Reddit, Inc. [RDDT] — Company Research

Tier-2 memo · framework: Criteria, 2026-07-29 · as of 2026-07-29 · spot $178.345 (close 2026-07-28)

This document issues no position verdict. It scores Criteria and outputs an analysis. Whether that analysis justifies a position is a question about a particular book, and the book decides.

0. What this memo adds to the Tier-1 screen — and what it corrects

The screen did fundamentals, multiples, reverse DCF and momentum from XBRL. This memo does the layer XBRL cannot see, and audits the screen's inputs against the filings. Two of the screen's inputs are wrong and one is mislabelled. All three move the answer in the same direction — the screen was too harsh.

Dual-class check, run because it failed on another name tonight. Reddit has three authorised classes (A, B, C). C is authorised 100,000,000 and zero issued. Dimensional XBRL splits A and B, which SEC aggregation does not surface. Cross-check as instructed: net income ÷ shares ≈ filed EPS — $203,981k ÷ 191,518,973 basic = $1.065 against a filed $1.07; ÷ 202,524,173 diluted = $1.007 against a filed $1.01. Both tie. The share count is confirmed from two independent directions.

Recency. Latest filing is the Q1 2026 10-Q, filed 2026-05-01 for the period ended 2026-03-31 — 89 days old. Reddit reported Q2 in the last week of July in each of 2024 (Aug 6) and 2025 (Jul 31). Q2 2026 results are imminent or have just landed and are not in this analysis. That is the single largest information gap in this memo and it is stated, not buried.

1. The mechanism — named, specific, evidenced

Reddit's revenue growth is a price mechanism, not a user mechanism, and the price is being taken in a market where Reddit is structurally under-monetised relative to its attention share.

The Q1 2026 10-Q states the decomposition directly:

"During the three months ended March 31, 2026, ad impressions delivered increased by approximately 32% compared to the prior year period driven in part by ad load optimization and improvements, including new ad placements, and by increases in the number of users and user engagement. During the three months ended March 31, 2026, the price of ads increased by approximately 32% compared to the prior year period driven by higher advertiser demand and performance across the full funnel of ad objectives."

1.32 × 1.32 = 1.74, against advertising revenue growth of +74.2% in the quarter. The identity closes. Half of Reddit's growth is price and half is ad load. Neither is user growth.

The named driver behind the price half is full-funnel ad-objective expansion — Reddit moving from brand/awareness inventory (cheap, CPM-priced) into performance and conversion objectives (expensive, outcome-priced). That is a product decision with a measurable output: ARPU.

(Source: Q1 2026 Form 10-Q, "Quarterly ARPU". Q1'26 levels: global $5.23, US $9.63, RoW $2.02.)

ARPU growth inflected in 2025Q2 and has not decayed for four quarters. That is the mechanism, and it is the only line in Reddit's operating record that is both large and accelerating.

Why it is not exhausted. US ARPU of $9.63 per daily unique per quarter is roughly $38.50 annualised. Meta's global ARPU is a multiple of that and its US/Canada ARPU is a large multiple of that. Reddit's own RoW ARPU of $2.02 against a US $9.63 is a 4.8x internal gap on the same product — the company's stated strategy (machine translation, international ad sales build-out) is to close an ARPU gap it can observe inside its own P&L. That is a specific, evidenced runway, not a TAM slide.

Why it is fragile. Ad load has a ceiling and users have a churn response to it. Half of the impression growth is coming from placing more ads in front of the same people. That half cannot compound for five years.

2. The trap, resolved: recurring advertising versus lumpy AI licensing

The brief's hypothesis is that a 48.9% CAGR carried by a handful of AI data-licensing contracts is not a growth rate. The filings refute the hypothesis, decisively and with numbers.

2.1 The split, from the revenue disaggregation note

(Source: FY2025 Form 10-K Note 3 and Q1 2026 Form 10-Q Note 3. "Other revenue" is the line that contains data/content licensing; Reddit does not disclose licensing separately from Reddit Premium and other items, so 5.9% is an upper bound on licensing, not the licensing figure itself.)

2.2 Contribution to the demonstrated CAGR

Over the CAGR window (FY2022 $666.7m → FY2025 $2,202.5m), total revenue grew by $1,535.8m. Other revenue grew from a figure below its FY2023 level of $15.2m to $140.0m — at most $129m, or 8.4%, of the increment. Advertising supplied at least 91.6%.

Narrow it to the most recent year, where the licensing thesis should be strongest: FY2024 → FY2025 total revenue +$902.3m, of which advertising +$877.0m (97.2%) and other +$25.3m (2.8%).

Advertising alone compounded at approximately 46.6% over the three-year window against a total of 48.9%. Strip every dollar of licensing out of both ends and the demonstrated growth rate falls by roughly 2.3pp. The CAGR is an advertising CAGR.

