Scaffold Briefing — June 2026

In this Briefing

1. Coursera closes the Udemy combination — a ~$2.5B bet that "skills" beats "courses"
2. Duolingo's user engine accelerates, but AI compute is bending margins down
3. Chegg's 48% revenue collapse is the clearest market signal of AI substitution yet
4. The AI-tutoring evidence base hardens — and the effect sizes are smaller than the hype
5. The "efficacy reckoning" rewires edtech venture funding
6. Pearson defends incumbency by selling skilling and embedding AI in the workflow

Coursera closes the Udemy combination — a ~$2.5B bet that "skills" beats "courses"

What's new. Coursera completed its all-stock combination with Udemy, effective 11 May 2026, folding the two largest Western online-course marketplaces into a single skills platform. The deal had been announced 17 December 2025 and cleared both shareholder votes on 9 April 2026. The strategic framing is explicit: pivot from selling individual courses toward an AI-era "skills" engine serving the enterprise L&D and workforce-reskilling buyer.

Evidence. The combined company carries an implied equity value of roughly $2.5B based on 16 December 2025 closing prices; Udemy holders received 0.800 Coursera shares each — a 26% premium to the 30-trading-day average — leaving former Coursera holders with ~59% and Udemy holders ~41% on a fully diluted basis [1][2]. Greg Hart remains CEO and Andrew Ng remains Chairman. Management is targeting ~$115M in run-rate annual cost synergies within 24 months, with a "significant majority" inside year one [2]. Coursera entered the deal from relative strength: Q1 2026 revenue of $196M (+9% YoY), a 57% gross margin (a three-year high), $12M net income, 7.6M new registered learners (a Q1 record, 205M cumulative), and ~$790M cash, no debt [3].

The Take: This is a defensive merger dressed as an offensive one. The synergy math, not the revenue math, is what closed it — $115M of run-rate cost-out against a combined revenue base of roughly $1.0B (Scaffold estimate: Coursera FY26 guidance midpoint ~$810M plus Udemy's ~$200M annualized) implies the deal is underwritten by ~11% of combined revenue in cost reduction, a margin story for a business whose top line grows high-single-digits. The real prize is consolidating the catalog-overlap problem: two marketplaces bidding against each other for the same instructors and the same SEO real estate now stop competing. Watch for instructor-economics friction — Udemy's long-tail creator base is not Coursera's university-partner model, and reconciling the two revenue-share schemes is the integration's quiet landmine.

Market read. COUR (NYSE) — Add · conviction Medium · 12–24 mo — at ~$5.47 (5 Jun 2026) and ~$1.6B market cap the stock prices in execution risk, not synergy capture; if cost-out lands and enterprise reaccelerates, the multiple re-rates. The thesis fails if integration distraction stalls the 10% consumer growth streak.

Duolingo's user engine accelerates, but AI compute is bending margins down

What's new. Duolingo's Q1 2026 print showed the consumer-engagement flywheel still compounding — but management's own guidance now flags AI inference cost as a structural drag on gross margin, an unusual admission from a company that has marketed AI features (Max, video call practice) as its growth story.

Evidence. Q1 2026 revenue reached $292.0M (+27% YoY); daily active users hit 56.5M (+21% YoY from 46.6M); paid subscribers rose to 12.5M (+21% YoY); bookings were $308.5M (+14% YoY) [4]. For FY2026 the company guides to ~$1.205B revenue and ~$1.28B bookings (a ~10.5% growth point estimate), with adjusted EBITDA of ~$310M (25.7% margin) [4]. Critically, gross-margin guidance steps down from 71% in Q2 toward 69% by year-end, attributed explicitly to rising AI-driven product adoption [4]. The stock closed $123.84 on 11 June 2026 (~$5.68B cap), off a 52-week high of $540.30 [5], and jumped ~10% on 8 June on value-rotation interest [6].

The Take: Duolingo is the first consumer-edtech name to publicly book the AI-margin tax, and the 200 bps of guided gross-margin compression is the number that matters more than the DAU beat. The bull case quietly inverts: the AI features that drive engagement and ARPU also raise variable cost per active user, so the model only works if AI lifts conversion and pricing power faster than it lifts compute. Scaffold estimate: at ~56.5M DAU and a ~200 bps full-year margin step-down on a ~$1.2B base, the implied incremental AI-serving cost is on the order of $20–25M annualized — small today, but it scales linearly with engagement, which is the opposite of the fixed-cost software leverage the stock was historically valued on. The bookings deceleration to ~10.5% is the tell that the easy DAU-to-bookings conversion is maturing.

