SPOT (Trans): Subs Top 300mn, Monetizing MAUs

Below is Dolphin Research's summary of the $Spotify(SPOT.US) FY26 Q2 earnings call.

Key Takeaways:

I. Core Financial Highlights

1. Shareholder returns: As of Aug 1, the company had repurchased $662 mn YTD, up 30% vs. 2025. Since buybacks resumed in 2025, it has repurchased nearly 2.2 mn shares, approx. 1% of the float. Cash and cash equivalents stood at €9.4 bn at quarter-end with no debt other than lease liabilities.Management expects stronger cash flow in coming years and intends to keep returning cash, even with M&A.

2. Q3 guidance

a. Users: MAUs of 788 mn (+11 mn QoQ) and Premium subs of 305 mn (net adds of 5 mn). MAU guidance embeds product optimization in emerging markets, while growth in developed markets remains steady.

b. Revenue and profit: Total revenue of approx. €5.0 bn (+14% YoY) and GPM of 32.9% (+~130 bps YoY). Operating profit of €670 mn.

c. Guide definitions: ARPU growth consistent with Q2. Guidance figures are rounded; excluding FX and social insurance effects, opex growth in Q3 is roughly in line with Q2.

3. This quarter's key metrics

a. Totals: Revenue of €4.8 bn, +15% YoY at constant FX (Q1: +14%). GPM of 33.4%, 30 bps above guidance and +193 bps YoY. OP of €655 mn, above the €630 mn guide, implying OPM of 13.7%. FCF of €797 mn (+14% YoY), slightly below Q1 on timing of cash taxes, with H1 working-capital benefits flat YoY.

b. Segments: Premium revenue rose ~16% YoY (Q1: +15%), driven by +9% subs and +7.4% ARPU. Ad-supported revenue grew +3% YoY, flat vs. Q1.

c. One-offs and upside drivers: GPM beat was mainly a timing mismatch in growth investments across quarters, plus a small one-time benefit from the reversal of prior accruals after Canada scrapped its digital services tax. Equity volatility drove a €9 mn non-forecastable benefit in social insurance expense, and OP still beat guidance by €16 mn ex. that.

4. Full-year opex and profit cadence: Marketing and AI-related spend will add ~€200 mn of incremental full-year opex. Opex will be temporarily higher in Q2–Q3 and then slow in Q4 as marketing peaks roll off.Headcount will be flat in 2026, and this investment cycle excludes structural staff increases. While the company does not guide full-year GPM/OPM, both are expected to improve YoY with a marked increase in FCF.Quarterly margin cadence will depend on launch timing, and new features may temporarily temper GPM expansion.

5. Mid/long-term targets (May Investor Day): Revenue CAGR in the mid-teens, GPM of 35%–40%, and OPM above 20%, with robust FCF growth. Since the 2022 Investor Day, revenue has compounded at +18% to €17 bn in 2025, while GPM has improved from 25% to above 33% this quarter. FCF for 2025 is €2.9 bn.

II. Detailed Call Commentary

2.1 Management Highlights

1. Users and engagement

a. Q2 MAUs grew +12% YoY, with Europe and North America materially beating internal expectations. Net MAU adds were 16 mn, 1 mn below internal forecasts.

b. Net Premium adds were 7 mn, with growth across all regions QoQ and outperformance in Rest of World and North America. Premium subs reached 300 mn, 1 mn above guidance, marking the first-ever break above 300 mn.

c. Over the past five years, net Premium adds have exceeded 25 mn annually, with balanced growth across developed and emerging markets. Current MAUs are about 777 mn (transcript misread as '77 mn'; inferred from Q3 guide of 788 mn minus +11 mn adds).d. Monthly active days among global Premium users continue to rise, which management views as the most critical engagement metric.

2. Emerging markets: shifting from 'growth lever' to 'monetization lever'

a. MAU outperformance over the past 4–5 quarters was driven by emerging markets such as India and Indonesia. These are large population pools with strong potential.

b. Actions include tightening sign-up flows to improve MAU quality, dropping support for low-end Android devices, carefully increasing ad load and feature limits on the free tier, and product optimization.

c. Intent and impact: proactively add friction to free usage to drive higher conversion and future revenue. Q3 MAUs will reflect this impact, though management stressed no near-term hit to Premium sub growth.

d. Free-to-paid cycles differ from mature markets and will take time, but the upside is substantial. This represents incremental capture of existing potential.

