3Q2026: Spend Per Payer Rose as the Payer Base Shrank

Thesis status: Weakening. The quarter beat every guided bookings, profit and cash-flow range and the over-18 mix moved for the first time, while the fourth-quarter guide leaves C2's failure trigger one print from firing and places C4's 26% line inside the guided range.

This is Memo Update No. 2 to the platform transition primer, after Memo No. 1 on 2Q2026. It covers the third-quarter results and the fourth-quarter guidance released on 29 October 2026.

Key Takeaways

1. Roblox ($RBLX) reported third-quarter bookings of $1.70B, down 12% year on year and $44M above the top of the guided range, with revenue up 16% to $1.57B as 2025's deferred bookings continued to release. Management attributed the beat largely to retention and engagement, stating that seasonally adjusted retention reached its highest level since 2021. Adjusted EBITDA of $138M beat the guide midpoint by almost exactly the amount revenue did, so costs landed close to plan. Free cash flow of $364M was flattered by about $140M of capital expenditure that moved into the fourth quarter.
2. Average monthly unique payers fell 16% to 30.3M, the first year-on-year decline in a data-sheet series that begins in 1Q2024, while spend per payer rose 5% to $18.69 after four consecutive declines. The rise is consistent with management's reading that active payers spent more. It fits equally the departure of low-spending payers recruited in the 2025 surge, and spend per payer remains 5% below 3Q2024 with 59% more payers. The provisional monetisation overlay opened in Memo No. 1 measured the variable that moved the quarter for a second time, and no numbered condition did.
3. Management guided fourth-quarter bookings down 10% to 15%, which puts FY2026 bookings 1% to 3% above FY2025, with adjusted EBITDA of -$70M to -$10M and free cash flow of -$190M to -$105M on $380M of capital expenditure. The guide assumes a seasonal fall in DAUs, no near-term improvement in monetisation and a modest, unsized headwind from the release of GTA VI.
4. The over-18 share of age-checked DAUs rose to 29% from 27%, its first move since the primer, and management now states the premium as 50% above under-18 users, the condition's floor, which meets C1's first strengthening trigger at its minimum while the cohort's growth slowed, so C1 holds at Developing with its trend flat. C2's failure trigger, DAU growth below 15% for three consecutive quarters, stands at two with DAUs down 9%, and the guide's seasonal decline would complete it at 4Q2026, the condition's own first-testable date. The 2025 comparison now sets that test, while the friction it was written to measure eased: more than 80% of DAUs aged 13 to 17 have age-checked, and DAUs that sent a chat message or connected to voice grew 24% on the quarter.
5. Infrastructure and trust and safety expense reached 24.5% of revenue, a second consecutive quarter above C4's 24% ceiling. Across the fourth-quarter revenue guide the 26% failure line falls between $382M and $411M, against $385M in the third quarter and an increase management has guided. Repurchases slowed to $200M from $380M, below C6's $250M quarterly marker, and the shareholder letter dropped the $1B twelve-month intent that the 10-Q retains. C6's free-cash-flow qualifier now binds on a fourth-quarter guide wholly below zero, and the condition moves to Developing.
6. Roblox accrued $49M for settlements and settlement talks with US states and territories in the quarter and $140M across nine months, excluded from adjusted EBITDA and equal to 36% of it over the same period. The FTC, Pennsylvania and Utah are considering claims. The memo opens a second standing overlay for this legal cost, with no status or thresholds.
7. At the 3 November close of $45.45, the price requires a flat free-cash-flow margin of 14.5% from FY2027 to FY2035, against 20.7% at the 30 July close that Memo No. 1 published. The model's 4.28% equity risk premium, against Memo No. 1's 5.50%, accounts for 4.0 points of the 6.2-point fall, the model's mechanics and the lower share price for 1.2 each, and the higher 10-year yield and other inputs return 0.2. Across betas of 1.00 to 1.80 the requirement spans 11.7% to 19.2%, against a reported margin of 27.7% in FY2025, a year whose deferral growth exceeded its free cash flow.

