A deferred-tax valuation allowance booked then released breaks two years of EPS

Claim

Qualcomm booked a $5.7B non-cash tax charge ($5.29/share) in FQ4 FY2025 when the One Big Beautiful Bill Act forced a valuation allowance against future deferred tax assets — and then appears to have substantially released it in FQ2 FY2026.

Fiscal quarter	Pretax income	Tax provision	GAAP net income	GAAP dil. EPS
FQ4 FY25 (Sep-25)	$2,971M	+$6,088M	−$3,117M	−$2.89
FQ1 FY26 (Dec-25)	$3,547M	+$543M	$3,004M	$2.78
FQ2 FY26 (Mar-26)	$2,232M	−$5,138M	$7,370M	$6.88
FQ3 FY26 (Jun-26)	$2,462M	+$460M	$2,002M	$1.87

Corroborating balance-sheet movement: non-current deferred tax assets ran $743M (Sep-25) → $5,968M (Mar-26), and the FQ2 cash-flow statement shows a −$6,148M deferred-tax line. Effective tax rate by fiscal year: 2.19% (FY24) → 56.24% (FY25) (roic.ai).

(The release is inferred from the tax provision, the DTA balance and the deferred-tax cash-flow line moving together; the charge itself is confirmed in Qualcomm's FY2025 Q4 release. Confirm the exact release mechanism in the FQ2 FY26 10-Q before quoting it as fact.)

What breaks

FY2025 GAAP EPS $5.01 understates the year by ~$5.29. Normalized net income was ~$11.2B against $10.14B in FY2024 — underlying earnings grew ~11% in a year that screens as a 45% collapse.
FY2026 GAAP EPS will overstate the year by roughly the same amount.
Net-income CAGR across the pair is pure noise. Qualcomm's 3-year net-income CAGR reads −24.6% while its 3-year FCF CAGR reads +23.3% — same company, same window.
The TTM trap: a trailing window containing both the charge and the release nets them out. QCOM's TTM EPS of $8.75 and P/E of 18.6x look unremarkable and are right by coincidence, not by construction. Roll the window one quarter either way and it is off by ~$5.
ROE inherits it twice: through net income and through an equity base that absorbed the swing ($21.2B Sep-25 → $27.7B Jun-26).

How to apply

On any name with an effective tax rate outside ~10–30%, stop and find out why before using EPS, P/E, ROE, or a net-income growth rate. The tax rate is the cheapest tell there is and it sits in roic.ai get_profitability_ratios.
Use operating income for the trend and FCF for the level. Both are untouched here: Qualcomm's FY2025 operating income rose to $12.39B while net income fell 45%, and FCF was never affected — the charge was non-cash.
Check whether the change is permanent or reversing. A valuation allowance is an estimate about future taxable income, so it can and does come back. A one-way charge (a fine, an impairment) needs one adjustment; a round trip needs two, in opposite directions.
The forward consequence is usually the opposite sign to the headline. Qualcomm guided to a 13–14% effective tax rate and lower cash taxes afterward — the "disaster" year set up a tailwind.

Related: [[pitfall-unrealized-equity-marks-break-headline-pe]] · [[pitfall-government-warrant-derivative-inverts-gaap-eps]] · [[pattern-ai-build-inflates-earnings-while-destroying-fcf]] · [[principle-primary-source-beats-vendor]]

Second confirmed instance — Dynatrace (DT), 2026-08-06

The mechanism is not a semiconductor or a tax-law-change artifact. Dynatrace ran the same round trip in software, with no OBBBA involvement, and it broke the screen just as badly.

Fiscal year (Mar-end)	Pretax income	Tax provision	GAAP net income	GAAP dil. EPS
FY2024	$154.9M	+$0.28M (0.18% rate)	$154.6M	$0.52
FY2025	$223.4M	−$260.3M (BENEFIT)	$483.7M	$1.59
FY2026	$299.8M	+$137.1M (45.7% rate)	$162.7M	$0.54

What it broke, as of 2026-08-06: trailing P/E 89.5x · reported "earnings growth" −25% · ROE 20.9% → 6.2% · every net-income CAGR spanning FY25–FY26. Underlying pretax income grew 34% in FY26 while GAAP EPS fell 66%. Normalized at a 21% rate, EPS went $0.58 → $0.78 and the 3-year normalized-EPS CAGR is +47%.