Market read. DUOL (NASDAQ) — Hold · conviction Medium · 6–12 mo — best-in-class engagement and a now-cheaper multiple after the 77% drawdown from highs, but the AI-margin trajectory and decelerating bookings cap near-term upside; constructive only if conversion offsets compute.

Chegg's 48% revenue collapse is the clearest market signal of AI substitution yet

What's new. Chegg's Q1 2026 results quantify, in real revenue terms, the substitution of a paid vertical-edtech product by general-purpose generative AI. The company itself names ChatGPT-class tools as the cause of accelerating subscriber loss — a rare instance of an incumbent attributing structural decline directly to a horizontal AI product.

Evidence. Q1 2026 total net revenue fell to $63.3M, down 48% YoY; Academic Services revenue fell 57% YoY on lower subscriptions and reduced traffic [7]. The one bright spot was Chegg Skilling at $17.6M (+9% YoY), and the company posted positive net income for the first time in two years on aggressive cost-cutting [7]. Context: the subscriber base had already dropped ~31% YoY to ~3.2M by Q1 2025 [7]. The stock closed $1.11 on 9 June 2026 (~$123M market cap), against a 2021 high above $113 [5].

The Take: Chegg is now a cost-structure story, not a growth story, and the 9% growth in Skilling is the only part of the business with a defensible moat — because skills credentials carry verification value that a chatbot answer does not. The instructive read for the sector: horizontal LLMs commoditize answer retrieval (homework help, Q&A) but not outcome verification (assessment, credentialing, proof-of-mastery). Every edtech business model should be stress-tested against that line. Chegg's survival path is to become a skilling/assessment company before the academic-services cash flow that funds the pivot runs out — a race it is currently losing on the top line even as it wins on net income.

Market read. CHGG (NYSE) — Avoid · conviction High · 6–18 mo — at a ~$123M cap the equity is effectively an option on a successful pivot to Skilling before core revenue zeroes out; positive net income is cost-driven, not demand-driven, and does not change the substitution trajectory.

The AI-tutoring evidence base hardens — and the effect sizes are smaller than the hype

What's new. The center of gravity in AI-tutoring research has shifted from lab demonstrations to randomized controlled trials in authentic classrooms, and a December 2025 RCT of Google's LearnLM across UK secondary schools is among the first to report a clean, like-for-like effect size against human tutoring. The picture that emerges is positive but modest — well short of the "two years of schooling" claims that circulated in 2025.

Evidence. In the LearnLM exploratory RCT (165 students, five UK secondary schools, math on the Eedi platform), students supported by LearnLM were 5.5 percentage points more likely to solve novel problems on subsequent topics — 66.2% vs 60.7% for human-tutored controls [8]. Stanford's Tutor CoPilot study found students whose tutors used the AI co-pilot were ~4 percentage points more likely to pass session assessments [9]. By contrast, the widely cited Harvard RCT (Kestin et al., Scientific Reports, 2025) reported students learned "significantly more in less time" with an AI tutor than in active-learning class, with higher engagement — a larger but lab-proximate result [10].

The Take: Triangulating the credible field RCTs, the durable classroom effect of current AI tutoring clusters around +4 to +6 percentage points on near-term problem-solving (Scaffold estimate, from the LearnLM and Tutor CoPilot deltas) — real, procurement-relevant, but an order of magnitude below the headline claims. The second-order point matters more: the LearnLM result is human-plus-AI versus human-alone, and Tutor CoPilot augments human tutors. The strongest evidence is for AI augmenting tutors, not replacing them — which inverts the cost thesis most edtech valuations rest on. If AI raises tutor productivity rather than eliminating tutor labor, the savings accrue to whoever already employs tutors (districts, tutoring firms) rather than to a pure-software vendor.

Market read. No clean single-name public read; LearnLM is embedded in Alphabet (GOOGL) at immaterial scale. The efficacy bar is a sector-wide input — it favors platforms that can run and publish trials (COUR, DUOL, PSO) over thin-wrapper startups, and is captured in Topic 5's funding read.

