The multiplicand

Thematic essay — week of June 22–28, 2026.

"Force multiplier" is the phrase everyone reaches for when they want to say AI is going to change everything. It is worth taking the metaphor literally, because the metaphor contains an argument most of its users skip. A multiplier does nothing on its own. It acts on a multiplicand. Two times nothing is nothing. Two times a small number is a slightly larger small number. Two times a very large base is the kind of change that reorganizes an economy. The force is the same in all three cases. What differs is what the force is applied to.

So the interesting question about AI and India is not the one the phrase invites — how powerful is the model, how fast is the curve. The model is getting more powerful and the curve is steep; both are true and both are mostly out of India's hands. The interesting question is the one the metaphor hides: what is the base the multiplier is being applied to, and is that base India's strength or its exposure? This week, read as a set of events about the multiplier, was unremarkable — a funding report, a regulatory draft, a leadership hire, a cost-discipline trend. Read as a set of events about the multiplicand, it was a fairly complete map of where the transformation is real, where it is promised, and where it cuts the wrong way.

The week, read as a multiplier

Strip the week to its AI-economic content and four things happened. Enterprises kept turning away from "tokenmaxxing" — the early-2026 posture of maximizing AI usage for its own right — toward token efficiency, rationing frontier-model spend and routing work to cheaper models. The Reserve Bank of India put AI and ML models inside a draft model-risk regime for regulated finance, with board accountability, risk-tiering, and kill-switch provisions. OpenAI named its first India Managing Director, an ex-Uber executive, with consumer growth as the first-named line of his mandate. And the half-year numbers landed: Indian tech funding rose 12% to $7.2 billion while the number of rounds fell 43%, capital concentrating hard onto a small AI core.

None of these is a capability event. Nothing in the week made a model smarter. What the week described instead was the machinery of applying a multiplier that already exists: cheaper to run, governed for safety, locally staffed, selectively funded. That is the tell of where we are in the cycle. The frontier is no longer the story for India because India does not set it. The story is the multiplicand — the base of work, of life, and of unit cost onto which a commoditizing multiplier is now being pressed. Take each in turn.

Multiplicand one: the desk

The most measured multiplier in India right now is the one applied to knowledge work, and the place to read it is the IT-services majors, because they instrument everything. In early June, Microsoft reported that Infosys, TCS, and Wipro had each put Microsoft 365 Copilot in front of more than 100,000 employees — a combined 300,000-plus seats in under six months, with reported monthly active usage between 86% and 95%. Wipro alone was generating roughly 7.5 million prompts a month. Strip the vendor framing and a structural fact survives: the three largest Indian services firms moved internal AI tooling from pilot to production headcount in two quarters, and people open it daily. That is a multiplier on output-per-worker, applied at a scale — well over a million employees across the cohort — that almost no other labor base on earth can match.

Then the same companies handed us the other half of the ledger. A Cognizant–Pearson workforce study reported that AI already performs 37% of entry-level tasks at Indian organizations, against a 33% global average; 18% of Indian HR leaders said AI now handles half or more of entry-level work. Read the 37% as a perception figure, not an audited count — it is what 750 managers report, and it carries the optimism of the people who bought the tooling. But the comparison is defensible and it is the point: India reports higher entry-level automation than the global pool, because India's labor base sits disproportionately in exactly the standardized, process-bound entry roles current AI reaches first. The multiplicand here is enormous, and it is concentrated at precisely the rung that is most multipliable.

That is the augmentation–displacement question, and it does not resolve, because both motions are happening to the same cohort at once. Per-person output rises — the Copilot numbers — while the number of persons the base of the pyramid needs may fall — the 37% figure. The sharpest illustration is inside a single company: Cognizant's own CEO spent early June publicly committing to hire more than 20,000 entry-level graduates and dismissing AI-usage metrics as a "vanity metric," in the same month his firm's research put a number on how much entry-level work AI is said to be doing. That contradiction is not confusion. It is the honest state of the question. The firm is hiring the cohort and measuring its automation simultaneously, because it does not yet know which curve wins.

What the multiplier does to the desk is therefore real and unsettled in equal measure. The capital markets have already picked a side: Cognition raised $1 billion at a $26 billion valuation on the thesis that an autonomous coding agent is a substitute for the offshore billable hour, and this week Vishal Sikka, who once ran Infosys, came out of stealth with Hang Ten Systems and a $32 million seed built to collapse the "build, customize, integrate, maintain" cost structure the Indian SIs bill on — selling against the model he used to run. The incumbents are not standing still; LTIMindtree's BlueVerse Currency reprices services from effort to outcomes, the incumbent answer to the same cost collapse. But notice what is contested and what is not. Nobody disputes that AI is a force multiplier on knowledge work. The dispute is entirely about the multiplicand: who captures the multiplied output — the worker, the firm, or the client — and what happens to the entry rung that has been India's escalator into the middle class for two decades. The multiplier is settled. The base is the fight.