Two refinements this instance adds to the original:

The near-zero rate is the leading indicator. DT's FY2024 effective rate of 0.18% is the tell that a full valuation allowance is in place, one year before the release lands. A rate below ~5% means the company is not paying tax because it is sheltered by an allowance against DTAs — the release is then a scheduled future event, not a surprise.
Confirm with the tripwire, which fires first and costs one subtraction. DT's FY25 net margin (28.5%) exceeded its operating margin (10.6%) → [[pattern-net-margin-above-operating-margin-is-a-tripwire]]. Run that check before ever opening the tax footnote.

Third and fourth instances — PEGA and MNDY, both found 2026-08-10, both in the same session

Two more in one /analyze run. This is not exotic; it is the base rate in 2026 software. Both companies established allowances during the 2022–2023 loss years and released them in Q4 2025, the same quarter — because that is when a cohort of software names crossed into sustained GAAP profitability at once.

PEGA — the cleanest specimen yet, because both ends of the round trip are visible

FY	Pretax income	Tax provision	Net income	Effective rate
2022	−$161.80M	+$183.79M (expense on a loss)	−$345.58M	n/m — allowance ESTABLISHED
2023	$95.44M	+$27.63M	$67.81M	29.0%
2024	$142.64M	+$43.45M	$99.19M	30.5%
2025	$280.63M	−$112.81M (BENEFIT)	$393.44M	−40.2%

Confirmed three independent ways: Q4'25 tax provision −$129.62M; cash-flow deferred-tax line −$171.05M; and non-current deferred tax assets jumping $3.97M (Sep-25) → $175.47M (Dec-25).

What breaks: trailing P/E 18.8x → ~40x normalized. FY2025 net income $393.4M → ~$211–222M. And the TTM trap fires in its pure form — the $129.6M benefit sits inside the current TTM window and rolls off after the Q3 2026 print, roughly doubling the reported P/E with no business change.

The 2022 charge is the part worth keeping. A tax expense of $183.8M booked on a pretax loss of $161.8M is the establishment event, and it is visible three years before the release. A company that took an inexplicably large tax expense in a loss year has a scheduled future earnings windfall sitting on its balance sheet. That is a screenable leading indicator, not just a trap.

MNDY — the extreme case: operations contributed nothing at all

($000)	FY2023	FY2024	FY2025
Operating income (loss)	(38,585)	(21,034)	(1,748)
Financial income, net	41,911	55,500	61,065
Income tax benefit (expense)	(5,203)	(2,094)	+59,425
Net income	(1,877)	32,372	118,742
Effective tax rate	156.0%	6.0%	(100.2)%

118,742 − (1,748) = 120,490 = 61,065 interest + 59,425 tax benefit — exact, no other items.

~51% of net income was the one-time non-cash release; ~49% was interest on the cash pile; operations lost $1.7M. Cash taxes actually paid: $4.9M. The 20-F states it plainly: "reversal of the valuation allowance… of $61.1 million… a non-cash income tax benefit."

FY2024's 6.0% effective rate is the near-zero leading indicator from the DT case, firing exactly one year ahead of the release — now 2-for-2 as a predictor.

A refinement MNDY adds: the largest DTA component was share-based compensation ($41.7M of $92.7M gross). The company booked a $61M profit by recognising tax assets created by the very SBC that suppressed its operating income. On heavy-SBC software names the release is mechanically larger, because SBC is what built the deferred tax asset in the first place. Expect this pattern to keep firing across the 2021 IPO cohort as it turns profitable.

History

2026-08-04 — Found during /analyze QCOM when FY2025 net income fell 45% on rising operating income. static — the mechanism is in the accounting standard, not the cycle.
2026-08-06 — Second instance confirmed at Dynatrace during /analyze DT, in a different sector with a different trigger. Confidence held at high; added the near-zero-rate leading indicator. Two instances in three days suggests this is common, not exotic — check the effective tax rate on every name before quoting a P/E.
2026-08-10 — Third and fourth instances, PEGA and MNDY, both found in a single /analyze run and both releasing in Q4 2025. PEGA supplies the first case where both ends of the round trip are visible (a $183.8M tax expense on a 2022 pretax loss establishing it), making "large tax expense in a loss year" a screenable leading indicator of a future windfall. MNDY supplies the extreme: operating income of −$1.7M with 100% of net income from the release plus interest, and establishes that heavy-SBC names get mechanically larger releases because SBC built the DTA. The near-zero-rate predictor is now 2-for-2. Four instances in six days — treat the effective-tax-rate check as mandatory on every software name.