3. Ads

a. The ad stack migration completed early this year, moving fully onto Spotify's proprietary stack, with ~99% of impressions on the in-house ad stack. With the inventory now on the in-house ad server, sales flows can be streamlined to capture more demand.

b. Automated channels contributed nearly 40% of ad-supported revenue (slightly above 30% in Q1), and management expects further gains. This upside was largely offset by an anticipated decline in direct sales, which is now stabilizing post pricing optimization.

c. Active advertisers reached 33k (+60% YoY). Plugins and MCP for Claude, ChatGPT, and Gemini are live, enabling prompt-based campaign and audio asset creation, with 7k advertisers already using AI audio-generation tools.

d. Supply and demand: supply is expanding via broader reach from user growth, new free-tier ad slots, personalized ad load, and deeper interaction. Demand-side self-serve and auctions remove the pricing and sell-through ceiling of the old email/IO/phone plus guaranteed model.

e. Outlook unchanged: ad biz is expected to return to double-digit growth in 2H26. Margins should advance from the current ~20% range toward ~40%.

4. AI and personalization (Taste foundation models)

a. The Taste models learn from 3.4 tn daily events on the platform. Spotify has invested in AI-driven personalization for 7+ years.

b. In the first two months of the new recommendation system, monthly active days increased despite a high base, with significant gains in Autoplay time and song saves, and lower Autoplay churn.

c. In parallel chat experiences, listening time grew double digits, with increases in active days and saves. These are tightly correlated with retention and LTV and are typically the hardest to move.

d. AI experiences such as DJ now reach roughly one-quarter of active users. Prompted Playlist has 14 mn users among the first 100 mn users onboarded, with encouraging early retention signals.

e. 'Talk to Spotify', personal podcast, and Studio by Spotify are live, and Prompted Playlist will expand into audiobooks in the coming weeks.

5. Premium value-add and new products

a. Reserved (ticket reservations): Launched with Live Nation in the US in Jun, supporting multiple tours and reserving nearly 100k tickets via Spotify. Some tour allocations sold out 100%, prompting Live Nation to add capacity mid-tour, and management called it one of the largest value adds ever put into Premium.

b. Audiobooks: Premium includes ~15 hours. Audiobook penetration among Premium listeners has more than doubled this year, and the Audiobooks+ add-on has surpassed $100 mn in ARR even though it is live in only a handful of markets.

c. Song DNA: Over 100 mn Premium users have used it, making it one of the fastest-adopted features ever, enabled by the WhoSampled acquisition.

d. Social (single-player → multiplayer): Messages launched so users can react, reply, and share within Spotify, with Listening Activity showing what friends are hearing in real time. Listening stats have shifted from private reports to shareable comparisons, and Jam's monthly users are nearing 50 mn.

e. Fitness: Running mode rolled out last week, letting users describe pace or intervals in natural language (e.g., 8-min/mile). It matches tempo and BPM to step cadence with seamless speed-mix transitions and optional coaching, updated weekly.

f. Music videos: After sustained investment, both experience and catalog are in place, and tracks with MVs (especially new releases) perform noticeably better.

6. Music remix/covers and rights partnerships

a. Following the May agreement with UMG and Universal Music Publishing, Spotify announced a deal with Merlin, the digital licensing partner for leading independent labels and distributors, covering 30k labels in its network.

b. The model rests on the 'three Cs': artist consent on which songs are eligible for remix, proper credit, and compensation across labels, publishers, artists, and songwriters. Management frames this as the first lawful path to ride the interactive music AI tailwind.

c. Response from the indie community has been strong, and management believes the model is winning industry buy-in. By design, nearly all rightsholders who want in should be able to participate.

d. Timeline: substantial work remains before a formal launch, and scaling to a material revenue driver will take time.

7. Engineering efficiency and AI cost control

a. Honk is an internal coding agent that lets engineers fix bugs or add features via Slack from a phone, delivering a testable build by the time they reach the office. This workflow is now widely adopted.

b. Chirp powers Honk and replaces direct calls to coding tools, enabling mid-task model switching and routing each job to the best value stack, including self-hosted open-source models, to avoid vendor lock-in.

c. Chirp also shares context across models, developers, and companies, preventing double payment for the same inference and avoiding data leakage, while surfacing per-developer inference spending.

d. Cost philosophy: to be an AI beneficiary, Spotify must also win on costs. Headcount has not increased over the past three years, and revenue per employee is set to double; opex growth is driven by compute and marketing, both variable and fully controllable.Inference volumes are company-determined, keeping the cost curve manageable. Investment hurdles are high: double down on what works and exit what does not.

e. Margins are managed outcomes, not byproducts.

