This is a widely repeated claim in Indian education: physical infrastructure is what killed the edtech companies. It is repeated because the headline events fit — Byju's tuition centres closing, FIITJEE branches shutting overnight, Unacademy walking away from its own centres. But the numbers underneath point somewhere narrower and more useful.
What the record actually says
Five of six fit the story. The sixth breaks it — and the sixth is the one expanding offline fastest.
Two corrections worth making
Byju's did not "win online and lose offline". Online was the core business, and it failed there. The offline push — its own tuition centres plus the Aakash acquisition — added a large fixed-cost base to a company whose unit economics were already broken. Physical centres made the fall faster and more visible; they were not the thing that was wrong.
Unacademy did not win online either. Its loss came down substantially, but revenue came down with it — from ₹907 Cr in FY23 to ₹701 Cr in FY25. That is a business getting smaller and cheaper, not a business winning. Exiting company-operated centres for a franchise model is the same move in physical form: convert a fixed cost into someone else's.
The number that looks like the answer, and isn't
The obvious metric is revenue per centre. Both expanding players get worse on it:
So revenue per centre cannot be the discriminator: it fell for the company that is failing and for the company that is fixing itself. It falls mechanically during expansion, because a centre opened in month nine contributes three months of revenue against twelve months of denominator, and because new centres start with a partial cohort.
The discriminator is the direction of the base
Put the two side by side on what was happening to the whole business while centres were being added:
That is the whole finding. Adding fixed cost to a growing business dilutes temporarily and recovers as centres mature. Adding fixed cost to a shrinking business does not recover, because there is no maturing cohort coming to fill the rooms you have already committed to paying for.
Why coaching is unusually exposed to this
Three features of the model stack the risk:
- Demand is seasonal and cohort-shaped. Enrolment happens in windows tied to exam cycles. A centre that misses an enrolment window does not get a second chance that year — it carries twelve months of cost against one intake.
- Faculty is the largest cost and the least flexible. Good teachers are recruited ahead of the cohort and cannot be scaled down mid-year without destroying the product.
- Expansion is the growth story investors fund. Centre count is legible and easy to announce, so it gets committed to before the demand that justifies it is proven.
The last one is the trap. Centre count is an input presented as an outcome. Aakash's network grew while its revenue shrank; the expansion was reported as progress while it was adding cost to a contracting base.
What PW appears to be doing differently
On the public numbers, two things. It expanded into demand it had already built online — an audience that existed before the rooms did, which is close to the opposite of opening centres to create demand. And it kept the blended cost base moving in the right direction while doing it: EBITDA margin went from 6.7% to 14.1% during a 78% increase in centre count.
Worth stating plainly: PW is not yet profitable, its offline average revenue per user has been falling, and a fast-expanding network is a forward commitment whose bill arrives later. A year of good direction is not a proven model. But it is enough to retire "offline infrastructure is what kills you" as an explanation.
What this means if you are buying capacity rather than selling classes
Most organisations running exams, assessments or training are not in the coaching business, but they face the same decision in smaller form: hold space, or book it.
The record above suggests three things:
- Commit fixed cost only behind demand you have already seen. Not forecast demand, not funded demand — observed. PW built the audience first.
- Treat centre count as a cost, not an achievement. The number that matters is what each location runs, not how many you have. "The network grew, revenue shrank" is the failure mode in one line.
- Keep the seasonal peak variable. Capacity you need for two weeks a year is the worst possible thing to own, and the easiest thing to book.
This is the same conclusion as our own infrastructure versus pop-up classrooms comparison, arrived at from the other end — there from first principles, here from the balance sheets of companies that ran the experiment in public.
Where OnSpot Infra comes in
We are the variable-cost side of this decision. The network runs to 730 centres across 225 cities in 35 states — exam labs, training and conference rooms and pop-up classrooms — booked for the days you need them and handed back. The fixed cost stays with whoever owns the building.
Three uses map directly onto the failure modes above.
That is the anchor-plus-pop-up model in practice: hold space only where volume is dense and continuous, and cover everything else on demand.
What this does not solve
Being straight about the limits, because the argument above only works if it is applied honestly in both directions:
- It does not create demand. Nothing here fixes a shrinking business. Booking capacity stops you compounding the problem with fixed cost; it does not fill the rooms.
- Per day, it costs more than a well-used owned site. If a location genuinely runs most of the year, owning it is the cheaper answer and we would tell you so.
- It is not the answer where the room is the product. A flagship centre, or a space with a fixed specialised configuration, is not delivery infrastructure and this comparison does not apply to it.
The decision framework for which of those you are in is in own infrastructure versus pop-up classrooms.
A note on the figures
All of the above is company-reported or press-reported and is cited below. Aakash's and Allen's numbers in particular come from reporting rather than from filings we have inspected, and per-centre figures are our own arithmetic on reported revenue and centre counts, not disclosed metrics. Nothing here is a forecast, and none of it is investment advice.
Sources
- Outlook Business — PhysicsWallah FY26 results and offline expansion
- Inc42 — PhysicsWallah FY25 loss and revenue
- Open — Aakash revenue, EBITDA and centre count post-Byju's
- The CapTable — Byju's downsizing its tuition-centre business
- Inc42 — Byju's locked out of tuition centres over unpaid rent
- Inc42 — Unacademy's reset, revenue and losses
- YourStory — Unacademy moving offline centres to a franchise model
- Entrackr — Allen FY24 revenue and profit
- Careers360 — FIITJEE centre closures and fee refunds