AI IS REWRITING OUTSOURCING ECONOMICS

AI is dismantling outsourcing by attacking the one thing the industry was built on: wage arbitrage. The entire model rested on paying a fraction of the onshore rate for the same routine work done somewhere cheaper. Generative AI drives the marginal cost of that work toward zero, which makes the wage gap meaningless. When the task itself is nearly free, it no longer matters where the cheap labor lives.

Why was outsourcing only ever a wage-arbitrage trade?

Strip away the language about "global delivery models" and outsourcing is one arbitrage: buy labor where it is cheap, sell the output where it is expensive, keep the spread.

That spread was enormous and durable. A process that cost an American firm $60 an hour onshore could be staffed offshore for a fraction of it, and the quality gap narrowed every year. The business-process and IT services giants scaled that single trade into hundreds of billions of dollars of revenue.

The trade had one structural assumption: the work still required a human. The whole edifice — the campuses, the training pipelines, the time-zone handoffs — existed to supply human labor more cheaply. Remove the human and you remove the trade, which is precisely what is now happening. This sits at the center of our economics coverage because it is a clean case of a business model losing its underlying input.

How big is the exposed layer?

The exposure is concentrated and large. India's IT and business-process sector employs roughly 5.5 million people and exports about $200 billion a year, according to NASSCOM, the industry's own trade body. A significant share of that is routine, rules-based, English-language work.

That description is also a near-perfect specification of what generative AI automates first. Ticket triage, basic code maintenance, data entry, document processing, first-line support, and template-driven content are the commodity tier of outsourcing — and the easiest tier for a model to absorb.

The point is not that 5.5 million jobs vanish at once. It is that the fastest-growing, highest-margin layer of the industry is the layer most directly in the path of automation that operates on language and logic.

Why doesn't this just send the work back onshore?

The reflexive assumption is reshoring: if offshore labor loses its edge, the work comes home. That misreads the mechanism.

Reshoring is a story about where a task is done. It assumes a human still performs it and only the location changes. AI is a story about whether a task is done by a human at all. When a model resolves the support ticket or writes the boilerplate code, there is no task left to relocate.

This is the claim you will not find on the first page of search results, which is crowded with "reshoring vs. offshoring" framing. The real shift is not geographic. The routine work is not moving — it is being deleted. What returns onshore is a thin band of oversight, not the volume that was sent away.

What happens to the vendor's business model?

The billing model is breaking before the headcount does. Outsourcing contracts were priced in full-time equivalents — a polite proxy for "how many people we are renting you." Revenue scaled with bodies.

AI severs revenue from bodies. A vendor that automates a process still wants to be paid for the outcome, so the industry is repricing toward per-transaction, per-resolved-ticket, and per-token models. That defends volume, but it compresses the margin that came from marking up cheap labor. You cannot mark up labor you no longer employ.

The firms that move first turn this into a durable competitive advantage by owning the automation layer and the client relationship. The firms that defend the old headcount model are defending a melting asset, much as incumbents do when platform dynamics shift the basis of competition beneath them.

Which outsourced work actually survives?

The work that survives is the work a human must own. Strip the function down to its essence and ask: can this be written as a clear input-output rule?

If yes, it is exposed. The more cleanly a task can be specified, the more cheaply a model performs it. If no — if the task depends on judgment, liability, trust, or the integration of messy systems that resist standardization — it is defensible.

Regulatory accountability does not delegate to a model. Neither does a high-stakes client relationship or the ownership of a result when something goes wrong. The durable outsourcing business is shrinking toward that core, which connects directly to the broader restructuring of knowledge work playing out across every white-collar function, a theme we trace in our analysis of AI and infrastructure economics.

What should operators and investors watch?

Watch the pricing disclosures, not the press releases. The signal is the share of revenue a vendor reports under outcome-based or consumption-based contracts versus headcount-based ones. A rising outcome share confirms the labor-to-software transition is real; a flat one suggests the firm is protecting an old model.

Watch headcount-to-revenue ratios. A vendor growing revenue while flattening or cutting headcount is monetizing automation. A vendor still adding bodies to add revenue is running the legacy arbitrage on borrowed time.

And watch which firms move up the value chain. The outsourcers that reposition around judgment, integration, and accountability inherit the defensible work. The ones clinging to commodity volume inherit the decline. This is the same competitive sorting we examined in the shift in how digital services create value — capability migrates, and the firms that read the migration early keep the margin.