GRIDINT Research · AI Economics

Fifteen events. Every number wears its class.

In fifteen months, fifteen companies — from Intel and Amazon to Nike and FedEx — filed or announced the workforce reductions on this ledger: 115,875 tracked job cuts, plus two filed restructurings that never disclosed a number. The deeper finding is about the numbers themselves: only about a third of those heads trace to a count a company actually put in a filing. This report grades every number by where it comes from — and every figure on this page is computed from the ledger at render time.

THE WAVE

Fifteen months of cuts, one picture.

Every event on the ledger, on the calendar it was announced. Bubble area is headcount; color is the receipt class — where the number actually comes from. The legend is also the control: switch classes off to see what survives on filings alone. Dashed hollow marks are companies that filed a restructuring and never disclosed a count — absence drawn as absence, never as zero.

A1The tracked ledger as a timeline — sized by headcount, colored and filterable by receipt class
WHAT YOU ARE LOOKING AT Fifteen layoff events, May 2025 → July 2026. Green counts appear verbatim in a filing; violet ride on filed charges with a press count; amber is press-only; pink is derived from a filed end-state and includes attrition. Click a legend chip to include or exclude its class; tap or hover any mark for its receipt.
sourceSEC EDGAR filings + receipted press, per event · classes printed in B1 and F1 methodcurated entry — a receipt or no entry · null count ≠ 0 · the filter changes what is DRAWN, never what is tracked
THE RECEIPTS

Where the numbers come from.

The same 115,875 heads, decomposed by the quality of their receipt. The filed-verbatim rows are the floor you can defend in any room — a number the company itself signed in a filing. Everything above that floor inherits the class of its weakest link.

B1Tracked heads by receipt class — the composition of the public number
WHAT YOU ARE LOOKING AT Each row is a receipt class: the bar is heads carried on that class, the line beneath names every company in it. The hollow dashed row holds the two filed restructurings with no public figure at all.
sourceclass per event, printed on the ledger, F1 methodfiled = count verbatim in a filing/regulatory release · split = charges filed, count press · press = count press-only · derived = computed from a filed end-state, incl. attrition · no figure = Item 2.05 filed, headcount undisclosed
THE PACE

The wave, accumulating.

Read cumulatively, the ledger shows two waves. 2025 was four events, three of them giants — Intel, Amazon and HP stacked 60,400 tracked heads inside four months; CrowdStrike’s 500 is the fourth. 2026 is many more companies cutting less each: eleven events across six industries, six of them in the first quarter alone. The wave did not grow taller — it grew wider.

C1Cumulative tracked heads by announcement month — with events per quarter
WHAT YOU ARE LOOKING AT The step line is the running total of tracked heads, dots colored by each event's class. The columns beneath count events per quarter — the breadth of the wave, independent of its size. The two no-figure events appear in the columns but can add nothing to the line: that gap is itself the finding.
sourcesame ledger · date_announced per event methodcumulative sum of counted heads in announcement order; no-figure events counted in the columns, excluded from the line — stated, not guessed
DEPTH VS SIZE

The giants trim. The mid-caps amputate.

A headcount says how big a cut is; the share of workforce says how deep it goes. WiseTech's 2,000 is a stated 30% of WiseTech — while some of the ledger's biggest headcounts state no share at all. Where companies stated the share, this exhibit draws both dimensions at once; where they didn't, the row says so instead of computing one from a denominator the company never gave.

D1Share of workforce cut, where stated — dot area is absolute headcount
WHAT YOU ARE LOOKING AT One row per event that stated a workforce share, sorted deepest first. The dot's position is the percent of the company cut; its area is the absolute headcount. Below the chart: the events whose share was never stated.

sourcesame ledger · pct_workforce as stated by the company or filing methodno share is ever computed from an assumed headcount — stated or absent
THE LAG

Announced is not effective.