8. Scale positioning and monetization framework

a. Scale advantage: very few companies globally have hundreds of millions of recurring paying users, and Spotify achieved this with a single product. Citing Munger, management views scale as the only structurally superior advantage, enabling the most cost-efficient, competitive product investment via cash generation.

b. Power-law usage: product, feature, and content consumption follows a power-law distribution, with tens of millions of users wanting more and willing to pay, first validated by audiobooks. The platform can decide how much inference to allocate to free vs. Premium, with paying users getting priority.

c. Inference-based products carry per-use marginal costs and need matching monetization. This leverages 20 years of Premium know-how via feature gates, usage-based thresholds, and value pricing; the first credit-based add-ons have launched, with each credit tied to a defined inference amount, aligning usage, costs, and monetization.

d. Coverage spans commuting, fitness, learning, gaming, dining, and sleep throughout the day. Most products only occupy one or two of these contexts.

2.2 Q&A

Q: Which products are you most excited about now, and which are most likely to gain traction over the next 3–5 years?

A: At a system level, software used to be an amortization game: build once and spread cost over more users. Inference introduces per-user variable costs, changing the landscape, which can be headwind or tailwind.So we shifted the model: usage follows a power law, with some users consuming far more than others. That did not matter in the amortization era, but it is critical now, so we built a platform that can choose how much inference to allocate to free and Premium. When users hit limits, the answer is not 'stop here', but 'pay to move into the future', letting those willing to pay run ahead.Audiobooks fully validated this model: Premium includes ~15 hours, and heavy listeners can pay to go well beyond that. This took time to build because we do not like shipping ideas without infrastructure, hence the Investor Day reveal of a single platform serving different user types, monetization levels, and needs, without being constrained by averages or the lowest-engagement cohort.

On products: first, Reserved is the feature most often called 'the best thing you have built', with strong early data and among the biggest value-adds ever placed in Premium, enabled by a unique deal structure. Second, the Taste model, which required material staffing and training spend, based on a bet that legacy ML had plateaued while sequence-based LLMs obey scaling laws, delivering better results with more compute, parameters, and data, now paying off in taste and recommendations. Third, Song DNA, enabled by the WhoSampled acquisition, which already has 100+ mn users and is similarly unique.

Also, music videos now have both experience and catalog in place after long-term investment, and songs with MVs, especially new releases, perform much better, matching our initial bet that MVs are disproportionately valuable in new-artist discovery. Gen Pod and running mode are roughly tied: nothing else matches users' running cadence with their favorite tracks while speed-mixing; and Gen Pod is a previously nonexistent podcast about the user, covering taste, new music, missed podcasts, books to read or publish, and even happenings in the user's own neighborhood. Signals are very early but promising.

Longer term, remix and covers are highly imaginative because others cannot do it at scale: generic generative music will happen with or without Spotify, but this specific product requires Spotify and enables existing artists to participate. It is a lot of work and needs more time. Lastly, Talk to Spotify is rolling out in the US, letting users ask about music, band members, tour dates, and song meanings, with higher answer quality than any general LLM across music, podcasts, and books.

Q: Despite programmatic exceeding 25% of ad-supported revenue, ad growth has been soft over the past year. What is the outlook over the next one to several years, and how can you re-accelerate this high-margin business?

A: Stepping back, ad sales have been in transition for two years, with the migration completed early this year. We now run entirely on Spotify's proprietary ad tech stack, with ~99% of impressions on our own ad stack, and the shift to automated channels is the bigger change, nearing 40% in Q2 vs. 30% in Q1, with further runway.

Supply has never been stronger, not only from user growth but also from MFT Plus and new free-tier ad slots, personalized ad load, and deeper interaction. Together these create ample supply to match demand.

