NextFin News - Philip Moyer, the former Google Cloud AI executive who took over as McGraw Hill's chief executive in February, has a blunt answer to the question that has haunted education publishers for two years: artificial intelligence will not replace teachers, and the companies betting that it will are chasing the wrong revolution. His argument is not a publisher's defensive reflex. It is the read of an operator who spent five years inside Alphabet helping launch Google's generative AI strategy, then ran Vimeo on an AI-first video playbook, and now runs one of the "Big 3" education companies from a desk in Columbus, Ohio.
The timing sharpens the stakes. McGraw Hill's fiscal third quarter of 2026 — the quarter ended December 31, 2025, in a fiscal year that closes in March — delivered revenue of $434.2 million, up 4.2% year over year, while re-occurring revenue, the subscription-like base that matters most to investors, climbed 14.8% to $357.5 million. Digital revenue rose 11% to $363.7 million. The company raised its full-year guidance. Its remaining performance obligation — contracted revenue not yet recognized — stood at $1.6968 billion at quarter-end, a backlog that gives the guidance teeth. And the company is already public: it priced 24.39 million shares at $17.00 on July 23, 2025, began trading on the New York Stock Exchange under the ticker MH the next day, and closed the offering at a market value of roughly $3.25 billion. In other words, the market is already paying for the very transition Moyer describes: not AI instead of teachers, but AI inside the tools teachers already use.
That is the tension this piece resolves. If AI can explain algebra, grade essays, and tutor one-on-one at near-zero marginal cost, why would a century-old curriculum publisher grow its recurring base by double digits and command a multibillion-dollar valuation? The answer lies in what Moyer calls the "last mile" of education — the part of learning that no model can reach, and where the real money is moving.
Why an AI Executive Took the Helm of a Textbook Giant
The question behind Moyer's appointment is simple: what does a former Google Cloud AI chief have to do with a company built on textbooks?
More than the board initially let on. Moyer became President and Chief Executive Officer on February 9, 2026, succeeding Simon Allen, who retired from the role but remains Chairman of the Board. The company framed the choice plainly: Moyer brings "deep expertise in technology and artificial intelligence, paired with his customer-centric approach," suited to McGraw Hill's "next phase of growth." His résumé reads like a map of the last three computing revolutions — 15 years at Microsoft managing global customer teams, a stint as Managing Director of Global Financial Services at Amazon Web Services, CEO of Vimeo where he launched an AI-first video strategy that produced double-digit improvements in revenue and profitability, and a role as Global VP of Applied AI Engineering and Business Development for Google Cloud, where he helped launch Google's generative AI strategy after joining in July 2019.
He also has an education pedigree that most Big Tech executives lack. Early in his career he co-founded IEP+ Orion System Group, which built one of the first digital Individualized Education Programs to help K-12 schools manage special-education curriculums, testing scores, and regulatory reporting. He later ran EDGAR Online, a financial-data company. That combination — classroom-facing software plus enterprise data platforms plus generative AI go-to-market — is exactly the profile a publisher needs when its product shifts from selling books to selling personalized learning outcomes.
Moyer said as much in his first statement as CEO: "We are a digital-first business with some of the world's most trusted global curricula. We have over a century of learning insights, and proprietary data and analytics. We have deep global customer relationships and, over the past 2 years, we have been rolling out AI solutions at scale and delivering real improvements in education outcomes." Note the sequencing: content and data first, AI second. The company is not an AI company that found education. It is an education company arming itself with AI.
That distinction matters because it inverts the dominant narrative. The panic story is that AI flattens content — that once a chatbot can teach anything, the publisher's moat evaporates. Moyer's thesis is the opposite: AI commoditizes explanation, which makes trusted, sequenced, efficacy-tested content more valuable, not less. A model can generate a lesson on fractions. It cannot, on its own, guarantee that the lesson aligns to a district's scope and sequence, meets state standards, fits a specific curriculum's pedagogy, and produces measurable gains on validated assessments. That is the product McGraw Hill sells, and it is why recurring revenue is compounding while the broader market hesitates.
The most recent numbers reinforce the mix shift. In the fiscal first quarter of 2027, ended June 30, 2026, the company reported revenue of $549.9 million and re-occurring revenue of $425.6 million — roughly 77% of the total — alongside net income of $57.9 million, a sharp swing from $0.5 million a year earlier, and adjusted EBITDA of $207.0 million. Recurring revenue is no longer a supporting act; it is the headline.
In the same release, Moyer put the AI thesis in the company's own words:
AI was a contributor to the momentum we're seeing across revenue growth, margin expansion, price realization, and market share gains. AI represents a genuine tailwind for our business, and our agentic strategy continues to progress, representing an opportunity for meaningful TAM expansion ahead.