The "efficacy reckoning" rewires edtech venture funding

What's new. Edtech venture capital has stabilized at a depressed level, and the qualitative shift among investors and institutional buyers — toward demanding verifiable learning gains rather than engagement metrics — is now reshaping where capital flows. The pandemic-era K-12 funding boom is decisively over; capital is rotating to AI-enabled workforce and teacher-productivity tools that can show outcomes.

Evidence. Global edtech-focused startups raised ~$2.8B in seed-through-growth funding in 2025, roughly flat versus 2024 and a fraction of the pandemic peak; US edtech drew ~$1.2B, near 2023 levels [11]. Early-2026 AI-education seed rounds reset upward — named seed raises of ~$7–12M (e.g., Vimi, Giant, Pensive) — while teacher-facing AI tools (MagicSchool, Brisk, Curipod and peers) collectively raised $90M+ [12]. Industry trackers frame the dominant 2026 themes as the "efficacy reckoning" and "agentic AI," with buyers increasingly requiring evidence of ≥10% gains in mastery or teacher productivity to win contracts [12].

The Take: The efficacy reckoning is a moat-creation event disguised as a funding slowdown. When buyers demand third-party evidence of outcomes, the binding constraint stops being product velocity and becomes access to classrooms for trials and the capital to run them — which structurally advantages incumbents with installed bases (Topic 4's "platforms that can publish trials" point) over thin LLM wrappers. Scaffold estimate: with a ~$10% outcome bar now common in procurement, the effective addressable market for unvalidated AI-tutoring tools is contracting even as total AI-education funding holds, because the trial-to-contract conversion window has lengthened by an estimated 6–12 months. Expect a barbell: well-capitalized platforms and teacher-productivity tools (which can show productivity gains cheaply) thrive; mid-stage student-facing tutors without evidence stall at Series A.

Market read. COUR (NYSE) — Add · conviction Medium · 12–24 mo (reinforces Topic 1: evidence + catalog scale is the moat). DUOL (NASDAQ) — Hold · conviction Medium · 6–12 mo — owns the data to run efficacy trials, but consumer engagement metrics are not the same as learning-outcome evidence buyers now want.

Pearson defends incumbency by selling skilling and embedding AI in the workflow

What's new. Pearson's Q1 2026 trading update shows the legacy education publisher executing a credible incumbent's-defense playbook: lean into enterprise skilling and assessment, and embed AI products inside the tools customers already use (notably Microsoft 365) rather than launching standalone consumer AI apps that would compete with the LLM majors head-on.

Evidence. Q1 2026 underlying Group sales rose 4%, led by Virtual Learning +21%, Enterprise Learning & Skills +8%, with Higher Education and English Language Learning each +2% [13]. Pearson reaffirmed FY2026 guidance: mid-single-digit underlying sales growth, adjusted operating profit of £640m–£685m (at FX as of end-2025, £:$1.35), and 90–100% free-cash-flow conversion [13]. On AI, Pearson launched Communication Coach, an AI product embedded in Microsoft 365, alongside skilling collaborations with IBM, Cognizant and Salesforce; capital returns include a £350m buyback (£219m completed at 964p) and a new £350m 10-year bond [13].

The Take: Pearson is quietly running the smartest large-cap edtech strategy of the cycle: it sells the picks-and-shovels (assessment, qualifications, embedded coaching) into channels it doesn't have to own, which sidesteps the Chegg trap of having a product LLMs can replace. The embed-in-Microsoft-365 move is the tell — Pearson is positioning as the content-and-assessment layer on top of horizontal AI rather than competing with it, which is the defensible place to stand (it monetizes the verification value that Topic 3 identified as AI-proof). The risk is that this same strategy depends on partners (Microsoft, IBM, Salesforce) who can disintermediate the content layer once they have the distribution. Mid-single-digit growth plus buybacks is a respectable, low-drama compounder profile — unfashionable, but the durability is real.

Market read. PSO (NYSE ADR / LSE: PSON) — Add · conviction Medium · 12–24 mo — guidance reaffirmed, enterprise/skilling mix shifting toward AI-resistant revenue, shareholder returns underpinning the floor; the incumbency-defense thesis is intact unless platform partners move up the stack.