Third-Quarter Results

Management credited new titles that built audiences quickly under the refined recommendation algorithm for the bookings beat (Exhibit 2). Revenue beat the guide midpoint by $120.5M and adjusted EBITDA by $117.5M, which leaves costs outside the excluded items within a few million dollars of plan in aggregate. The margin beat is a revenue beat. Revenue outran bookings against the guide because less of the quarter's sales went into the deferred balance, net of what it released.

Capital expenditure moved into the fourth quarter accounts for about $140M of the free-cash-flow beat and operating cash flow for the rest, so free cash flow excluding the shift was about $224M (Exhibit 8). Against the prior year the cash picture is weaker, because net deferrals added $136M to cash against $458M a year earlier, so the float that bridges the GAAP loss to positive cash flow grew at under a third of its pace in 3Q2025.

What moved bookings

The 10-Q attributes the bookings decline primarily to average daily unique paying users, which fell to about 1.8M from about 2.3M, against a 3Q2025 payer base that had grown 88% (Exhibits 3 and 4).

The 5% rise in spend per payer is consistent with management's statement that spending improved among active payers. It is equally consistent with a change in who pays. Spend per payer reached its low of $17.88 in 3Q2025, the quarter payers grew 88%, and an average rises when its low spenders leave. Roblox Plus, launched in April, adds subscription bookings to the same average, and without a subscriber count its contribution cannot be separated. Daily paying users fell faster than monthly payers on the 10-Q's rounded figures, so the average payer paid on fewer days and spent more on each.

Management frames the year-on-year declines as a comparison with the "extraordinary growth" of 2025 and points to two-year growth instead, on which DAUs, hours and payers all rose faster than bookings (Exhibit 5). Bookings per hour engaged rose 3% on rounded hours, about 4% as management states it, and remain 8% below 3Q2024, so the recovery from the surge's dilution is partial.

US and Canada, where management stated that monetisation among younger cohorts remained challenged against the prior year, generated 53% of bookings and grew 29% over two years against 85% for the other three regions combined (Exhibit 6). The over-18 premium is disclosed for US users only, inside a region growing at about a third of the others' two-year rate.

Where the costs went

The revenue line still releases bookings taken across 2025, so a cost ratio on revenue improves as bookings fall, which is the denominator problem Memo No. 1 proposed fixing (P1). On bookings every main cost line rose, led by infrastructure (Exhibit 7). The 10-Q attributes $47M of the $64M rise in infrastructure to data centre, technical infrastructure and hosting costs, and the letter attributes the growth in the company measure to AI training and inference. Sales and marketing rose 66%, $31M of it from advertising and promotion.

Outlook

Management assumed a fall in DAUs between the third and fourth quarters similar to recent years. The guide builds in no short-term improvement in monetisation and excludes the strength management saw entering the quarter from hit games and Halloween events. It includes a modest headwind from GTA VI, which management stated it could not size. Revenue is guided to grow 4% to 12% while bookings fall, because most bookings are recognised over the 27-month estimated paying-user life and the fourth quarter still releases 2025's.

Management expects "fixed cost deleveraging" against the year-ago quarter alongside continued investment in the core app, Universal Player and Roblox Everywhere. Infrastructure expense is guided up from the third quarter for AI initiatives including Build, Roblox Reality and Universal Player, and the CFO stated that infrastructure should grow again year on year in 2027. Management expects marketing aimed at over-18 users to keep rising.

The fourth quarter carries $380M of capital expenditure, including the $140M moved from the third, and without that shift the guide would put free cash flow between -$50M and $35M. Most of the swing from 4Q2025 comes from net deferrals, guided to add $390M against $696M a year earlier, which leaves the float growing at 56% of its year-ago pace (Exhibit 10). FY2026 capital expenditure of about $468M sits at the low end of the expectation management gave at the start of the year, so the build moved within the year without growing.

The spending funds a product calendar whose launches fall mostly in 2027. Management expects Build to reach a broader US release in December and a global rollout in 2027. Roblox Everywhere, which will publish standalone versions of Roblox games to third-party app stores, is due to launch its first phase in the first half of 2027 with select top titles that Roblox curates, and Roblox Reality to reach early access in the same half. Roblox has disclosed no revenue or cost figures for any of them. Roblox Plus has "scaled to millions of active subscribers" since its April launch, and management stated that subscribers spend seven times more Robux than other users, a figure that mixes the effect of the subscription with the selection of heavy spenders into it.