Between the press release and the last day there is a window in which a headline is true but nothing has happened yet. For same-day filings that window is zero; for HP it is nearly three years. Reading layoff totals as if announcement equaled effect compresses years of the labor market into one news cycle.

E1Days from announcement to effect, where both dates are tracked
WHAT YOU ARE LOOKING AT Each bar is one event's announced→effective gap in days, longest first. Events whose effective date is not yet tracked are absent by honesty — the newest entries are still inside their windows.
sourcesame ledger · date_announced → date_effective
THE LEDGER

All fifteen, every field.

The record behind every exhibit above — one row per event, receipt class printed on each. The estimated-payroll column is the desk's estimate and is labeled as one. This table is deliberately the last view: a table is the record, not the story.

F1The complete tracked ledger — edition 2
sourceGRIDINT economics/layoffs.json · receipted per event · retrieved 2026-08-26 methodest. payroll saved is the desk's estimate, method on file — labeled, never presented as a filed figure
METHOD

What this is, and what it refuses to do.

The data: a curated ledger. Every event enters with a receipt — an SEC filing (8-K, 10-Q, 10-K), a regulatory release, or reputable press — and the receipt is graded by class, printed on every exhibit. We read the filings first, fetched from SEC EDGAR with a declared research agent; the exact filed sentence governs. The discipline: we never create a number. No share of workforce is computed from an assumed headcount; no undisclosed count is estimated; null is never rendered as zero; the largest figure in the ledger is labeled as the derivation it is (Intel's 26,400 — press-derived from a filed target of "about 75,000" core employees, attrition included). The limit that matters most: whether any of these cuts is attributable to AI is an editorial judgment, not a data operation — and it has not been made yet.

Attribution is pending — deliberately. Not one event on this ledger is yet marked "attributable to AI." That column is reserved for a per-event human editorial read, and it is printed as pending until the editor makes each call. A ledger that auto-attributed every tech layoff to AI would be committing the exact error this report exists to catch.
G1The review queue — what is not in the ledger, and why
WHAT YOU ARE LOOKING AT The pipeline that keeps the ledger honest: filings found and awaiting a read, proposals checked and rejected, and companies this route cannot reach. The next edition enters whatever survives the same read this one did.
sourcelayoffs-review-queue.json · method + caveat on file
What the next edition watches: UiPath's unread filing; whether C3.ai or FedEx put a number on their filed reductions (the class for it is already waiting); and the attribution column, which fills only by the editor's hand.
FINDINGS

Five things the receipts show.

1The wave is broader than tech's front row — and early 2026 was its busiest stretch. Six of the fifteen events landed in the first quarter of 2026 alone, across semiconductors, software, retail infrastructure, logistics, and apparel. → the pace, C1
2Only 35% of the tracked heads trace to a filed count. 40,100 of 115,875 carry a headcount stated verbatim in a filing. The public picture of this wave is mostly built from numbers no company has put its name to. → the receipts, B1
3The single largest figure in the ledger is not a layoff count. Intel's 26,400 — 23% of all tracked heads — is press-derived from a filed end-state and explicitly includes attrition. Any total quoting it as a cut count overstates what has been filed. → the wave, A1
4The giants trim; the mid-caps amputate. Where the share is stated, megacaps cut 2–10% — while WiseTech cut 30% of its entire company in one board action. Size and depth are different stories, drawn together. → depth vs size, D1
5Announced is not effective. Where both dates are tracked, the gap between headline and last day runs from zero days to nearly three years — HP's program stretches to fiscal 2028. → the lag, E1
A GRIDINT RESEARCH REPORT · AI ECONOMICS · AUTHORED FROM OFFICIAL RECORDS · RENDERED BY THE GRIDINT BUILDER
published 2026-08-27 · data cutoff 2026-08-27 · edition 2 — supersedes 2026-08-26 ed.1 — published before the edition archive existed, not preserved · updated only by a new dated edition · attribution: PENDING HUMAN READ on all 15 events, by design