Multiplicand two: the last mile

If the desk is where the multiplier is most measured, the last mile is where it is most India, and it is the part of the story the displacement anxiety tends to crowd out. Here the multiplier multiplies reach rather than replacement — and here India holds a multiplicand almost no other country has.

Three things make that base unique. First, linguistic surface: a billion people across 22 official languages, most of whom were never reachable by English-first software at all. Second, digital public infrastructure: UPI, Aadhaar, and the identity-and-payments rails that already put a billion people inside a single addressable digital system. Third, a frontline-delivery layer — ASHA health workers, Anganwadi workers, agricultural extension agents, telephony-based customer service — that runs on human intermediaries doing structured, language-bound, repetitive cognitive work. None of these is the multiplier. All three are multiplicand, and they are the part of India's base that is genuinely scarce globally.

The week's longer arc is full of the multiplier being pressed onto exactly this surface. BHASHINI, India's public language-AI stack, signed an MoU to build a voice-first platform — and is being pitched outward, to Nepal, as exportable digital infrastructure. Gnani.ai shipped Prisma v2.5, an Indic speech-to-text model built specifically for telephony, which is the channel through which most of non-metro India actually transacts with institutions. Sarvam, now a unicorn on HCLTech's strategic cheque, builds the Indic foundation layer beneath all of this. In agriculture, the Rajasthan Agriculture Department signed a three-year MoU with Wadhwani AI for state-level agricultural AI — advisories to farmers in their own language, on their own crops, through workers they already trust. And Pine Labs launched P3P, an agentic payment protocol riding on UPI, the first serious attempt to let an AI agent initiate a real transaction on India's public rails.

This is a different multiplication than the one happening at the desk, and the difference matters morally and economically. When AI translates a government scheme into spoken Marathi for a first-time digital user, or gives an ASHA worker a screening prompt in her own language, or hands a farmer a pest-management advisory grounded in his district's conditions, it is not substituting for a worker who was already there. For most of this surface, there was no service to displace. The multiplicand is unmet need, and a multiplier applied to unmet need is pure expansion of capability. This is the strongest, most defensible version of the force-multiplier thesis, and it is the one most specific to India, because the base — a billion people, 22 languages, a public-rails delivery system — is something the US and China largely do not have in this form. It is also the version that gets the least airtime, because it does not come with a valuation or a layoff.

The honesty tax on this section is real, though. Most of these are MoUs, pilots, and v2.5 releases, not audited outcomes at population scale. The reach is potential until a deployment publishes how many farmers actually changed a decision, how many screenings actually happened, how many transactions an agent actually completed without a liability dispute — the unresolved RBI and NPCI questions on consent and liability when an agent initiates a payment are the reason P3P is a protocol launch and not yet a payments shift. The multiplicand is genuinely India's to multiply. Whether the multiplication has happened is still, mostly, a forward claim. But the base is real, it is large, and it is the part of the Indian AI story where the transformation, when it lands, lands on lives rather than on margins.

Multiplicand three: the unit cost

A multiplier you cannot afford to apply is not a multiplier. The third base — and the one that quietly determines whether the first two happen at all — is unit cost, and this is where the week's "tokenmaxxing"-to-efficiency turn connects to everything else.

India has never had a tokenmaxxing phase, because India deploys AI against ARPUs and public-sector budgets that never supported it. Cost-per-useful-output has always been the binding metric here, not raw capability. So the global correction that the week documented — enterprises rationing token spend, routing to cheaper models, demanding ROI — is not a correction in India. It is the operating condition. And the supply side has been moving to meet it. DeepSeek made a 75% discount permanent at $0.435 per million input tokens. Moonshot's Kimi K2.7-Code shipped agentic coding at a twelfth of frontier-API output pricing. Avataar priced a distilled video model at ₹0.48 per second. The effective cost of useful AI is falling faster than the sticker price of any single model, because an optimization layer — model routing, distillation, open weights — is doing work the price cuts alone did not.

The arithmetic that follows is the whole essay in one line. The multiplier is getting cheap exactly as India's multiplicand is largest. A force multiplier whose unit cost is collapsing, applied to the planet's biggest base of routine cognitive work and unmet frontline need, is not a marginal change. It is the mechanism by which the first two multiplicands actually get multiplied, because cost is the gate. A model at frontier prices reaches the enterprise desk and stops; the same capability at a twelfth of the price reaches the telephony line, the extension agent, the ₹99-a-month consumer. The efficiency turn that looks, globally, like a margin problem for the labs looks, from India, like the thing that finally lets the multiplier touch the base that was always too price-sensitive to reach.