On demand, since inception until two years ago, buying ads on Spotify meant phone or email, purchasing fixed guaranteed volumes, which capped both pricing/sell-through and access for buyers who prefer automation over emails/IOs/calls. Both constraints are now removed: most buying is self-serve and automated, and buyers who see value can bid prices up. As a result, it is not just migration of incumbents but also more active advertisers, now 33k (+60% YoY), aided by our plugins and MCP for Claude, ChatGPT, and Gemini that let advertisers create campaigns and audio assets via prompts, with 7k already using AI audio tools. The plan is right; it is about continued execution.On margins, with automation and self-serve, ads become a scale business. As scale builds and music/podcast ad monetization improves, profitability should rise; consistent with Investor Day, we aim to move from ~20% toward ~40%. As music ads scale in emerging markets, ad sales should show healthy margin flow-through.

Q: Are you surprised by the number of artists opting in to the AI music layer? What are the barriers to artist participation?

A: After the May deals with UMG and Universal Music Publishing, we announced a Merlin partnership today, giving 30k labels in its network the opportunity to participate in this covers/remix product. We are deliberate and structured around the 'three Cs': consent to include works in the library, credit, and compensation for labels, publishers, artists, and songwriters.Momentum is strong. It is important to distinguish concerns around 'from-scratch artificial music' from what Spotify is doing, which is different: this is for real artists, not fake ones, and in the remix case it is real artists' real vocals, which is a fundamentally different value proposition not available today.Artists and consumers are receptive, though the landscape is evolving. Next, expect a research preview to gather preference data before full catalog is needed, not as a product test but to enable reinforcement learning to tune the models, leveraging our 777 mn users and music fans for RL. We will launch only when it is mature and compelling for consumers.

Q: By 2026, open-source models will have advanced meaningfully. How do you weigh costs and benefits of broader open-source deployment vs. using models like Anthropic Claude today?

A: The open-source movement helps us. Our Taste model is based on open source with CPT (continuous pre-training), taking an open model and continuing training on proprietary data, and it performs well. Open-source models are arriving faster, better, and cheaper, which benefits consumer products.It also helps our dev environment and R&D costs. Internally, Chirp can switch between paid and self-hosted open-source models while preserving project context, which is how we maintain best value. We plan to offer Chirp to other companies because we are hearing similar needs.Also, the industry is cutting prices, with per-token costs falling. For a given unit of functional quality, costs are dropping meaningfully. That does not mean we will avoid more advanced models, but same-quality costs are declining rapidly.

Q: Why does Q3 MAU growth guidance slow? Does this reflect changes in competition or marketing acceptance, and what will drive MAU growth going forward?

A: Fundamentals are strong; we just surpassed 300 mn Premium subs and continue to grow. MAUs have beaten expectations for years, with outperformance over the past 4–5 quarters from emerging markets such as India and Indonesia, which we have called out before.We are now adjusting product, value proposition, and strategy in these markets: tightening sign-ups for higher-quality MAUs, ceasing support for low-end Android devices to improve efficiency, and adding measured friction via ad load and free-tier feature limits. These moves prepare for better monetization, shifting from the growth lever to the monetization lever.This takes time, but the curve follows a familiar pattern: establish product-market fit and initial MAUs; MAU growth rises steadily; a convertable base emerges and Premium conversion starts; then, with calibrated product and value, growth follows a LATAM-like path where early conversion was low but MAU growth was strong. Importantly, this will not affect near-term Premium sub growth.

Q: After the Role Model campaign, how many tours have used Reserved? Have you seen higher conversion to paid for Reserved, and how will Spotify's role in live evolve?

A: Users rate Reserved highly, and Live Nation has also received positive feedback on the feature and partnership. The logic is a three-way win: superfans get tickets, artists fill shows with their most committed fans, and Spotify offers exclusive value to Premium subscribers, creating differentiation and improving perceived value-for-money.It is still early, with only a handful of tours and currently limited to US Premium members. Roughly 100k tickets have been reserved through Reserved; in some cases the allocation sold through 100%, and Live Nation added capacity mid-tour. We look forward to more tours and markets, and the near-term goal is to enhance Premium value.

Q: With deals in place with UMG and Merlin, do you need all majors onboard before launching AI services? Why?

A: We do not need all majors. We want as many artists as possible, but we do not expect everyone.In Spotify's early days, key catalog gaps persisted for years, with acts like the Beatles or Metallica arriving much later, so a full catalog is not a prerequisite. More is better for both consumers and creators, and we will proceed with research previews in public to improve the product and start collecting preference data that can automatically improve models. Launch timing will depend on progress, and momentum is strong.

