The AI factory: the rewiring of India’s tech industry

Thirty years of growth in outsourced tech services has transformed India’s economy. However, AI’s expansion is forcing the tech sector to evolve. The FT’s Krishn Kaushik investigates whether data annotation to train AI models and humanoid robots is enough for India to thrive in the AI era.

— Channel description, FT Film (Financial Times)

A ~20.5-minute Financial Times documentary (FT Film, YouTube, 25 June 2026), reported by Krishn Kaushik (FT’s Mumbai correspondent, ~15 years reporting on India). The film interleaves voices from textile-factory floors in Karur, Tamil Nadu; data-annotation entrepreneurs; rural-BPO operators (NextWealth); a robotics-data startup (Object Ways); a multinational Global Capability Center (Tesco Business Solutions, Bangalore); and AI-political-economy critics. The reporting question: as AI threatens the IT-services outsourcing model that built modern urban India, is the new data-annotation-and-robot-training economy a genuine step up the value chain — or a re-run of the extractive back-office role under a new name?

Source quality note: auto-generated (ASR) English captions, cleaned at acquire time (proper-noun and mistranscription fixes logged in the raw file’s notes:). Speaker attribution is partial — the film does not label most interviewees in the caption track, so beyond Kaushik and the named companies, individual quotes are left unattributed.

TL;DR

• The thesis: India is becoming “the AI factory of the world” — human-in-the-loop labor at population scale. “The booming AI industry needs more and more humans in the loop.” India is “probably the second largest AI workforce in the world today,” supplying the data-annotation, video/audio annotation, multi-turn-conversation, and RLHF labor that trains models built elsewhere. “Almost all the products that are there globally, at some point, will be trained by somebody in India.”
• The new frontier: egocentric data for humanoid robots. In a Karur textile factory, ~60 of 200 workers wear GoPros and Meta glasses 6–8 hours/day (for an extra ~10,000 rupees/month) to capture egocentric data — folding clothes, sweeping, washing dishes — to train robots. Object Ways (founded 2019 as a data-annotation company, now robotics) collects “gripper data, teleoperations data” and claims expectations of “500 million hours [of data] in a single day.” A worker’s framing: “for a robot to know that this is a cup, San Francisco offices don’t train the robot to do that — it’s people sitting in small-town India.”
• The displacement threat to IT services is existential. India’s IT/IT-enabled services export ~$330–340bn — the country’s largest single export and a primary source of dollar income, built over 25+ years on “under-promise, over-deliver” at “the right price point” (“if you are paying $1… I can get it done for 20 cents”). AI is “the toughest challenge that India’s IT services sector has faced,” threatening high-volume, error-tolerant work that is “a prime candidate for AI substitutability.” The sector “were not the vanguard of bringing AI to India… they were not investing in AI.”
• The political-economy critique: extractive supply chain, concentrated power. “AI is fundamentally a marketing term… a way to unlock vast amounts of capital.” “AI should be understood as a technology that by design concentrates power.” The data-work is “a supply chain which is extractive by design”; workers may be “asked to wear these cameras… with no extra money,” and “all this data can be used against the workers.” The recurring question: “are we data-working our ways out of employment… into a world where the [technology] companies run away with the lion’s share of the value?”
• The development/gender upside (the steel-man). Rural data-annotation work (NextWealth’s model: “take the work to where the people are, rather than people traveling to where the work is”) reaches first-generation graduates in small towns — 60% of India’s engineering colleges and graduates are in small towns — and gives women jobs they can do without leaving their hometowns, in a socially conservative context. “If you educate a girl, you educate the entire village.” Multiple women describe economic independence and funding their children’s education.
• The sovereignty worry. “Offering up the scale of our population as a carrot to attract foreign tech companies is not a pathway to anything resembling sovereignty or resilience longer-term.” India lacks the chip/hardware base; “$25bn worth of investment has left the country” in early 2026 toward Taiwan/South Korea. “India will become the use-case capital of the world… but not at the cost of sovereignty.” Echoes of the BPO era and “the backroom.”
• The counter-frame (the optimists). A lump-of-labor-style rebuttal appears too: “while certain kinds of tech jobs may go out, there will be certain other newer kinds of tech jobs which will get created”; “the threat to humanity is not AI — [it’s] the inability to learn”; “a distinct possibility that the overall number of jobs in the tech sector in India will actually go up.” And the binding constraint stated plainly: “India has talent at very large scale. But if you don’t have enough jobs for these people, what is the value of the scale? It’s nothing.”