2.3 The licensing stream is small, contracted, and stepping down

The remaining-performance-obligation disclosure is the only forward view Reddit gives on licensing, and it is explicit:

"As of March 31, 2026, the aggregate amount of remaining performance obligations in contracts with an original expected duration exceeding one year was $120.6 million. This amount consists primarily of long-term content licensing contracts… We expect to recognize $91.6 million in the remainder of 2026 and $29.0 million in 2027." — Q1 2026 10-Q Note 3

At FY2025 year-end the same disclosure read $143.7m, split $118.9m in 2026 and $24.8m in 2027.

Two readings, both worth stating: - The licensing book is $120.6m against $2,473.6m of TTM revenue — 4.9%. It cannot carry a growth rate and it never did. - It steps down 68% from 2026 to 2027 on contracted terms. If nothing is renewed, roughly $60–90m of annualised revenue rolls off in 2027 — a 2–3% headwind to total revenue, not a thesis break. Renewals are unknowable from the filings and are not assumed here in either direction.

2.4 Management has stopped talking about it — mention frequency, §4

"Data licensing" appears 11.6 times per 10,000 words in the Q1 2024 shareholder letter and zero times in every letter from Q4 2024 onward. A stream that was a headline at IPO is now not mentioned at all. That is consistent with the numbers above — it is small — but it is also a disclosure choice that removes the investor's ability to track it.

2.5 Customer concentration — settled from the filing

"No customer accounted for greater than 10% of our revenues for the years ended December 31, 2025, 2024, and 2023. No customer accounted for greater than 10% of our accounts receivable as of December 31, 2025 and 2024." — FY2025 10-K

There is no Google-sized or OpenAI-sized single counterparty inside the revenue line. The concentration risk the brief was probing does not exist at the 10% threshold in any of the last three years.

3. The trap, part two: what happens to traffic when search referral shifts

This is where the real risk sits, and the filings quantify it precisely.

3.1 The user data, exactly as filed

(Source: Q1 2026 Form 10-Q, "Quarterly Average DAUq". Q1'26 levels: global DAUq 126.8m, WAUq ~490m, DAUq/WAUq 26%.)

The United States — 81% of FY2025 revenue and the geography with a $9.63 ARPU — has stopped adding logged-in daily users. US logged-in DAUq growth is +1%. It was +32% eight quarters ago. This is the single most important operating fact about Reddit and it is not in the screen.

3.2 The company names the cause itself

"The growth in global DAUq in the three months ended March 31, 2026, compared to the prior year and prior quarter periods, was primarily driven by the combination of third-party search engine algorithm changes and continued traction from our growth initiatives, particularly in machine translation, marketing, and product improvements." — Q1 2026 10-Q

Reddit attributes its own user trajectory, in the first instance, to someone else's algorithm. The FY2025 10-K carries the matching risk factor: "potential harm caused by changes in internet search engines' methodologies, particularly search engine optimization methodologies and policies."

The exposure is structural, not cyclical. Reddit's logged-out DAUq — the half of the user base that arrives from a search result, does not have an account, and is worth materially less per head ("logged-in users generally contribute significantly more to ARPU than logged-out users due to the lower monetization opportunity of logged-out users", 10-Q) — is the swing factor in the headline number. Total DAUq grows +17% while logged-in grows +7%: the delta is logged-out traffic, and logged-out traffic is a function of how often a search engine or an AI answer surface decides to send a click rather than answer inline.

3.3 The disclosure is being withdrawn — flag this

"…beginning in the quarter ended September 30, 2026, we will no longer report logged-in and logged-out DAUq." — Q1 2026 10-Q

Reddit is retiring the logged-in/logged-out split one quarter after US logged-in DAUq growth printed +1%. The stated reason ("our focus has shifted to delivering immediate value for all users") is plausible on its face. The effect is not in dispute: from Q3 2026 there will be no filed series that separates durable account-holding users from search-referred drive-by traffic. The metric that would tell an owner whether the franchise is compounding or being rented from Google is being switched off.

This is the accounting-quality finding on RDDT. It is not fraud and it is not aggressive revenue recognition — Reddit's revenue recognition is clean, its receivables are unremarkable, and its cash conversion is better than its earnings. The quality issue is a metric withdrawal at the precise moment the metric turned.

3.4 What breaks if referral shifts further

Reddit's revenue is DAUq × ARPU. If logged-out DAUq falls, the ARPU denominator shrinks and reported ARPU rises mechanically — because the users being lost are the low-monetising ones. A referral shock would show up in Reddit's headline ARPU as an improvement. Anyone underwriting this name on ARPU alone will read a traffic loss as a monetisation win. The honest read requires the two series side by side, which is exactly the pair being retired in Q3 2026.