Market Calls

References

1. Udemy, Inc. — Form 425 / joint press release, "Coursera to Combine with Udemy to Empower the Global Workforce with Skills for the AI Era" (announced 17 Dec 2025). https://investors.udemy.com/news-releases/news-release-details/coursera-combine-udemy-empower-global-workforce-skills-ai-era
2. Coursera, Inc. — Form 8-K, completion of Udemy merger (effective 11 May 2026); deal terms, exchange ratio, ownership split, ~$115M synergy target. https://www.stocktitan.net/sec-filings/COUR/8-k-coursera-inc-reports-material-event-0f805b46228c.html
3. Coursera, Inc. — Q1 2026 results (Form 8-K, quarter ended 31 Mar 2026): $196M revenue, 57% gross margin, $12M net income, 7.6M new learners, 205M cumulative, ~$790M cash. https://www.sec.gov/Archives/edgar/data/0001651562/000165156226000023/cour-20260331xexx991.htm
4. Duolingo, Inc. — Q1 2026 shareholder letter / Form 8-K (quarter ended 31 Mar 2026): $292.0M revenue, 56.5M DAU, 12.5M paid subs, $308.5M bookings, FY26 guidance and gross-margin trajectory. https://www.stocktitan.net/sec-filings/DUOL/8-k-duolingo-inc-reports-material-event-6974ab47316e.html
5. Stock prices as of 9–11 June 2026 (DUOL, COUR, CHGG closing prices and market caps), per Yahoo Finance / public market data summaries. https://finance.yahoo.com/quote/DUOL/
6. "Duolingo Stock Jumps 10%: What You Need to Know," The Motley Fool, 8 June 2026. https://www.fool.com/investing/2026/06/08/duolingo-stock-jumps-10-what-you-need-to-know/
7. Chegg, Inc. — Q1 2026 results (Form 8-K / 10-Q, quarter ended 31 Mar 2026): $63.3M revenue (−48% YoY), Academic Services −57%, Chegg Skilling $17.6M (+9%). https://www.sec.gov/Archives/edgar/data/0001364954/000136495426000045/a9901-financialresultsq120.htm
8. "AI tutoring can safely and effectively support students: An exploratory RCT in UK classrooms" (LearnLM, Eedi platform), arXiv:2512.23633, Dec 2025: +5.5 pp on novel problems (66.2% vs 60.7%). https://arxiv.org/abs/2512.23633
9. Stanford SCALE Initiative — Tutor CoPilot RCT: students of AI-assisted tutors ~4 pp more likely to pass session assessments. https://scale.stanford.edu/news/how-ai-can-improve-tutor-effectiveness
10. Kestin et al., "AI tutoring outperforms in-class active learning: an RCT…," Scientific Reports, 2025. https://www.nature.com/articles/s41598-025-97652-6
11. "Edtech-Specific Startup Funding Stays Low," Crunchbase News, 2025: ~$2.8B global, ~$1.2B US edtech funding (2025). https://news.crunchbase.com/venture/edtech-funding-stays-low/
12. "Top AI in Education Startups by Fundraising (2026)" / New Market Pitch and related 2026 funding trackers: early-2026 seed raises ~$7–12M; teacher-AI tools $90M+; "efficacy reckoning" / ≥10% outcome bar. https://newmarketpitch.com/blogs/news/ai-education-top-startups-fundraising
13. Pearson plc — Q1 2026 Trading Update (unaudited): Group sales +4%, Virtual Learning +21%, Enterprise Learning & Skills +8%; FY26 guidance reaffirmed; Communication Coach (Microsoft 365); £350m buyback. https://www.prnewswire.com/news-releases/pearson-q1-2026-trading-update-unaudited-302759498.html

Disclosures & Disclaimer

This report is general commentary published for information purposes only. It is not investment advice, a recommendation, or a solicitation to buy or sell any security, and it does not account for the objectives or circumstances of any individual. Scaffold is a research publication, not a registered investment adviser or broker-dealer. Views are the publication's own analytical opinions, are subject to change, and may prove wrong. Markets involve risk of loss; past performance does not indicate future results. Readers should do their own research and consult a licensed financial professional before acting. The publication and/or its principals may hold positions in securities mentioned. Company facts and figures are drawn from public sources believed reliable but are not guaranteed. © Scaffold.

About Scaffold

Scaffold publishes independent technical and techno-economic research in its field. Reports are prepared for subscribers and are provided for information only. © Scaffold. All rights reserved.