The Coaching Nation Meets the Machine

The size of a funnel built on scarcity

India's households spent roughly Rs 58,000 crore, about $7 billion, on private coaching in 2024, an estimate anchored to National Sample Survey Office household-expenditure data and cited in Parliament that year. The number covers everything from a neighbourhood tutor charging a few hundred rupees a month to the multi-lakh annual fee at a residential coaching campus, and it buys entry into three gates that decide a large share of what a middle-class Indian family considers a secure future: a medical seat, a seat at an elite engineering institute, or a place in the civil services.

Kota, the coaching town in Rajasthan that became shorthand for the whole industry, shows the scale and the strain in one place. Its coaching economy peaked near Rs 6,000 to 7,000 crore in 2023, the year student inflow also peaked at roughly 2 to 2.5 lakh a year. By 2024 that inflow had fallen to about 85,000 to 1 lakh students, and reported revenue had fallen with it, to about Rs 3,500 crore. Coaching operators and local reporting attribute the drop mainly to two forces: online coaching platforms pulling students away from the residential-campus model, and a normalisation after the pandemic-era spike that had briefly pushed numbers higher than the town could comfortably hold. Neither explanation involves AI tutoring, which barely existed as a consumer product in India in 2023.

The scarcity that makes coaching worth paying for is measurable in the exams themselves. NEET-UG, the entrance test for medical school, drew about 24 lakh registrations in 2024 for roughly 1.1 lakh MBBS seats nationwide, a ratio near 22 applicants for every seat. JEE Main, the first stage of engineering admission, drew about 13.2 lakh registrations that narrowed to roughly 24,000 seats across the IITs and NITs. The Union Public Service Commission's Civil Services Examination drew about 13 lakh applicants in the 2023 cycle, of whom 1,016 were finally selected, a selection rate near one-tenth of one percent. A family does not pay for coaching because it wants coaching. It pays because these three numbers make self-preparation feel like a bet nobody wants to take alone.

None of this scarcity is new; NEET and JEE have run at similarly steep ratios for years, and the civil services examination has been a lottery-odds contest for decades. What is comparatively new is the size of the industry built to sell a family a better chance at those odds, and how quickly two very different forces, first cheap online delivery, now AI tutoring, have reached that industry inside the space of a few admission cycles. The rest of this account tries to keep those two forces separate, because the coaching business has already lived through one of them breaking it, before the other had any real presence in the market.

The edtech bubble that already broke, on its own terms

Before AI tutoring reached the market at any scale, India's edtech sector had already run through a full boom-and-bust cycle on its own. Funding into Indian edtech companies fell from about $4.1 billion in 2021, near the peak of pandemic-era online learning, to about $215 million in 2024, an 87 percent collapse. No Indian edtech company has reached unicorn status, a $1 billion valuation, since 2023. The bust predates any meaningful AI product in the category and is better explained by what happened once classrooms reopened: the online-only model that had looked unstoppable in 2021 stopped converting free users into paying ones once students had somewhere else to be.

BYJU'S is the clearest case. The company reached a $22 billion valuation in 2022, raised more than $5 billion in total, and spent about $1 billion of it acquiring the coaching chain Aakash Institute, a bet that its online product and Aakash's physical centres would reinforce each other. Instead the debt load outran the business: Prosus, an early investor, wrote its stake down to zero, BlackRock marked its holding down by about 95 percent, and BYJU'S US entity filed for Chapter 11 bankruptcy protection in February 2024 while the Indian parent entered insolvency proceedings. Within two years, a company that had briefly ranked among India's most valuable private firms was valued at effectively nothing.

Unacademy followed a milder version of the same arc. It peaked at a $3.44 billion valuation in 2021 and, by 2026, agreed to an all-stock sale to upGrad valued at roughly Rs 2,055 crore, about $218 million, roughly 94 percent below its peak; that figure is recent enough to read as emerging rather than settled fact. Against both, PhysicsWallah stands out as the survivor: it listed publicly in November 2025 at about a $5 billion valuation, reported FY25 revenue of Rs 28.9 billion, up 49 percent, and counted 4.5 million paying subscribers split roughly evenly between its online courses and its physical coaching centres. Where BYJU'S borrowed heavily to buy an offline network, PhysicsWallah built a lower-cost online product first and layered offline centres on top of it, funded largely out of its own revenue.

Where AI actually shows up in the funnel today

AI tutoring has entered this market, but through the survivor, not the casualties. PhysicsWallah's "Alakh AI" suite, an adaptive tutoring tool built on a general-purpose AI model, resolved roughly 3 million student queries between September 2025 and early 2026, and the company has partnered with a major AI research lab to develop the product further. It is presented as a supplement to the live and recorded classes that remain the core product, answering doubts at odd hours and walking a student through a problem step by step rather than replacing an instructor.

BYJU'S is a reminder that an AI tutoring announcement is not, by itself, a business strategy. The company had announced generative-AI tutoring models of its own before its collapse, evidence that the idea was already circulating in the sector well before its current wave of attention. The technology did not save BYJU'S, and its absence was not what sank it either; the failure sat in the balance sheet, not in what the product could or could not do.

For the industry as a whole, AI tutoring remains a feature attached to a much larger live-and-recorded-class business, not a separate revenue line reported at scale. PhysicsWallah's own FY25 growth was driven by subscriber additions across its conventional online and offline courses; the AI tutor is described in company communications as an engagement and support tool layered on top, not the reason students enrol. Reading a single usage number, 3 million resolved queries, as proof that AI has changed how India prepares for its exams would overstate what the figure shows. It shows adoption of a feature. It does not yet show a shift in how the money moves.

