published and partially internal, and those classifications may not match what company disclosures, permit filings, and government records show at the facility level. The trackers report a pattern; the pattern's resolution depends on classification choices the reader cannot fully audit.

Pipeline depletion and the shrinking risk pool

Set the definitional questions aside and take Rhodium's quarterly series at face value.

Q1 2026 is the lowest in the series, and it has been read as evidence that the cancellation wave has crested. The series cannot support that inference without a denominator, and the denominator column is almost entirely empty.

The Q4 2025 spike shows the mechanism. That $8.4 billion consisted of five projects, all EV supply chain, all pre-operation. Once those exit the at-risk pool, the following quarter's cancellations are drawn from what remains, which is a smaller pool that has already shed its highest-dollar members. The dollar figure falls mechanically, not because the surviving projects are healthier.

Rhodium's own May 2026 analysis quantified the cumulative effect. Remaining outstanding clean-manufacturing investment, meaning projects that had neither broken ground nor completed construction, stood at roughly $98 billion against $279 billion in cumulative announcements. Something on the order of $181 billion has already left the pipeline through completion, cancellation, or reclassification. The at-risk pool is under 36% of the announced total. Announced capacity systematically overstates what will reach production, and the gap widens with every quarter of pipeline activity, since completions and cancellations alike deplete the pool the headline announcement figure implies still exists.

A declining quarterly cancellation figure is consistent with improving project survivability. It is equally consistent with a constant or worsening cancellation probability applied to a shrinking pool. The quarterly dollar series cannot distinguish between the two. What would distinguish them is a hazard rate: the probability that a project still in the at-risk pool transitions to cancellation during a defined period. Neither tracker publishes one.

What a hazard rate would require

Computing a cancellation hazard for the battery manufacturing pipeline requires four things that no public source provides.

First, defined cohorts. All battery-cell programs meeting an announcement threshold in a given quarter or year, fixed at entry and observed forward. A rolling universe, where projects enter and exit the tracked population across reporting periods, cannot produce a rate at all.

Second, stage classification at each observation point: announced, site selected, construction started, equipment installed, commissioning, customer qualification, commercial operation. Cancellation probability is not constant across those stages. A program that has poured foundations and installed dry-room equipment carries different risk than one that has named a site and done nothing else. Collapsing all stages into a single pool destroys the most operationally useful signal, which is at which stage projects fail and whether that stage-specific rate is moving.

Third, a surviving risk set at each interval — the projects, dollars, or GWh still eligible to experience a transition. This is the denominator. Without it you have a numerator (cancellations this quarter) divided by nothing in particular.

Fourth, defined transition outcomes. Cancellation, closure, downsize, product conversion, completion, and continued construction are distinct events with distinct implications. As we argued in Issue #8, a plant retooling from EV cells to grid storage cells removes EV-addressable capacity without destroying the facility, the equipment, or the workforce. Whether that registers as a cancellation depends on the tracker's classification rules rather than on the industrial event underneath them.

Both count-weighted and capital-weighted results would be needed. A single $4 billion cancellation dominates a dollar series while saying nothing about how often sponsors are failing across the pipeline. The two weightings answer different questions.

The distance between what current trackers report and what a hazard rate requires looks like this:

What trackers report:
  Quarter → Newly classified reversal dollars

What a hazard rate requires:
  Announcement vintage
    × Stage at quarter-start
    × Surviving projects / dollars / GWh at risk
    → Cancellation | Downsize | Conversion | Completion | Still at risk
    → Rate per interval, by cohort and stage

The trackers supply the top line. The cohort, stage, exposure, and transition architecture underneath it does not exist in any published source.

The registry as cross-section, not time series

Our 42-facility registry is a single-cutoff cross-section of non-producing US battery cell and materials programs as of August 8, 2026. Each record is classified by the nearest public evidence gate it has cleared: 20 at financing or construction, 7 installing or commissioning equipment, 9 at qualification or pre-commercial output, and the remainder elsewhere. It does not track the same universe as Rhodium or E2, and its status rules are deliberately different. Cancellation, downsizing, idling, and conversion stay in separate categories, and a product conversion does not register as a cancellation unless both the original and successor programs stop.

The registry also differs in source discipline. It records facility-level status from company disclosures, permit records, and government filings rather than assigning projects to categories through internal classification rules. That does not make it more comprehensive. It is narrower in scope, covering only battery-related programs, and it says nothing about solar, wind, or hydrogen. What it means is that each status assignment can be audited against a named source document.

The registry is not a time series. Its stage distribution is bottom-heavy, concentrated at financing and construction, and that pattern supports several readings. Most of these programs may simply be too early to have failed yet. Stronger sponsors may be advancing while weaker ones stall in place, producing selective survival. Or the concentration may reflect nothing more than the fact that qualification takes years and the pipeline is young. A directional claim would require applying the same denominator and classification rules to an earlier registry date, and we have not done that.

What neither series settles

Two credible sources report cancellation figures that are not contradictory and are also not comparable. The tracked universes differ by five years of announcement history and roughly $150 billion of cumulative investment. The cancellation definitions differ, the dollar bases differ, and one reports a flow while the other reports a stock. Neither establishes whether a project entering the pipeline today faces better or worse odds than one that entered in 2023, because the analytical structure that would support that inference — vintage-specific, stage-specific, exposure-adjusted — is not published anywhere.

Anyone citing the Q1 2026 decline as evidence that the wave has crested should state what they are assuming about the denominator. And anyone citing E2's cumulative count as evidence of acceleration should say which announcement population they are measuring against, and whether it includes the pre-IRA projects that E2's reversal universe captures but its announcement universe does not.

The pipeline is losing projects. Whether it is losing them faster or slower than two years ago cannot be determined from what is currently published, and will not be until someone tracks cohorts forward instead of counting reversals as they arrive.