4. Transcript / prepared-remarks mention frequency — required core metric

Source substitution, stated. The Alpha Vantage EARNINGS_CALL_TRANSCRIPT endpoint was rate-limited to zero on 2026-07-29 (the shared free-tier cap of 25 requests/day was already exhausted by parallel runs; the API returned an Information rate-limit body, which the fetcher correctly refused to cache). No earnings- call transcript could be obtained. In its place, mention frequency is computed across the quarterly Letter to Shareholders furnished as Exhibit 99.2 to each Form 8-K Item 2.02 — nine consecutive quarters of management's own prepared written remarks, retrieved from EDGAR. This is a primary source and it is the closest available analogue to prepared remarks; it is not the same instrument as a call transcript and is labelled as such wherever cited.

Counts are normalised per 10,000 words because letter length varies (5,716–6,875 words). Read the normalised row, never the raw one.

Per 10,000 words, Reddit Letter to Shareholders:

Four readings, each load-bearing:

1. "Data licensing" went from the third-most-emphasised topic to absent, in one quarter, and stayed absent for six quarters. Combined with §2, this is a stream management no longer sells to investors because it is no longer material. Anyone underwriting RDDT as an AI-data-licensing story is underwriting a narrative the company itself abandoned in Q4 2024.
2. "AI" rose 70% (64.7 → 110.2) over the same period. Reddit reframed from selling data to AI companies to using AI in the product. The frame moved; the revenue line did not follow.
3. "Answers" — the named AI search product — spiked to 27.6 in 2025Q2 and collapsed to 3.5. That is a product-cycle signature: heavy promotion, then de-emphasis. Either it worked and became plumbing, or it did not work. The letters do not say, and no disclosed metric tracks it. This is an unresolved gap.
4. "advertis*" fell from 31.5 to 17.5 while advertising rose from 98% to 94% of revenue. Management talks less about the thing that is 94% of the business and more about the thing that is 6% of it. Weigh the filings over the letters.

5. Product-cycle intelligence — named products, adoption evidence

No product event inside 12 months is disclosed with a date. Reddit does not run a launch calendar in its filings. The catalyst calendar therefore contains dated reporting events and dated disclosure changes, and no invented product dates. See RDDT_Catalyst_Calendar.md.

6. Accounting quality

Reddit's accrual quality is good, with one genuine distortion and one presentational issue.

Clean: - Cash conversion exceeds earnings. FY2025 CFO $690.9m vs net income $529.7m (1.30x). Q1'26 CFO $312.3m vs net income $204.0m (1.53x). Sloan accruals are negative — the direction associated with higher future returns, not lower. - No customer >10% of revenue or receivables in FY2023–FY2025 (10-K). - Deferred revenue is trivial and shrinking in relative terms ($28.3m at Q1'26 on $663.4m of quarterly revenue). No channel-stuffing surface: Reddit sells auction advertising, not units into a distributor. The Applied Optoelectronics failure mode (a private stocking distributor at 53% of revenue, ~264-day DSO, named in no release) has no analogue here — DSO is ~71 days on Q1'26 revenue and receivables fell from $590.2m to $522.9m over the quarter while revenue rose. - Revenue recognition is point-in-time auction advertising. There is no percentage-of-completion, no bill-and-hold, no vendor-rebate offset in cost of revenue.

The genuine distortion — the +63.2pp operating-margin expansion is mostly an IPO artefact:

FY2024 carried a $577.5m single-quarter SBC charge in Q1 2024 — the IPO RSU catch-up. The screen's op_margin_delta_pp = 63.2 is arithmetically correct and economically misleading: roughly three-quarters of the headline margin inflection is the absence of a one-time charge, not operating leverage. The real underlying expansion is +17.1pp, which is still excellent and still the strongest evidence for the mechanism — but it is not +63pp, and a screen that ranks on +63pp is ranking partly on an accounting event.

The presentational issue — earnings are untaxed and will not stay that way: - FY2025 effective tax rate: 0.1% on $530.7m of pre-tax income. Q1'26: 0.8%. - Valuation allowance at 2025-12-31: $779.0m, against gross deferred tax assets of $794.3m — still essentially fully reserved. - Federal NOL carryforwards $1.7bn, state $806.1m; R&D credit DTA $199.0m. - The FY2025 rate reconciliation shows a +$172.4m change in valuation allowance (+32.6pp) offsetting a −$286.0m SBC deduction (−54.1pp).

Two consequences an owner must hold simultaneously: (a) cash taxes stay near zero for years — a real, durable FCF benefit; (b) if the valuation allowance is released, GAAP net income takes a large one-time non-cash gain and the reported EPS for that quarter is not a run-rate. A company with $442m of operating income, positive cumulative three-year earnings and $1.7bn of NOLs is squarely in release territory. Treat any large EPS beat in the next four quarters as a candidate VA release until the tax note is read.