The scale behind that statement is not trivial: the company said it serves more than 100 million active curriculum licenses. When AI becomes a tailwind on a base that large, the second-order effects are what investors should be underwriting.
The Human Bottleneck No Model Can Bypass
Moyer's case rests on a mechanism, not a sentiment. In an April 2026 commentary published under his own byline, he laid out the argument in three parts, and each one points to a bottleneck that keeps the teacher at the center of the learning loop.
First, learning is not a data problem.
Learning isn't a data problem. It is physical, social, and emotional—shaped by age, culture, and even what happened at recess. No algorithm captures that. Only a teacher does.
He anchored the point with a striking efficiency comparison drawn from research published in the journal Science: scientists mapped just one cubic millimeter of human brain tissue — smaller than a grain of rice — and found 57,000 cells and 150 million synapses, with the capacity to store more than 705 million bits of information. That rice-sized sliver sifts through billions of bits of data every second on roughly 20 watts of power. Training a frontier AI model, by contrast, requires what he described as "a nuclear power plant and a small lake's worth of water." His punchline: "A day of learning for an average 8-year-old? Fueled by grilled cheese and a cup of tomato soup."
Second, the combinatorics of a classroom defeat any static model. In Algebra 2 alone, he noted, there are 2,384 potential knowledge states. An effective teacher must navigate "trillions of distinct learning pathways to comprehension—in real time, for 20 or 30 kids at once." The teacher does not compute this consciously; she intuits it from the look on a student's face, the mood the child walked in with, the error pattern in yesterday's work. That is the moment Moyer says no large language model has ever felt, and no LLM ever will: the instant a student looks up and says, "I get it."
Third is what he calls the "last mile" of education — the gap between a working technology and a working classroom. "Every generation of startups makes the same mistake: they assume that enough code—or today, enough prompts—can replicate deep domain knowledge, community trust, and the irreducibly human texture that makes a company built to last. It never does." He cited an analysis of venture-backed unicorns showing that only 1% of them sustained their success over the past 20 years, and argued the failure rate is especially brutal in education, where pedagogy "shifts by zip code," by instructor, and by "the look on a child's face at 2 p.m. on a Tuesday."
This is where the cyclical-versus-structural call crystallizes. The fear that AI will replace teachers is cyclical hype — a panic that follows every general-purpose technology, from the PC to the internet to the cloud, and it mean-reverts as the limits of the technology become visible. But the shift underneath the panic is structural and will not revert: learning is moving toward personalized, data-driven, adaptive experiences, and AI is the engine that finally makes personalization scalable. Those are two different forces, and conflating them produces bad investing. The cyclical leg is the chatbot tutor bubble; the structural leg is the rewiring of curriculum, assessment, and instructional time around continuous student-level data.
Moyer's formulation captures both: "machine learning has helped teachers identify comprehension gaps for decades, and now LLMs are powering personalized content, adaptive labs, and immersive games—not to answer questions for students, but to give teachers better tools to ask them." The tool changes. The teacher's role as the designer of the learning experience does not.
Where the Money Actually Is: Content, Data, and the Recurring-Revenue Moat
Follow the mechanism to the money, and a second-order picture emerges. The first-order effect of generative AI in education is obvious: cheaper content creation and cheaper tutoring. Every investor has priced that in. The second-order effect is less discussed: as AI drives the marginal cost of content toward zero, the scarce asset flips from content production to content curation, validation, and integration.
McGraw Hill is positioning for exactly that flip. Its moat is not that it can generate content — any frontier model can do that. Its moat is a century of vetted curricula, proprietary data from billions of student interactions, and deep relationships with school districts and universities that buy on multi-year procurement cycles. When Moyer says the company has been "rolling out AI solutions at scale," he means AI layered onto assets that took decades to build and that competitors cannot reproduce with a prompt. The company has described pilots that let students engage with McGraw Hill's vetted content through their own AI tools — a concrete example of the strategy: own the content layer, remain model-agnostic on top.
The financials already reflect the shift. In the fiscal third quarter of 2026, total revenue grew 4.2%, but re-occurring revenue grew 14.8% and digital revenue grew 11%. Re-occurring revenue of $357.5 million represented roughly 82% of the quarterly base — the kind of mix shift that changes how the market should value the company. A publisher selling one-off textbooks trades on cycles; a company selling persistent, data-rich learning relationships trades on retention and lifetime value. The nine-month picture reinforces it: revenue of $1.639 billion versus $1.628 billion a year earlier, net income of $85.6 million versus $71.0 million, and adjusted EBITDA of $613.7 million, a 37.4% margin. The $1.6968 billion remaining performance obligation at December 31, 2025 adds a forward-looking anchor: this is contracted revenue the company has not yet had to earn.
