By Ramachandran Rajeev Kumar — 2026-07-06

The Ladder and the Racket

Bharath Manthan - Episode 14

By Ramachandran Rajeev Kumar

The rung that built India's middle class is the rung AI automates first. Our education industry is selling tickets to it anyway.

On February 16, 2026, India's biggest AI showcase opened in Delhi. At one pavilion, a private university unveiled what it described as a ₹350-crore AI ecosystem. At the centre of the stall stood a four-legged robot named ORION - the "Operational Robotic Intelligence Node". It walked. It responded to commands. It charmed the cameras.

In an interview to DD News, the university's professor of communications said ORION had been developed by the institution's Centre of Excellence.

Within a day, viewers had identified the machine. It was a Unitree Go2 - a commercially available quadruped built by a Chinese robotics company, purchasable online by anyone with a credit card. Observers then identified a second exhibit at the same stall, a soccer-playing drone, as a South Korean product. On February 18, Reuters reported that the university had been asked to remove its display. Officials reportedly cut power to the stall and barricaded it. The university first disputed reports it had been ousted, then issued an apology, calling the episode a miscommunication.

The robot did nothing wrong. It walked, it obeyed, it performed exactly as built. The question worth losing sleep over is why the institution presenting it did the same.

This is not an article about one university. Institutions, like robots, execute their programming. The programming of Indian higher education - particularly its vast IT wing - is the subject of this churning. Because the incident at that pavilion was not an aberration. It was a business model caught on camera.

The Ticket Business

Start with the arithmetic we already know.

India produces roughly 1.5 million engineering graduates a year - more than any country on Earth. What happens to them next depends on whose survey you buy. Mercer-Mettl's Graduate Skill Index 2025 finds 42.6% of graduates employable - down from 44.3% two years earlier. The India Skills Report 2025, prepared in collaboration with AICTE itself, offers a far rosier 71.5% for engineers. When the regulator helps produce the flattering number, treat it the way you would treat a placement brochure. Meanwhile, in the 2025-26 academic year alone, 58 engineering and technical colleges entered progressive closure - barred by AICTE from admitting a single fresh batch.

Hold those numbers together and the picture resolves. Most of this system does not sell education. It sells a ticket.

The ticket has a precise face value: admission, a degree, a placement day, an offer letter. For three decades the ticket was honoured, because at the top of the ladder stood an industry with a bottomless appetite for ordinary graduates. BPO and IT services took lakhs of average engineers from average colleges, trained them for twelve weeks, and put them on client work. It was the greatest social mobility machine in independent India's history. One generation - from clerk's salary to Bengaluru flat.

I wrote in The Education Emergency that Macaulay's system was designed to produce obedient clerks, not creative minds. The IT boom did not break that system. It rescued it. The mass-recruitment model meant the clerk-factory never had to reform, because its output - compliant, credentialed, trainable - was exactly what the services industry ordered. The colleges did not need to teach computer science. They needed to teach enough to pass the recruiter's aptitude test.

That is the racket: an education industry whose real product is placement optics, whose curriculum is an afterthought, and whose economics depend entirely on someone else's hiring spree.

And the hiring spree is ending.

The Rung That Vanished

The numbers from the past year tell a story most commentary has missed.

An EY analysis estimates that entry-level IT roles - the coding and testing jobs that were the ticket's redemption counter - have already shrunk by 20 to 25 percent under automation. The tools writing that code now are the same class of AI systems the colleges claim to be teaching.

Meanwhile, the Naukri JobSpeak index closed FY26 with white-collar hiring up 8 percent - the strongest in three years. Read the fine print. The growth was led by hospitality, BPO operations, oil and gas, and real estate - not IT services. AI and machine-learning roles grew 34 percent. And the sharpest number of all: fresher hiring in the ₹20-lakh-plus bracket grew 23 percent.

Put plainly: the ladder has not been removed. It has been pulled up.

Elite entry thrives. A fresher who can genuinely build - train a model, optimise inference, design a data pipeline - commands packages the racket's brochures can only photograph. Mass entry shrinks. The ordinary graduate with the ordinary certificate faces an entry rung that is 20 percent gone and going.

Employers have noticed, and adjusted faster than the academy. Recruiters increasingly screen for portfolios and demonstrable projects over degrees and certificates. The market has stopped honouring the ticket. The colleges are still printing it.

The New Brochure

Confronted with this, what has the education industry done? It has rebranded the ticket.

Every second institution now advertises "AI training". Look closely at the syllabus and you will find, in most cases, AI usage training - prompt writing, tool walkthroughs, certificate courses in operating chatbots.

Let me say this plainly, because nobody selling the courses will: AI usage is not a skill you need a semester to learn. It is barely a skill at all. Using a modern AI system is common sense, and it becomes more like ordinary human conversation with every model release. The entire, deliberate design goal of this technology is that it should require no training to use. Selling a six-month certificate in talking to a machine that is engineered to understand you is not education. It is the racket's newest brochure.

