Here is a number that should worry every investor and policymaker in India more than it currently does: 17 lakh. This is roughly how much capital has been invested for every direct and indirect job generated under India’s Production-Linked Incentive scheme so far. Now, the PLI is a strategic scheme whose primary objective is to revive manufacturing in the country and not jobs, but it is important to understand that investment in manufacturing does not automatically create jobs. As a matter of fact, the higher the capital investment, the fewer jobs are created. This decoupling of business and jobs is happening with AI now across service jobs in the country. This demands a rethink of the country's job engine. AI will bring a similar shift to the services economy, which has historically been India’s more reliable engine of white-collar employment.

Ironically and not coincidentally, both these trends are being driven by the same technologies. Shop-floor robots and office AI are increasingly converging technologically. Nvidia's "physical AI" and its Isaac GR00T humanoid foundation models use the same reasoning architecture that powers enterprise AI agents; SAP is integrating its office-facing Joule agent platform directly into industrial and domestic robots built by NEURA Robotics. The distinction between "services jobs are safe because they're not physical" and "manufacturing jobs are exposed because they're physical" is becoming harder to sustain.

Back to the numbers, as of March 2026, India's Production-Linked Incentive scheme — the country's flagship manufacturing job-creation program — had drawn actual investment of over ₹2.40 lakh crore ($25.55 billion) across 14 sectors and generated more than 14.15 lakh direct and indirect jobs, per the government's own disclosure to the Lok Sabha. Divide one by the other, and roughly ₹17 lakh capital has been spent— about $20,000 — for every direct and indirect job created. This is for a blue-collar job that, on average, may not pay more than Rs 4 lakhs per annum. Thus, it is not a very efficient scheme if viewed only from a job perspective.

That ratio is not uniform across industries. In large-scale electronics — PLI's flagship success, credited with a 2.4x rise in domestic mobile phone production and a 77% fall in mobile imports — capital intensity per job is far higher than in food processing, where a much smaller ₹9,207 crore in investment generated 3.35 lakh jobs, beating its own target. Same government program, same incentive structure, wildly different capital-to-job ratios — depending entirely on how automatable the underlying sector is.

That distinction matters because India is entering one of its biggest capex cycles in decades. Investors tend to measure factories by capacity, revenue, margins and return on capital. Policymakers also need to ask another question: how much employment and income does each incremental rupee of capital create?

This is not an isolated Indian anomaly. It is the local instance of a pattern that has been building in the world's most capital-intensive economy for two decades, and is now appearing in India's own data and in the investment theses that VCs and family offices are writing right now. The relationship between capital investment and job creation — the assumption embedded in every industrial policy document, every PLI scheme design, every "jobs will follow growth" argument — has broken. Understanding exactly how and where is now a due diligence question for policymakers looking at deploying India's demographic dividend.

The relationship first collapsed in US

Between 2002 and 2022, the number of American manufacturing firms fell by 21%, and manufacturing employment fell by roughly 2.4 million jobs — a 17% decline — even as the broader US economy grew over 14% in headcount terms, per Census Bureau data analyzed by the Information Technology and Innovation Foundation. Only three sub-sectors — food and beverages, tobacco, and chemicals — added net jobs over those twenty years. Reporting in 2025 noted that manufacturing employment kept falling even as "billions of dollars pour into new plants and reshoring initiatives nationwide" — the Reindustrialize America capital story and the jobs data are now moving in opposite directions, because the new plants are automated by design.

The mechanism has a coefficient attached to it, and it is the single most rigorous number in this entire argument: Daron Acemoglu and Pascual Restrepo's "Robots and Jobs: Evidence from US Labor Markets" (Journal of Political Economy, 2020) found that one additional industrial robot per thousand workers in a US local labour market reduces the employment-to-population ratio by roughly 0.2 percentage points and wages by about 0.4 per cent. A robot effect statistically distinct from the damage done by Chinese import competition. It is some of the strongest empirical evidence that automation can raise productive capacity without generating a commensurate increase in employment.

India's own numbers confirm this is not an imported anxiety. NSSO data show that manufacturing employment actually fell by roughly 3 million between 2004–05 and 2009–10, at the tail end of what economists were already calling "near-jobless growth." Manufacturing's share of GDP has been stuck at 14–17% since 1991; its share of employment has stagnated at around 12%.