Why this matters for the wiki

This is the wiki’s first field-journalism, Global-South-vantage source on AI’s labor economy. It anchors three threads the corpus has otherwise carried mostly from US/European empirical and operator vantages:

1. The human-in-the-loop training-labor economy. The corpus treats RLHF and data annotation as a technical pipeline step inside agent development and generative AI; this source surfaces the labor behind it — who does it, where, for what pay, under what framing. It is the wiki’s clearest material instance of the “AI factory” / ghost-work layer. (A candidate future concept page — data-labor / human-in-the-loop economy — is flagged below; held as single-source for now.)
2. AI employment effects, offshoring vantage. It corroborates the ai-employment-effects displacement thread from the receiving end of the offshore-outsourcing chain — the IT-services sector that the wiki’s empirical sources (Brynjolfsson Canaries, AI Index 2026) measure in US payroll data is the same work that India built an export economy around. See the dedicated section added there.
3. Responsible AI, as political economy. Extraction, power concentration, data-sovereignty, and consent-to-surveillance (“data used against workers”) are RAI concerns the wiki has under-attended at the labor-and-geopolitics layer. See the section added there.

Related sources

• 2026-05-19-palicha-zepto-stanford-or-startup-india-quick-commerce — the optimistic India-AI-economy counterpart. Where Zepto narrates India building a frontier AI-native operation (in-house ML, SaaS spend cut to zero), this film narrates India supplying the labor that trains everyone else’s models. Holding both is the wiki’s two-sided read on India’s place in the AI economy.
• 2026-06-25-guilbeault-stanford-gsb-what-ai-cant-do-and-why — the AI-hype critique at the cognitive-science level. Guilbeault’s “grandiose narrative” objection is the intellectual sibling of this film’s “AI is fundamentally a marketing term… a way to unlock vast amounts of capital” and “concentrates power by design” — the film supplies the material/labor instantiation of the same critique.

Linked entities and concepts

• Concepts: ai-employment-effects (IT-services displacement; the data-annotation/robot-training labor economy), responsible-ai (extractive supply chain, power concentration, data sovereignty, surveillance-consent), automation-vs-augmentation (the “60% automated… I need only 40 people” line), generative-ai and agent-development-lifecycle (the RLHF / data-annotation pipeline this labor feeds), strategy / dynamic-capabilities (India’s sector-level value-chain repositioning).
• Candidate concept (dangling, single-source): a human-in-the-loop data-labor economy / “AI factory” / ghost-work concept. Deferred until a second source substantiates it (e.g. a future ingest on Scale AI / Surge / Sama / Mechanical-Turk-style annotation labor). Flagged in the log for promotion.
• Dangling entities (single-source mention, deferred per author/second-source rule): Krishn Kaushik (FT Mumbai correspondent / reporter), Object Ways (data-annotation→robotics startup, founded 2019), NextWealth (rural-BPO / impact-sourcing data-annotation operator), Tesco Business Solutions (Tesco’s Bangalore Global Capability Center). No entity pages created this ingest.

Dynamic-capabilities mapping

• contextual/external-triggers — AI is the disruptive digital technology reshaping India’s IT-services sector and labor market: “AI is probably the toughest challenge that India’s IT services sector has faced in their existence… absolutely threatening the dominance that they had for the last 25 years.” The film also names the capital-flight trigger (“$25bn… has left the country” toward chip-producing economies) and shifting global demand for hardware India lacks.
• strategic-renewal/business-model — the sector-level question of refreshing India’s tech value-creation/value-capture logic: moving from back-office IT services toward “build our own large language models… our own products,” the “AI factory of the world” / “use-case capital of the world” positioning, and the in-house Global Capability Center model (Tesco doing “100% of all architectural design… right here in Bangalore,” AI cost-intelligence and personalization) that replaces the outsourced-vendor logic with captive in-house capability. The film’s skeptics frame this renewal as incomplete — “India has a lot of ambitions for AI but… hasn’t put in the effort to really take the lead.”

(Role-relevance inherits the cell defaults: cso, cdo for external-triggers; ceo, cso, cmo for business-model renewal.)