What the Price Implies

This section characterises the financial conditions the current enterprise value implies. It is not a price target, a fair-value estimate, or a view on whether the shares are mispriced. At the 3 November close of $45.45, the market values Roblox's operations at $29.1B after $5.23B of net cash including long-term investments. Holding the model's revenue path, which compounds at 15.5% a year from FY2026 to FY2035, that value requires a flat free-cash-flow margin of 14.5% from FY2027 to FY2035, then 3.0% growth in perpetuity, at a WACC of 10.34%. The close fell 18.1% to $42.99 in the session after the release, 1.5% above the 30 September quarter-end close of $42.37, and stood at $45.45 by 3 November (Exhibit 11).

Memo No. 1 published 20.7% at the 30 July close of $48.63 and 17.3% at the after-hours price of $41.76. The walk restates both at this mark, changing one input at a time (Exhibit 12). Memo No. 1 built its 11.585% WACC on a 5.50% equity risk premium, and the model's 4.28% takes 4.04 points off the requirement from the close. The model's mechanics, a stub period from the mark to year-end and mid-period discounting, and the share price each take about 1.2 points, and the other rate inputs, chiefly the higher 10-year yield, return 0.76.

Beta is the 60-month regression of 1.458 adjusted toward the market to 1.307, and the regression's R-squared of 0.13 leaves a wide band. Across betas of 1.00 to 1.80 the requirement runs from 11.7% to 19.2%, and every point in that range sits below FY2025's reported margin of 27.7% and below the model's path, which averages 26.8% from FY2027 to FY2035 (Exhibit 13). The terminal value now has to capitalise free cash flow of $1.5B, 1.1 times FY2025's, where the primer published $3.2B, or 2.3 times, at its own price and frame, which this memo does not walk.

The two solves are separate readings of one price. Holding the margin path, it needs revenue to compound at 7.0% a year from FY2026 to FY2035, against the path's own 15.5% and bookings growth of 29.2% a year across FY2020 to FY2025. Each solve grants the other's variable in full, and the curve between them traces the combinations that meet the price (Exhibit 14). Held at FY2025's reported 27.7%, the margin needs revenue growth of 6.2% a year.

The 14.5% requirement sits below the free cash flow FY2026 reaches on the guide, about 18% of revenue at the midpoints. That year includes $901M of growth in net deferrals, and FY2025's 27.7% rested on $1,603M, more than its $1,353M of free cash flow (Exhibit 9). Deferrals grow only while bookings run ahead of the revenue they release, so both benchmarks are years in which the float added to cash.

Since Memo No. 1's close, the share price has lowered the requirement by 1.2 points and the equity risk premium by 4.0. The model's path still opens with 20% revenue growth in FY2027, which takes revenue to about $7.2B, above the $6.9B to $7.0B of FY2026 bookings the guide implies.

The Conditions

C6 is the only condition to change status, moving to Developing from Affirmed. C2 and C3 deteriorated within their statuses, and C1, C4 and C5 are unchanged (Exhibit 15). The legal cost of the safety question now runs beside the monetisation overlay as a second overlay.

C2: the failure trigger now measures 2025

To clear 15% growth in the fourth quarter Roblox needs 165.6M DAUs against 138M in the third, while the guide assumes a seasonal decline similar to recent years (Exhibit 16). If the guide holds, the trigger completes with the 4Q2026 results in February 2027, the condition's own first-testable date, and the memo will score the failure on the terms Memo No. 1 set. On management's two-year comparison DAUs are 55% higher, which the trigger as written cannot see.

The friction C2 was built to observe eased, as age-check penetration rose to 64% of global DAUs from 57%, with Australia at 80% and the US and UK at 75%, and passed 80% among DAUs aged 13 to 17. Users who sent a chat message or connected to voice grew 24% on the quarter, twice the platform's growth, and the sequential-growth strengthening trigger is one quarter in, from 123M to 138M DAUs with Russia restored in June, which the guided seasonal decline would reset. On its literal terms penetration has sat below 65% for two quarters, at 57% and 64%. Memo No. 1 read "stalls" as stagnation, and penetration has risen 13 points since 1Q2026, so the stall trigger stays unfired. No third-quarter document discloses the passive age-estimation system as operational, which the condition lists as a weakening sign, so on its literal terms that trigger reads against C2, and both go to the restatement (P5).