The capital side has moved independently. A 2025 Observer Research Foundation analysis found India's capital stock grew 74% over the last decade while employment grew only 36% — capital compounding at roughly double the pace of jobs. A survey of Industry analysis found that the share of workers' wages in organized manufacturing gross value added fell from 22.2% in 2000–01 to 14.3% in 2011–12. The labour's share nearly halved even as output rose. More recent cross-state ASI data (2023–24) confirms that the more productive a state's manufacturing sector, the smaller the share of output that flows to workers as wages. Capital improves productivity through automation and reduces not just labour deployment but also wages. Fewer jobs at lower salaries.

The PLI arithmetic above is illustrative rather than causal. But the lesson is important: the sector matters as much as the manufacturing label. Capital deployed into labour-intensive categories can still create significant employment. Capital deployed into highly automated categories may generate much more output than labour demand.

The robot data explains why

The International Federation of Robotics' World Robotics dataset shows global annual robot installations more than doubling over the last decade — from roughly 254,000 units in the mid-2010s to 542,000 in 2024, the second-highest year on record, after 2022's peak of 553,052. Installations have now exceeded 500,000 units per year for four consecutive years — a historically unprecedented continuous hike. Global operational stock reached 4.66 million robots in 2024, up 9% year-on-year, with Asia now accounting for 74% of new deployments, up from 67% in 2019.

China alone installed 295,000 robots in 2024 — 54% of the global total — and its operational stock surpassed 2 million units, roughly 4.5x Japan's, the world's second-largest stock. For the first time in 2024, domestic Chinese robot manufacturers outsold foreign suppliers in China. Japan, the US, South Korea and Germany — the rest of the "big five" — have been essentially flat to declining since 2019, in sharp contrast.

India is the fastest-growing major market off a low base: 5,353 robots installed in 2022, 8,510 in 2023 (+59%, moving India to 7th globally), and 9,100 in 2024 (+7%, moving India to 6th, overtaking Germany).

Operational stock nearly doubled since 2018, reaching 44,958 units by 2023 — still 32 times smaller than China's 2023 single-year installation count. Automotive alone accounted for 45% of India's 2024 installations. The IFR's own 2024 report explicitly forecasts a contraction in Indian installations in 2026, tied directly to the PLI scheme's scheduled wind-down — a rare instance of an automation-adoption dataset naming a specific Indian industrial-policy program as its demand driver, which should tell every investor tracking India's capex cycle exactly how tightly these two curves are now linked.

Dark factories are reaching every sector

Full automation — "lights-out manufacturing," running with no human presence on the floor — began in semiconductors, not because robots were cheaper than labour but because human presence itself is a contamination risk in a cleanroom. It has since spread to electronics and precision components. It is now reaching garments, a manufacturing task that resisted automation for two centuries because soft, deformable fabric does not behave like sheet metal. Advances in computer vision, robotics and AI are beginning to change that.

We are now entering the era of Physical AI—a paradigm that combines perception, reasoning, and physical control into a single integrated system. This differs fundamentally from classical automation, which operates in closed, pre-defined environments following predetermined trajectories. Physical AI must operate in open-ended environments, interacting with unknown objects, adapting to unforeseen situations, and achieving long-horizon objectives

Atlanta-based Softwear Automation's "Sewbot" technology — machine-vision-guided robotic sewing that can build a finished T-shirt in about four minutes — is not a lab demo. Tianyuan Garments, Adidas's largest apparel supplier, installed 21 Sewbot lines in Little Rock, Arkansas in 2017–18, targeting 800,000 shirts a day at roughly 33 cents of labour cost per unit. Tianyuan's own chairman told China Daily the plant was built explicitly to out-compete Bangladesh and Cambodia on cost — from an Arkansas address. Last year, the company raised strategic funding from Bestseller, affirming that its technology has reached a critical stage.

For Asian workforces, the implications extend beyond factory productivity.

