India's Unbilled Future

For three decades, foreign clients paid India’s graduates to become professionals. AI promises to do some of their work more cheaply, leaving an awkward question about who will pay for them to learn the rest.

In September, Wipro’s chief technology officer, Sandhya Arun, told Reuters that AI had freed capacity equivalent to the output of 20,000 employees, permitting the redeployment of staff within the company. For anyone worried about mass redundancies, this was reassuring news. A graduate waiting to join the industry, however, might have heard something rather less encouraging.

Much of the argument about AI and employment concerns people already at work, whose occupations can be inspected for tasks a machine might perform. The person who has yet to be hired is harder to count. There is no dismissal to announce when a company discovers it can manage without its next intake, and the decision to hire fewer graduates may barely register amid good news about productivity and profits. India has particular reason to worry about that combination because its technology industry built itself on taking in people who did not yet know enough.

Since the 1990s, Indian IT firms have assembled large teams of engineers and sold their time to foreign clients who would have paid considerably more for comparable work at home. And what began with a difference in wages later grew into a formidable concentration of experience. Nasscom’s February forecast put the wider technology sector at $315 billion in annual revenue and almost six million employees. Behind those totals lay a familiar family formula for success whereby engineering college might lead to a campus interview, and an offer from a large employer could turn years of educational expense into a regular salary. The first job mattered partly because it offered a way to become qualified for the second.

Beginners entered an industry that had uses for them. Routine testing and maintenance could be assigned to junior staff under supervision, allowing firms to charge for work while the people doing it became more capable. There was nothing charitable about the arrangement. Clients wanted a service at an acceptable price, and Indian providers needed a large workforce they could afford to employ. Between them they financed an apprenticeship on a scale neither had set out to provide. Essentially, much of the time logged on a foreign invoice were hours in which an Indian graduate was learning how to be useful.

AI arrives in this arrangement as something almost everyone has a reason to welcome. An engineer can get through a task faster, a manager can promise an earlier delivery, and a client can ask why a project still requires so many people. Even the geopolitics need not get in the way. Microsoft offers DeepSeek models through its cloud, so an Indian business can buy Chinese-developed intelligence on an American invoice. For a company already paying Microsoft, trying the software may amount to little more than adding a service to an existing account.

The resulting productivity gain, however, belongs to no country in particular. A capable model can help an Indian team, but it can also assist a smaller competitor elsewhere or the client’s own technology department. India’s engineers may become better at their work even while the commercial advantage of employing so many of them diminishes. And once a client knows that an assignment takes fewer hours, the old bill becomes harder to defend. Charging for outcomes offers providers a way forward if they can deliver something their rivals cannot readily reproduce. Where competitors have access to much the same tools, however, they too can offer an outcome at a lower price.

None of this makes the large Indian firms dispensable. Years spent inside a client’s systems are worth more than the ability to generate a piece of code, particularly when those systems belong to a bank that cannot afford a failed migration. Established providers understand how the parts fit together and can be held responsible when something goes wrong. Such obligations should preserve plenty of valuable work. The difficulty is that the people best equipped to undertake it generally acquired their experience while doing work of a less exalted kind.

This is where Wipro’s announcement becomes interesting. Its estimate does not tell us that AI has eliminated 20,000 jobs, nor does redeployment tell us how many graduates it will recruit. It does describe a company finding additional labor inside its existing payroll. If other firms do the same, the industry could absorb more business for a time without expanding recruitment as much as it once would have. The employees who already know the clients would become busier and more productive while opportunities for those outside grew scarcer.

For a manager with a project to deliver, using an experienced employee is an understandable choice. Recruiting a beginner entails supervision and mistakes, followed by the possibility that the employee will leave just as the investment starts to pay off. Outsourcing made that risk easier to bear because junior work could earn money during the learning. If AI reduces what clients will pay for that work, the training bill becomes harder to recover. A firm can respond by hiring people whom somebody else has trained, which works very well until too many employers arrive at the same solution.

Graduates are advised to acquire greater judgment and move into more demanding work. They have good reason to try, though it is difficult to learn how to handle a failing project without ever being allowed near one. Practice with a model may improve technical ability, and better tools could let a recruit take on useful tasks sooner. Whether that recruit gets the chance still depends on an employer willing to put them on a live assignment, with a colleague responsible for supervision. No amount of exhortation to become more skilled settles who will provide the first opportunity.

It would be convenient to blame the entire predicament on a new technology, but campus hiring was already weakening in 2023 and 2024 as clients cut discretionary spending. In 2025, TCS announced plans to shed around 12,000 jobs. AI has entered an industry already reconsidering how many people it needs, which makes it harder to distinguish a temporary hiring slump from a change that will survive the return of demand. If companies discover during a downturn that they can deliver projects with fewer recruits, the next recovery may bring back the revenue without reopening all the places.

There are reasons it might work out differently. Cheaper software could attract customers who previously could not afford to commission it, giving Indian firms enough additional business to offset the reduction in work required for each project. Multinationals’ centers in India may also take on some of the training that outsourcing firms provide. Within an individual team, AI could reduce the burden of supervising beginners sufficiently to make hiring them attractive again. An employer that once needed months to make a graduate useful might find the cost of taking one on has fallen. The same technology that removes some junior tasks could help beginners attempt others.

But their effects will have to be sought in recruitment, as well as the work recruits actually receive. A new customer does not bring a fixed number of graduate places with them. The assignment might sustain several beginners under supervision, or go to an experienced engineer using better tools. Both arrangements could produce a satisfactory result and contribute to revenue growth, but only one would give those beginners the experience they need.

The sector’s aggregate figures cannot show which arrangement is prevailing. Nasscom’s February forecast included 135,000 net additional jobs in the 2025–26 financial year and put services firms’ AI revenue at $10 billion to $12 billion. Those are substantial numbers, but a net employment increase does not tell us how many newcomers entered or what happened to the balance between junior and experienced staff. India’s Economic Survey has itself observed that growth in high-end services produces less than proportionate employment. A successful industry can still become a less generous employer, particularly to people who need it to teach them something.

For families, the change would be felt well before it became a national diagnosis. The engineering degree would still cost money and the firms would still be there, announcing new contracts, but the passage from college into paid work would become less dependable. Vacancies asking for experience could coexist with graduates unable to obtain any. Several years later, employers might complain of a shortage of qualified candidates without recognizing their own recruitment decisions in its history.

Trying to keep foreign AI out would do little to help. Indian companies would lose access to tools their competitors could use, while a domestic model capable of doing the same work would create much the same difficulty. Nor can graduates sensibly be asked to preserve their employability by refusing technology their employers want them to master. India will need the productivity gains, as well as employers to continue taking responsibility for people whose contribution is initially smaller than the trouble of training them.

That responsibility used to fit reasonably well inside a commercial transaction. The client bought hours, the firm employed a graduate, and enough useful work got done for both to accept the arrangement. As fewer hours are needed, maintaining an intake may require a training expense that someone has to defend explicitly. All of which means that companies will have to decide how much of their saving they are prepared to spend on people they could, for the present, manage without. Foreign clients have no particular reason to volunteer for the bill. Unless employers or the state find a way to carry it, India’s graduates will be left paying for qualifications while waiting for the work that turns them into professionals.