Q: With Premium subs reaching 300 mn in Q2, how much monetization upside can interactive tools and live integration drive across the base?

A: This spans opportunity, TAM, pricing, and product, and we are still early. Hitting 300 mn is rare air; very few companies globally bring 300 mn recurring customers back monthly on a single product, approaching 4% of the world's population.We might never cover 90% of the globe, but 15% penetration is not out of the question. The priority is to push value-for-money higher by building interactive tools like covers and remix, plus taste profiles and personalized podcasts.Audiobooks are a good example: add-ons structurally lift ARPU versus simple price increases. On MAUs, we will differentiate between developed and developing markets. Overall, the opportunity remains significant.

Q: Spotify had 211 bn hours of content consumption in 2025, above Netflix's 191 bn. With expansion into fitness and deeper personalization, how will time spent evolve?

A: A key difference vs. streaming peers like Netflix (and sometimes YouTube) is device and context coverage. While others are predominantly big-screen (plus small-screen), Spotify spans speakers, consoles, and cars, and contexts like sleep and study, with far broader compatibility.This creates a 'universe of contexts' spread across billions of hours. Among engagement metrics, we focus on active days: over 100 mn Premium users use Spotify more than 20 days per month, and this ties strongly to LTV.This metric continues to climb, and monthly active days for Premium users improved again this quarter.

Q: You have launched multiple AI products in the last six months, including Prompted Playlist. What have you seen on listening time, conversion, and churn?

A: AI experiences now reach about one-quarter of active users. 'Talk to Spotify', Studio, personal podcast, and Prompted Playlist launched this year are scaling quickly, with adoption above 25% among actives.Prompted Playlist has reached ~14 mn out of the first 100 mn active users onboarded, which is fast for such a feature. The Taste model is moving the hardest metric, monthly active days, which we prioritize over instant engagement as the key driver, and we are seeing the impact now.Since 2010, personalization has shown a clear correlation with retention.

Q: What is behind the implied acceleration in Q3 opex?

A: The ~€200 mn opex uplift this year is fully controlled and non-structural. More precisely, excluding FX and social insurance impacts, Q3 opex growth should be roughly in line with Q2, and remember guidance is rounded, with revenue at ~€5.0 bn (+14% YoY) and ARPU growth consistent across Q2 and Q3.So Q3 stays consistent, with scope to slow in Q4. Costs are non-structural with a stable headcount base.The spend is aimed at engagement and LTV and marketing, with many new features launching and a bit more AI, as flagged at Investor Day. Nothing unusual and fully controllable.

To add, we invested early to lead this AI application wave, and feedback suggests we are ahead. We are now entering phase two: focusing on cost and efficiency, which Chirp addresses by giving us control over model choice and spend.The Q4 opex deceleration message stands. These are compute and marketing costs within control, not multi-year capex commitments.

Q: How are you thinking about launch timing for category add-ons, and what is the impact on Premium ARPU from 2H this year through 2027?

A: We will not comment on timing or guide ARPU. Framing it differently, usage-driven monetization is critical for both free-to-Premium and Premium-to-add-on conversion.One example we can share is Audiobooks+, which has doubled since last disclosure despite being in only a few markets, and it continues to grow. This adds another ARPU lever: beyond list-price hikes, successful add-ons structurally lift ARPU through quantity, a different kind of 'price increase' than simply raising Premium rates.

Q: Apple Music's Jul price increase — any read-through, and how does it affect your thinking on pricing?

A: We will not comment on others' pricing, but it signals ongoing value-add in music streaming, which is healthy for the ecosystem. Our focus remains category leadership, not only in users and subs but also in engagement.The more value we deliver, the stronger our pricing power. We are comfortable being the price leader in the category.

Q: Labels noted slower Premium growth for the industry in Q2, but your net adds were solid. How confident are you in long-term Premium sub growth for Spotify and the industry?

A: Very confident. While we do not provide long-term guidance, we are optimizing a healthy funnel to grow, not slow, Premium subs.This supports both our own business and the broader subscriber base.

<End of transcript>

Risk Disclosure and Disclaimer:Dolphin Research Disclaimer and General Disclosure