Bangladesh, Vietnam and India together employ millions of people in garment manufacturing — the sector is 72% of manufacturing employment in Cambodia, 55% in Bangladesh, 42% in Vietnam — yet none of these countries built many globally distributed consumer apparel brands during the decades they dominated production. Europe and the US kept the brand, the design IP, and the margin; Asia kept the factory floor. If automated garment production scales, Asian manufacturers face a double exposure: losing the jobs while never having captured the brand value that would have cushioned the loss. Over 20,000 Bangladeshi garment workers were reported to have lost jobs to retrenchment or factory closures in the first half of 2026, according to the Bangladesh Human Rights Centre, amid a broader reshuffling of global apparel sourcing. Automation is not the only force at work, but it is steadily weakening the technological barrier that once protected labour-intensive manufacturing.

AI could bring the job threat equation to the office

Robotic Process Automation compressed the low-skill end of India's BPO industry over the 2015–2025 decade — data entry, invoice matching, tier-1 customer service. The office-floor equivalent of Acemoglu-Restrepo now exists, too: Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen's Stanford Digital Economy Lab study, using ADP payroll microdata across millions of US workers, found a roughly 13% relative decline in employment among workers aged 22–25 in the most AI-exposed occupations. LinkedIn data show US entry-level hiring down 23% versus pre-pandemic levels, steeper than the 18% decline in overall hiring.

Some researchers cite a slowdown in entry-level hiring even at Indian IT companies, but there is not yet enough evidence to conclude that AI is causing net employment destruction in India's services sector. Hiring cycles, global technology spending and economic conditions are also influencing recruitment.

That distinction matters. Manufacturing gives us evidence of what sustained automation can do to labour intensity. In Indian services, AI-driven decoupling remains a risk and signs of it are appearing, though not in data.

Challenger, Gray & Christmas (which tracks employer-announced layoffs monthly) shows AI has been the single leading stated reason for US job cuts for most of 2026: cited in 8% of cuts in February, 25% in March, roughly 40% of May's cuts, and holding the top spot for five consecutive months through July. For the full year 2026, AI-attributed cuts had already surpassed all of 2025's total by May. Worth being precise about what this measures, though: it's employer-stated reasons on layoff announcement filings, not a verified causal test. But taken alongside falling entry-level hiring and rapidly improving AI capabilities, it is enough to question an assumption that has underpinned India's economic model for decades: that services will continue absorbing educated workers even as manufacturing becomes more automated.

How should policymakers create jobs

The argument is not for protecting the old jobs — that fight is being lost in garment cut-make-trim and tier-1 services work as we speak. A subsidy or an incentive cannot reverse a cost curve; it can only temporarily pause it. The need is for a policy discipline that investors and entrepreneurs can apply directly. The need is to be circumspect about using incentives for employment-generating sectors like food processing, restaurants and tourism.

Capital aimed at AI-enabled MSMEs and one-person or ten-person companies — where AI collapses the cost of sales, accounting, design, compliance and customer service — behaves like the food-processing case. This is where new entrepreneurs can create value and jobs, incentives should be targeted here.

It is also where India has run this multiplier before: NASSCOM-era IT income gains built a new middle class whose spending, in turn, built retail investment, EMI-financed consumer durables, and installment-financed housing. Reproducing that effect deliberately, rather than leaving it to chance, means favouring Indian AI infrastructure — affordable compute, open models, Indian-language datasets — built for Indian costs and Indian scale, over imported automation defaults tuned for high-wage economies. It means treating India's AI ecosystem more like Ukraine's distributed drone-innovation model — thousands of small teams iterating in parallel — than a strategy built around two or three national champions.

It also means watching concentration risk directly: AI investment that ends with most of the gains captured by a few large incumbents, Indian or global, is closer to the electronics case than the food-processing one. And it means measuring what the money actually buys at the district level — new businesses, new jobs, rising incomes, survival rates — rather than GDP growth contribution alone, since GDP is exactly the metric that has kept rising even as the capital-jobs relationship quietly broke over two decades.

The next phase of India's growth story will therefore require a different question. Not simply: how much capital did we attract?

But: what did that capital ultimately create?

Harish Mehta is co-founder of Nasscom and a founder member of the Hundred Million Jobs Mission (HMJM), and K Yatish Rajawat is the founder of Center for Innovation in Public Policy (CIPP) and a founder member of the Hundred Million Jobs Mission. www.hundredmillionjobs.org