C4: the fourth-quarter guide straddles the failure line

Infrastructure and trust and safety expense was 24.5% of revenue, the second consecutive quarter above the 24% ceiling (Exhibit 17). On the fourth-quarter revenue guide the 26% failure line sits at $382M at the low end and $411M at the high end, so the third quarter's $385M already sits above the low end before the increase management has guided. A first quarter above 26% starts the two-quarter clock, and 1Q2027 is the earliest it could complete. Blended cost per engagement hour fell to about $0.0114 from $0.0126 on hours that rose 17% seasonally, and remains under the $0.015 failure level.

The cost-per-hour test assumes a variable cost, and management now describes the growth in infrastructure as largely fixed model-training spend. A fixed cost moves the ratio with revenue and leaves hours a poor guide to it, which tightens the link Memo No. 1 drew between the bookings path and C4. In June Roblox signed non-cancellable agreements to buy $845M of hosting services over three years from 1 July 2026, $248M to $296M a year, so part of the cost C4 divides by revenue is already committed. Capital expenditure was not raised for Roblox Reality, so the condition's capex trigger has not fired. Self-hosted models handled 77% of AI requests, according to the letter, which describes the migration the primer counted as a strengthening sign without giving its cost.

C1: the trigger is met at the floor

Management has stated the over-18 premium at or above 50% for three consecutive quarters while the over-18 share of age-checked DAUs rose to 29% from 27%, and to 37% from 33% in the US, which meets C1's first strengthening trigger on its literal terms. The third-quarter wording moved from "over 50% higher" to "50% above", the condition's minimum. The premium is a US figure extrapolated from age-checked to non-age-checked users, Roblox discloses no over-18 bookings, and the ratio widens whenever under-18 spend weakens, which management stated it did among younger US cohorts. Those three limits keep C1 at Developing.

The cohort's growth also slowed, with US over-18 DAUs up 18% and hours up 13% against 32% and 27% in the second quarter, and the 18 to 34 cohort up 26% against 42%. The mix gain came with marketing aimed at over-18 users at nearly three times the year-earlier level, so part of it may be bought, and that cost sits in sales and marketing, outside every condition. The share improved while its composition did not, so the trend is flat.

C3: no visible step from the over-18 rate

Developer exchange fees were 25.6% of revenue in the first full quarter under the June over-18 rate, inside the 33% ceiling, and 23.7% of bookings against 23.3% in the second quarter and 22.3% a year earlier. The 10-Q names the September 2025 increase of 8.5% and the June over-18 rate as offsets to lower bookings without sizing either, and the year-on-year rise on bookings is consistent with the September increase alone, since 3Q2025 carried it for under four weeks. That is a consistency result, so the over-18 tier's cost may be small or offset by mix. Both ratios rose on the quarter, which sets the trend down, and the first-testable date remains 2Q2027.

On the competitive side, Roblox Everywhere's terms for creators are not yet disclosed. Epic's published creator terms step down twice in the first quarter of 2027. On 1 January Sponsored Row revenue enters the Fortnite engagement pool at 50% instead of 100%, and on 1 February the share of in-island V-Bucks value developers keep falls to 50% from 100%, which would ease the competition for creator time that the primer named as C3's structural risk. Neither of C3's other weakening triggers has fired, because the 10-Q lists no rate increase after June's and no change to Creator Rewards. The 10-Q also introduces Roblox Wallet, absent from both earlier 2026 10-Qs, a future programme under which eligible creators could take their revenue share in fiat directly instead of exchanging earned Robux. It gives no date or terms, and if it displaces exchange payouts, the expense line C3 reads would change with it.

C6: pace, wording and cash

Roblox repurchased 4.4M shares for $200M, all of them in September at an average of $45.87, below the $250M quarterly marker and above the $150M level the condition treats as conservation. Buying in a single month may reflect blackout timing, which the primer treats as noise. The programme has retired 12.6M shares for $580M since the May authorisation, so the $1B twelve-month intent needs $420M more by 19 May 2027, about $166M a quarter, which the third quarter's pace would cover. Fully diluted shares rose 2% to 755M and net cash stood at $5.23B, above the $4.5B comfort level.