There is even a domestic opportunity hiding inside the global pain. Roughly 95% of enterprise AI usage still runs on frontier models; the routing-and-optimization layer that squeezes frontier-quality output from right-sized models is barely built. That layer — cost-engineering intelligence to fit a budget — is precisely the kind of work Indian engineering has done well for thirty years. The efficiency shift is not just a tailwind for Indian deployment economics. It is a product surface Indian firms are unusually positioned to own, if they treat token-efficiency as a capability to sell rather than an advantage to leave on the table.

What a multiplier also multiplies

Here is the discipline the force-multiplier phrase is built to avoid. A multiplier is sign-neutral. It does not know what it is multiplying. Apply it to a strength and the strength compounds; apply it to a gap and the gap compounds just as fast. Every optimistic section above has a shadow that runs on the identical arithmetic.

The most important shadow is control. On June 12, the US Commerce Department revoked, overnight, a model capability that paying Indian customers had been using for three days — the Fable 5 export-control shutoff, a hard global off-switch on a frontier model in India's second-largest market, thrown from outside India. The force-multiplier story assumes the multiplier is yours to apply. The off-switch was the week of June reminding everyone that, at the frontier layer, it is not. A base multiplied by a capability you do not control is a base whose multiplier can be set to zero by a government you do not vote for. India's sovereign-AI build — compute, domestic models, BHASHINI, Sarvam — is, read this way, an attempt to own enough of the multiplier that the multiplicand cannot be switched off.

The second shadow is the base narrowing under you. This week's funding data showed Indian tech capital concentrating into a small core — funding up, but rounds down 43% and active institutional investors down to 488 from a peak of 824. Capital selecting hard for an AI elite is depth at the top; it is also a thinning of the early-stage layer that grows the next Sarvam. A multiplier applied to a narrowing base produces spectacular winners and a shallower bench, and the bench is where resilience lives.

The third shadow is the one the desk section left open: displacement without a replacement escalator. The augmentation read — entry roles evolve into AI-supervision, new categories appear — may hold. But 96% of managers expecting roles to evolve is a statement of intent, not evidence of net job creation, and the entry rung the multiplier compresses fastest is the precise rung that has carried Indians from college into the middle class. If the multiplier multiplies output-per-worker faster than the economy invents new on-ramps, the transformation is real and the human arithmetic underneath it is still a subtraction.

It is not an accident that the same week carried the RBI's draft model-risk regime. Naming AI models explicitly, assigning board accountability, requiring an off-switch and a risk tier for every model in production — that is the institutional layer trying to make a powerful, sign-neutral multiplier safe to apply at scale in the part of the economy where a bad multiplication wipes out savings. Governance is not friction opposed to the force-multiplier story. Governance is what the force-multiplier story requires once you admit the multiplier multiplies fragility as readily as output.

The transformation that's actually underway

So is AI a force multiplier for enormous change in India? Yes — and the honest version of that yes is more interesting than the slogan, because it is specific about where.

It is already multiplying knowledge-work output, measurably, at a scale only India's labor base makes possible — with the distribution of the gains, and the fate of the entry rung, still genuinely undecided. It is poised to multiply reach into a billion-person, 22-language, public-rails base that almost no other country possesses — the strongest and most India-specific version of the thesis, and still mostly a forward claim awaiting deployments at population scale. And it is getting cheap exactly as that base is largest, which is the mechanism that turns the first two from possibility into fact, because cost was always the gate. Those are three different transformations at three different stages of reality, and collapsing them into one word — "transformation" — is how the conversation loses precision.

The strategic point is that India does not need to win the multiplier. The multiplier is commoditizing, cheapening, and arriving on its own; this week's events were almost entirely about applying it, not building it. What India has to do is the unglamorous work on the multiplicand: deepen the base that is its genuine advantage — the languages, the rails, the frontline layer — own enough of the frontier that the multiplier cannot be switched off from outside, widen the capital bench instead of letting it narrow, and decide, deliberately and soon, who captures the surplus when output-per-worker doubles. A force multiplier applied to a large, well-tended, sovereign base is the most consequential economic event of the decade. A force multiplier applied to a base you do not control, do not broaden, and do not govern is the same arithmetic running the other way.

The multiplier is here. It is cheap, it is powerful, and it is mostly someone else's. The transformation will be exactly as large as the base India builds for it to act on — no larger, and no kinder than the choices made about who the multiplied surplus belongs to. That is not a forecast. It is just the metaphor, taken at its word.