The far end of the funnel is narrowing too

The reason this matters beyond the coaching industry itself is where the funnel leads. A large share of the students who pay for JEE and other engineering-entrance coaching are ultimately buying access to India's technology and services workforce, the sector that has absorbed the country's engineering graduates for two decades. That workforce is now reporting a different kind of pressure. Entry-level information-technology roles in India were reported down by 20 to 25 percent in 2025, attributed to automation of the routine coding, testing, and support work that has historically been the first job for a fresh graduate. India's technology-sector workforce grew by just 2.3 percent in FY26, even as sector revenue continued to grow, a gap between output and headcount that is itself a signal of the same automation.

Put together, this suggests a chain worth stating plainly, and testing carefully: families spend on coaching because the exams gate degrees, and the degrees gate jobs; if AI narrows the entry-level jobs those degrees were bought to reach, the incentive to spend heavily on exam coaching for that path could weaken over time. That is a thesis, not an observed fact. It describes a plausible multi-year mechanism, not something visible yet in enrolment or spending data; Kota's decline, as noted above, is attributed to online competition and post-pandemic normalisation, not to automation anxiety.

The chain is also not uniform across the three gates. Medicine and government service sit at the more insulated end: a doctor's clinical judgment and legal authority to practise are not substitutable by an AI system in the way that entry-level code review or first-line customer support are, and a government post is created and filled by policy rather than by a company's hiring budget. If the thesis holds anywhere first, it is more likely to show up as coaching demand concentrating toward NEET and UPSC, exams that lead to work automation reaches more slowly, and thinning at the JEE end, which leads most directly toward the technology jobs now under the most visible pressure.

A family weighing where to spend a limited coaching budget is, in effect, being asked to forecast which of three professions holds its ground longest against automation, years before their child sits any exam. That is a harder calculation than the one families were making a decade ago, when the three gates looked roughly equally secure once cleared. It is also not a calculation the current enrolment data shows families making yet; registration numbers for NEET, JEE, and the civil services examination have moved on their own separate cycles of growth and plateau, without a visible reallocation from one gate toward another that a rising automation-consciousness would predict.

The human cost, stated within what the data supports

No account of India's coaching economy is complete without its cost in student lives, and the counts here are the most solid, and the most sobering, numbers in this entire picture. The National Crime Records Bureau recorded about 13,800 student suicides nationally in 2023, a decade high. Kota itself recorded between 29 and 32 student suicides that same year, a town of roughly 2 lakh coaching students carrying a toll that made national news repeatedly through 2023 and into 2024.

What these counts do not do is assign a cause. The National Crime Records Bureau's data classifies deaths by broad category, such as "failure in examination," but it does not code a case specifically to coaching-centre pressure as distinct from family expectation, academic stress more broadly, or the many other factors that contribute to a young person's distress. It would be wrong, on this data alone, to attribute any individual death to coaching, and it would be equally wrong to read the national figure as evidence about Kota specifically, or the reverse. What the counts responsibly establish is scale: a large and, by the national figure, worsening number of Indian students are dying by suicide in years tied to exam pressure, and coaching towns are where that pressure is most visibly concentrated.

Whatever happens to enrolment numbers as AI tutoring and automation reshape the funnel's economics, this is the stake that sits underneath the business story. A funnel that narrows because fewer students see a payoff worth the pressure is not obviously a worse outcome than a funnel that keeps every seat full.

What the evidence supports, and what it does not yet

The temptation, writing about AI and a coaching industry already once broken by economics, is to draw a straight line: AI erodes the jobs coaching was bought to reach, so coaching demand falls. That line is not yet visible in the data. Kota's enrolment decline began and largely played out before AI tutoring products reached any scale in India, and the explanations offered by the industry itself, online competition and pandemic-era over-correction, do not require AI to make sense. The edtech funding collapse of 2021 to 2024 also predates AI tutoring as a mainstream product category. Both busts have already happened, for reasons that have little to do with AI.

Various market-research vendors publish forward estimates for India's AI-in-education market, and they disagree with each other by an order of magnitude depending on what gets counted as "AI-in-education" and over what horizon. None of those projections are treated here as fact; where a number cannot be checked against a disclosed underlying method, it does not belong beside the National Sample Survey Office figure or the exam boards' own registration counts, which is why this account leads with those instead.

What the evidence does support is a market in the early stage of a second disruption, arriving before the wounds from the first one, the edtech funding bust, have fully closed. Coaching survives because the exams it prepares students for remain scarce and consequential, and that scarcity is not going anywhere on its own. What may change is where the money concentrates: toward the gates, medicine and government service chief among them, that automation reaches more slowly, and away from the pathway, engineering coaching aimed at entry-level technology work, that leads most directly into the part of the job market already reporting the steepest AI-linked contraction. That is a shift worth watching across the next few admission cycles, not one the current data has confirmed.

For a coaching operator, an edtech platform, or a college weighing where to put its own effort next, the reading has two parts. Nothing here says the coaching economy is about to shrink; a Rs 58,000 crore market anchored to exams as scarce as NEET, JEE, and the civil services examination does not disappear because a new tool arrives. What it does say is that the composition of that market, which gate families pay hardest to clear, is the more likely place for AI to leave a mark first, well before it shows up as a fall in the aggregate spending figure.