7. Archetype and Quality Criteria

Archetype: COMPOUNDER. Reddit is profitable (TTM operating margin 25.1%, TTM net income ~$707m), growing fast, with stable and rising gross margin. The screen's classification is correct. It is at the INFLECTION/COMPOUNDER boundary — it crossed into profitability only in Q3 2024 — so both standards are reported.

Quality Criteria (BINDING): PASS. The one blemish is deceleration, which the framework scores under growth acceleration and which is real: +69% → +44% guided is a 25pp deceleration in one quarter.

8. Downside case with a named cause (Downside Criteria, MEASURED — blocks nothing)

The permanent-impairment scenario is not a volatility figure. It is this:

Named cause: a further step-down in search-engine and AI-answer referral traffic to Reddit, in the United States, that removes logged-out DAUq faster than ARPU can compensate — while the disclosure that would let an owner see it is withdrawn in Q3 2026.

The mechanism is already visible in the filed series, not hypothesised: US logged-in DAUq growth +1%, US total DAUq growth +7%, and the company itself citing "third-party search engine algorithm changes" as the first-named driver of its user trajectory. Reddit does not control the input.

How it becomes permanent rather than cyclical. Advertising revenue = impressions × price. Impressions = users × sessions × ad load. If US users stop growing, impression growth must come entirely from ad load, which has a hard ceiling and a user-experience cost. Once load is spent, revenue growth collapses to price growth alone — and ad price growth is itself a function of advertiser competition for a growing audience. The two legs fail together, not independently. A platform whose audience is rented from a competitor's algorithm (Google, whose own AI Overviews compete for the same query) has no structural defence.

Quantification. At a 5-year horizon, if US DAUq is flat and RoW DAUq compounds at 20% while ARPU growth decays to +10%/yr, revenue reaches roughly $4.6–5.0bn by 2031 against the $8.8bn the current price requires (§ RDDT_Valuation.md). On the growth-matched exit multiple that is an equity value near $95–115/share.

• Bear-case price: $100 (−44% from $178.345).
• Probability: 30%.
• This is not a going-concern case. Reddit has $2.77bn of cash and securities, no debt, positive and rising free cash flow, and 91% gross margins. It cannot be forced. The impairment is a permanent de-rating of a franchise that turns out to be a distribution tenant rather than a destination — not insolvency.

Logged per the framework to the ledger with its cause. Downside Criteria is MEASURED and blocks nothing.

9. What could not be established — stated, not filled in

1. Q2 2026 results. Not filed as of 2026-07-29. Prior-year analogue landed 2025-07-31. Every figure here is as of the Q1'26 10-Q.
2. Street consensus for FY2026/FY2027 revenue and EPS. Alpha Vantage EARNINGS_ESTIMATES shares the same exhausted 25/day cap as the transcript endpoint. No Street number is quoted anywhere in this memo. The NTM revenue used in the 12-month target is a house build off the company's own guided Q2'26 and is labelled as such. Consensus Criteria is therefore INDETERMINATE — a quota gap leaves it blank and blocks nothing (criteria.md).
3. Earnings-call transcripts. Same cap. Mention frequency is computed on shareholder letters instead and is labelled as such throughout (§4).
4. Licensing revenue as a discrete line. Reddit discloses "Other revenue", not licensing. 5.9% of TTM revenue is an upper bound, not a measurement.
5. Reddit Answers adoption. No disclosed metric of any kind.
6. Logged-out DAUq levels in absolute millions. The 10-Q publishes growth rates and a chart; the absolute logged-out level is not in the machine-readable text. Growth rates are used throughout and are exact.

Sources

All figures traced to filings retrieved from SEC EDGAR on 2026-07-29: - Form 10-Q, quarter ended 2026-03-31, filed 2026-05-01, accession 0001713445-26-000069. - Form 10-K, year ended 2025-12-31, filed 2026-02-06, accession 0001713445-26-000022. - Form 10-K, year ended 2024-12-31, filed 2025-02-13, accession 0001713445-25-000018 (FY2022 revenue). - Form 8-K Item 2.02, filed 2026-04-30, accession 0001713445-26-000067, Exhibits 99.1 and 99.2 (Q2 2026 guidance; Q1 2026 Letter to Shareholders). - Quarterly Letters to Shareholders, Exhibit 99.2 to each 8-K Item 2.02, 2024Q1 through 2026Q1 (mention frequency, §4). - data.sec.gov XBRL companyfacts, CIK 0001713445 (quarterly series cross-checks). - Alpaca Markets daily bars and options snapshots, 2026-07-28 close (price, volatility, chain). - Tier-1 screen: reports/scan_all_v2/RDDT_analysis.json, as of 2026-07-28.