There is a third-order implication that most coverage misses. If McGraw Hill's bet is right, the winner in education AI is not the company with the biggest model. It is the company whose content and student-level data make any model more effective inside a real classroom. That turns the industry's central anxiety on its head: open models do not threaten the publisher so much as they need the publisher. A chatbot trained on unvetted internet text will confidently teach misconceptions; a system anchored in expert-reviewed curriculum and calibrated by billions of recorded student interactions can be trusted by a district procurement office. The asymmetry is subtle but decisive. Model capability is a commodity racing toward parity. Trusted learning data is not.
That is also why Moyer frames the opportunity as teacher-centric rather than student-alone:
What we actually need is a new generation of teacher-centric tools—built for educators, not as replacements for them—that customize learning at the individual level and finally blow up the one-size-fits-all model that has held education back for generations.
The buyer of those tools is the district, not the teenager with a chatbot. And district buyers pay for efficacy, alignment, and accountability — the exact attributes that keep the moat intact.
The Counter-Case: Open Models and the Free-Content Threat
The strongest argument against this thesis does not come from Luddites. It comes from the economics of open source. A sufficiently capable open-weight model, paired with free open educational resources, could give students high-quality explanations and practice at zero cost. If a motivated student can learn calculus from a free chatbot and free problem sets, why would a district — or a parent — pay McGraw Hill? If that substitution accelerates, the publisher's content moat becomes a museum piece: respected, but no longer essential.
Some investors and edtech skeptics make precisely this case: AI disintermediates the middleman, and the curriculum publisher is the middleman. The bear scenario is not that AI fails in education. It is that AI succeeds so well that the layer that used to package and sell content loses its pricing power.
There are three answers, and only time will tell which holds. First, efficacy and accountability are not optional in formal education. Districts must demonstrate learning gains, meet state standards, and defend procurement decisions to school boards. A free chatbot comes with no efficacy guarantee, no alignment documentation, and no one to sue when it fails. Second, integration is the real product. Teachers do not adopt tools; they adopt workflows. McGraw Hill's advantage is that its AI features sit inside platforms teachers already use, with rostering, gradebooks, assessment, and reporting already connected. A standalone chatbot requires the teacher to build the workflow. Third, the data flywheel: every student interaction improves the product's understanding of how learning actually happens, and that feedback loop is proprietary. A free model trained on static internet text does not get smarter from a classroom in Columbus or Chennai.
Here is the falsifying signal. If McGraw Hill's re-occurring revenue growth decelerates below mid-single digits for two consecutive quarters while market share shifts to open-AI-first entrants, the moat thesis is wrong and the bear case is winning. Conversely, if re-occurring revenue holds in the double digits while the company continues to raise guidance, the content-plus-data model is proving more durable than the disintermediation story allows. Watch the quarterly re-occurring revenue line, not the headlines about AI tutors.
What Comes Next: Beneficiaries, the Exposed, and Three Scenarios
The mechanism cashes out into a clear map of winners and losers. Over the short term — the next two to four quarters — sentiment will swing with every AI-in-education announcement, and the whole sector will trade on the hype cycle rather than fundamentals. That is the cyclical leg, and it cuts both ways. Over the medium term — two to four years — procurement budgets will sort the field: tools that demonstrate measurable learning gains and fit existing workflows will win multi-year contracts; standalone chatbot tutors without efficacy data will struggle to monetize. Over the long term, the structural leg dominates: education becomes continuously personalized and data-driven, and the companies that own validated content plus student-level learning data capture the durable margin.
Beneficiaries include content owners with proprietary, vetted curricula and the data infrastructure to personalize at scale; teacher-facing workflow platforms that embed AI rather than replace the instructor; and assessment companies that can prove efficacy. The exposed are pure-play AI tutor startups without proprietary content or outcome data, generic content aggregators whose product a chatbot can replicate, and any publisher that treats AI as a cost-cutting exercise rather than a product redesign.
Three scenarios frame the path. The base case: AI becomes a feature layer inside existing curriculum platforms, recurring revenue continues to grow in the low-to-mid teens, and the publisher model re-rates as a retention business rather than a cyclical content seller. The upside case: AI-driven personalization produces measurable, publishable learning gains that accelerate district adoption, pushing re-occurring growth above 20% and pulling forward the guidance raises. The downside case: free, capable open models plus open educational resources erode willingness to pay, re-occurring growth stalls, and the sector compresses back to textbook multiples.
The trigger to watch is not the next AI demo. It is the next quarterly re-occurring revenue print, and whether it holds above the mid-teens while the company keeps raising guidance.
Moyer's central judgment survives the scrutiny: AI will transform education, but it will not replace the teacher — because learning is a human relationship that no amount of compute can replicate, and because the economics of schooling reward trust over cheapness. The companies that understand that AI's job is to make teachers more powerful, not obsolete, are the ones building the recurring revenue. The rest are just generating content.
Explore more exclusive insights at nextfin.ai.
