Edtech Startup Funding Slides 26% as AI Tutoring Evidence Splits

A smaller market, still shrinking

AI has been the most talked-about technology in classrooms for two years, but that attention has not brought money back to education startups. Funding for companies built specifically for education stayed weak through most of 2026. The research on AI tutoring, which investors and districts hoped would settle the question of what works, is now pointing in different directions.

HolonIQ's tracking puts global edtech venture funding at $1 billion for the first half of 2026. That is 26% below the $1.35 billion recorded in the same period of 2025.1 The year started slowly. The first quarter brought $512 million across 63 deals, down 24% in value and 10% in deal count from a year earlier.2 Those totals follow a full-year 2025 figure of $2.6 billion, about 11% above 2024. HolonIQ had called 2024's $2.4 billion the lowest annual total since 2014.4 Compared with the 2021 peak, the sector is down roughly 89%.5

Other trackers report larger totals but the same direction. Crunchbase counted about $2.8 billion in seed through growth-stage funding for edtech-focused startups in 2025 as of late November, roughly flat with 2024.13 Tracxn data reported by Rest of World says venture capital fell from a 2021 peak of $16.7 billion to under $3 billion by 2025.11 For the first half of 2026, Crunchbase reports just under $1.8 billion raised by edtech and education-related startups.20 That is well above HolonIQ's $1 billion, which most likely reflects how broadly each firm defines "education-related." Whichever number you use, the conclusion is the same: the money has not come back.

Deals keep coming, but checks shrink

The more telling detail is how the money is spread. HolonIQ found deal volume in the first half of 2026 roughly stable even as total value fell. It reads that as continued investor appetite combined with smaller average deals.1 An independent analysis of 12 months of deals through September 2026 reached a sharper conclusion. It counted 100 deals against 99 the year before, while disclosed capital fell 41.9%, from $1.334 billion to $775.6 million.12 The median round in that dataset dropped from $4.4 million to $3.05 million. Series B funding fell from about $521 million to $72 million.12

The weakest part of the market is now the middle. In the same analysis, rounds under $20 million made up 92% of deals. Only four financings landed between $20 million and $100 million, compared with 16 a year earlier.12 Founders are also raising later: the median company age at first financing doubled from about 1.5 years to three.12 In short, getting a first check is still possible, but raising the larger follow-on rounds needed to grow has become much harder.

The new-company pipeline has also shrunk. Tracxn data shows 645 edtech companies founded in 2025, compared with almost 10,500 in 2020.11 K-12 has been hit hardest. Rest of World reports that K-12 funding has fallen 82%.11

Where the money goes

The capital that is still flowing goes mostly to adult and workforce learning. Workforce training took about 70% of venture value in the first quarter of 2026.2 It stayed the largest category in the first half, led by Preply's $150 million round and Multiverse's $70 million.1 Preply's Series D valued the company at $1.2 billion and made it HolonIQ's newest edtech unicorn. The company says it had reached EBITDA profitability before raising.417 Eruditus raised a $150 million Series F only after becoming profitable, which one analysis describes as the new minimum for large rounds.15

Looking back at 2025, HolonIQ described investors putting money into AI-enabled products, workforce-focused platforms and K-12 tools that address cost and staffing pressures.11 Crunchbase similarly noted interest in healthcare training. In K-12, it saw investors backing AI tools that personalize lessons and take routine work off teachers. Coding bootcamps, by contrast, have fallen out of favor as coding automation spreads.13 The largest K-12-adjacent AI rounds in 2025 were modest. MagicSchool AI raised a $45 million Series B for teacher productivity tools.13

That is where the AI-tutoring funding story stands. One tracker followed only companies whose core product is AI-driven teaching, learning or assessment. Between August 2025 and September 2026 it found just 32 disclosed equity rounds worth $232 million across 30 companies. None was larger than $50 million, and the median was $5.2 million.14 Rounds for tutoring companies specifically are small. Third Space Learning raised about $5.95 million, partly to develop its spoken AI tutor. AILA raised $3 million. The median digital tutoring round fell from $4 million to about $3.18 million.12 Medly AI raised $8 million for an AI exam tutor used by more than 400,000 students, citing early randomized-trial evidence of higher science scores.17

These figures may understate interest in AI for education. Crunchbase has pointed out that edtech-specific totals leave out general-purpose AI platforms that students and teachers use heavily.13 This is the key structural point. Much of the money flowing into AI learning goes to frontier model makers, not to edtech startups. Specialized vendors have to justify their existence next to free general-purpose chatbots.

The evidence on AI tutoring cuts both ways

Investors increasingly want proof that products improve learning. Rest of World cites a startup-failure analysis listing low retention linked to "murky learning outcomes" among edtech's long-running problems, along with expensive customer acquisition and slow institutional sales.11 So the research matters for funding, and it does not all agree.

The positive results are real. In a Harvard randomized trial published in Scientific Reports, undergraduate physics students using a custom AI tutor learned significantly more, in less time, than students in a well-run active-learning class. They also reported feeling more engaged. Reported effect sizes range from 0.73 to 1.3 standard deviations, and median time on task was 49 minutes for the AI group compared with 60 for the class. Stanford's National Student Support Accelerator described two newer trials in live, chat-based math tutoring. In one, students tutored with Google's LearnLM did better on later, harder topics: 66% succeeded, compared with 61% of students with human tutors only and 56% of those given static hints. Supervising tutors approved about 76% of the model's responses with little or no editing.21 In the other trial, Tutor CoPilot raised topic mastery by four percentage points. Gains reached up to nine points for students whose tutors were less experienced or lower-rated.21

A new trial points the other way. A working paper released in October 2026 studied a generative AI tutor built into courses for 2,379 undergraduates and 30 instructors at a large U.S. public university. Students with access to the tutor ended up with final grades 0.37 standard deviations lower than peers in sections of the same course. Their participation in the learning management system fell by 0.90 standard deviations. First-generation students lost the most. Other reviews describe a similar pattern: gains are concentrated in structured subjects such as algebra. AI tutors tend to tie or slightly trail small-group human tutoring when both are well staffed. Short pilots often end before students build steady habits.27 Some commentary also warns that always-available AI help can reduce students' self-monitoring and increase procrastination.26

These results are not really in conflict. Together they suggest that design and implementation decide the outcome. The Harvard tutor was built around the same teaching principles as the class it was compared with. The Stanford trials kept humans in charge of the tutoring.21 The negative trial tested a tutor added to existing courses at scale, and it appears to have replaced effort students would otherwise have made. The finding that holds across the studies is that AI tutoring helps when it is carefully designed and supervised, and can hurt when it is simply made available.

Why it matters

The funding data and the research data point to the same conclusion. Capital is going to companies whose results can be measured, such as language skills, job placement, exam scores or profit. Large rounds increasingly require visible business results, not growth stories.17 One market analysis of K-12 AI companies says the startups still raising are those building tools for teachers, not around them. Stand-alone consumer apps used without teacher oversight are struggling.15 The trial evidence supports that investor preference. The tools with the strongest results keep a teacher or tutor involved.

HolonIQ expects the current pattern to continue: money concentrated in workforce training, with AI products still attracting some investment in K-12 and higher education. It does not expect deal sizes to grow much unless market conditions change.2 It also notes that investors are still waiting for returns on bets made during the pandemic.1 That leaves edtech in the position it has been in for the past three years. Education startups get less money than general AI companies, they have to show evidence of results, and they have to stay small until they can show it. For AI tutoring startups, the October trial result means investors will likely want rigorous, independent proof of learning gains before backing them with larger rounds.