The wording around the programme changed between letters. The second-quarter letter stated an intent to repurchase $1B over the twelve months after the announcement, which the third-quarter letter omits and the 10-Q retains as an intent to repurchase up to $1.0B. The CFO ranked capital uses as organic investment first, then optionality for strategic growth that could include "bigger things", then the buyback, which he credited with holding dilution at an acceptable level. P3 expected the free-cash-flow qualifier to bind on the third-quarter guide, and the quarter came in positive. It now binds on a guide wholly below zero, so with the pace below its marker and the intent gone from the letter, C6 moves to Developing.

C5: supply grew inside Other revenue

C5 stays Dormant until its FY2027 first-testable date, and the 10-Q still describes advertising and licensing revenue as insignificant. More than 70% of the top 100 games by Robux spent ran Sponsored Ads, more than 20% of games using rewarded video earned over 10% of their in-game earnings from advertising. The demand management cited on the call came from creators buying traffic, and creators earn a share of rewarded-video revenue. The primer treated advertising as the release valve because it carries no developer payout, so the creator share needs a rate before the valve can be sized, and the filings give none.

The overlays

Tested together as Memo No. 1 proposed, payers and spend per payer net to the 12% fall in bookings (Exhibit 18). Payer conversion and the Roblox Plus subscriber count both belong in the basket, and both remain undisclosed.

The legal cost of the safety question sits in no condition, because C2 tests how age checks affect users and the cost of the regulatory response runs through general and administrative expense. Roblox had settled with at least six states by 29 October, and ten states and two California counties are suing. In October the FTC, Pennsylvania and Utah notified Roblox that they are considering claims under the FTC Act, COPPA and state consumer protection law, and Roblox states that it cannot estimate the loss or range of loss from those or the pending suits. Settlements also carry commitments that are not accrued, including public service campaigns, headcount for law enforcement liaison and platform and policy changes, which will be expensed as incurred. P4 reports the accruals in Exhibit 18, with the count of settled and pending jurisdictions and any federal action, each quarter until the restatement decides whether to formalise it (Exhibit 19).

What Decides It

The 4Q2026 results in February 2027 can complete C2's failure trigger and start C4's two-quarter clock in the same print, and the FY2026 restatement follows them with eight proposals to settle (Exhibit 20).

The February print compares payers with a high base and spend per payer with a low one. Payers grew 94% in 4Q2025 while spend per payer fell 16% to $20.18, the same offset the third quarter carried, in the quarter the guide expects bookings to fall 10% to 15%. A payer count above 36.7M would return payer growth to positive, and Roblox discloses no retention measure that would show retention turning into payers sooner.

If C2 and C4 both fire, the restatement has to decide whether the framework failed or the thesis did. C2 would fail on a comparison with 2025 that was already on the record when the primer set the threshold, and C4 on a cost base management now calls fixed. The overlay that has measured the last two quarters is the instrument that would answer it, and it becomes a condition only if the restatement adopts it.

Sources: Roblox shareholder letter (Form 8-K, Exhibit 99.1), Form 10-Q for the quarter ended 30 September 2026, Investor Data Sheet and supplemental materials, and earnings call, 29 October 2026. Prior-period figures from the Q3 2026 Investor Data Sheet. Roblox 2Q2026 shareholder letter and Forms 10-Q for the first and second quarters of 2026. Memo No. 1 (2Q2026) and the platform transition primer. Epic's published Fortnite creator terms, as at 6 September 2026. Share prices of $48.63 at the 30 July close and $41.76 after hours at Memo No. 1, and $52.46, $42.99 and $45.45 at the 29 October, 30 October and 3 November NYSE closes. Roblox reverse DCF model v3.5, struck at the 3 November close with a US 10-year yield of 4.94%, the Treasury par yield of 30 October, and Damodaran's implied equity risk premium of 4.28% as of 1 August, on 755M fully diluted shares and net cash of $5,229M including long-term investments. Bookings, adjusted EBITDA and free cash flow are non-GAAP measures as Roblox defines them. Age-checked metrics are company estimates and not comparable with the self-reported data used before 2026.