There are three layers here, and the racket profits from blurring them:

Using AI - common sense, self-taught in a weekend, improving as the machines improve.

Applying AI in a domain - real and valuable, but learned in the domain: the agronomist, the radiologist, the lawyer who folds AI into their actual work.

Building AI - the scarce skill. Model architecture, training infrastructure, chip design, power systems, data engineering. The nuts and bolts.

The racket sells the first, prices it like the second, and markets it with the vocabulary of the third. A robot dog, bought abroad and rebadged, is simply this practice performed on a national stage.

The Nuts and Bolts

Here is what building actually looks like, and - credit where due - the state has started buying it.

The IndiaAI Mission, funded at roughly $1.2 billion, is backing twelve organisations to build sovereign foundation models. In February 2026, Sarvam AI announced two open-source models trained entirely on Mission compute - a 30-billion-parameter and a 105-billion-parameter mixture-of-experts system, weights published under Apache 2.0. The IIT Bombay-led BharatGen consortium shipped a 17-billion-parameter multilingual model, backed by over ₹1,000 crore - the largest single allocation in the programme. The national common-compute pool has crossed 38,000 GPUs, with a stated target of one lakh by the end of 2026. Private infrastructure is following: gigawatt-scale AI data-centre projects are under construction, with tens of thousands of latest-generation GPUs being deployed in Greater Noida, Navi Mumbai and Chennai.

Be honest about the gap, too. Training a genuinely frontier model requires sustained compute at a scale India does not yet possess. Our best models are competent, not competitive. The distance is real.

But notice what closed the distance that has been closed: procurement, infrastructure, energy, silicon. Nuts and bolts. Not certificates.

Now ask the obvious question. In this entire national buildout - the models, the GPU clusters, the data centres, the power engineering - where is the university system? It should be the feeder: the faculty seconded to model teams, the campus compute contributed to the pool, the Indic-language datasets curated and published, the graduates who arrive knowing how a transformer is trained because they helped train one. A handful of IITs are in the arena. The other thousands of institutions are printing AI brochures.

A nation whose universities showcase rented robots while its startups train sovereign models has its institutions exactly backwards.

Force the Issue

The polite view is that the academy will adapt on its own. The evidence - 190 years of it, by my earlier count - says otherwise. The clerk-factory has survived every wave of reform by absorbing its vocabulary and changing nothing. It will do the same with AI unless it is forced.

So force it. The government funds, regulates, and accredits this industry. Use the leverage.

One. Pay for capability, not enrolment. Tie AICTE approval and public funding to measurable contribution: faculty on national model teams, campus compute contributed to the Mission pool, published datasets in Indian languages, working hardware labs. An institution that contributes nothing to national capability should find expansion approvals hard to come by.

Two. Count what matters. Placement percentages and global-ranking advertisements are marketing. Count models shipped, silicon taped out, patents actually worked, systems in production. Publish the league table. Let parents see which colleges build and which ones photograph.

Three. Route compute through universities that earn it. GPU-hours are the new laboratory grant. Allocate national compute to campuses the way beam-time is allocated at CERN - on merit, with published results.

Four. Teach the unglamorous. AI is electricity plus silicon plus cooling. India needs power engineers, thermal engineers, fabrication technicians, and data-centre architects in the tens of thousands. These courses do not photograph well at pavilions. They are sovereignty.

Five. Make fraud expensive. The summit's organisers did one thing right in February: the display came down, in public. Institutional dishonesty on national platforms should carry consequences that outlast the news cycle - in accreditation, in funding, in expansion rights.

The Churn

In The Education Emergency I wrote: we were not uncreative, we were interrupted. Macaulay interrupted us; the empire needed clerks and built a machine to produce them.

The uncomfortable churning of this episode is that the interruption is no longer someone else's doing. No coloniser ordered us to sell six-month certificates in conversation. No foreign power compelled a university to rebadge an imported robot on state television. The clerk-factory found a new empire to serve - the placement brochure - and it did so entirely of its own volition, for a quick buck, with no national pride to restrain it.

The nectar in this churn is real, and it is new: for the first time, India is spending serious money on the nuts and bolts - models, compute, energy, silicon. The poison is also real: an education industry positioned to divert that moment into its oldest trade, selling tickets to a ladder it did not build and cannot see has been pulled up.

Somewhere in a tier-3 college tonight is a student with Ramanujan's hunger and a brochure's education. She deserves institutions that teach her to build the robot.

Not rent one, paint a name on it, and call it India's.

Previous Episode: Episode 13: Old Before Rich?

Next Episode: Episode 15: The Unbuilt Engine

Series Home: Bharath Manthan - Churning the Indian Pot

The author is Founder & Editor-in-Chief of BarathVector.