The 'Agentic AI Factory' Is a Distribution Deal. Here's Who Really Gets Paid.

Uniphore and Tech Mahindra have announced a partnership to deliver an "agentic AI factory" for the enterprise. On its face the deal is simple: Uniphore's AI agent platform, Tech Mahindra's enterprise delivery machine, one press release. The word that rewards attention is "factory."

"Factory" is not a product and not a data center. It is the current industry metaphor for industrializing the deployment of AI agents — NVIDIA popularized "AI factory" for accelerated data centers, and McKinsey told CEOs to build "agentic factories," imagining a ratio of 100 AI agents supervised by five humans and cost cuts above 50%. The reality under the metaphor is more mundane: most enterprises are not building dedicated internal factories; they are licensing a platform and paying an integrator to install it.

That is exactly what this agreement is: a license-and-integrate engagement wearing a factory's name. That makes the interesting questions economic, not technological. What does each side need from the other, at what margin, and where does the money in "agentic AI" actually go?

Start with the private company, because its numbers are the least exposed to daylight. Uniphore was founded in 2008 at India's IIT Madras and moved to Palo Alto; today it sells a "Business AI Cloud" of enterprise-grade, governed agents. It raised $400 million in February 2022 at a $2.5 billion valuation, then a further $260 million in October 2025 — from NVIDIA, AMD, Snowflake, and Databricks — at that same $2.5 billion. Three and a half years, a full product pivot, roughly $870 million raised in total, and later investors paid the same price, not a richer one. A flat round is shorthand for the private market declining to mark a company up. Revenue was not disclosed in either round; the only figures in public circulation are a 2022 report that annual recurring revenue was "approaching $100 million" and an unverified third-party tracker claiming $500 million today. A five-fold swing in the only numbers available is not a sign of financial transparency.

The investor list explains part of the picture. NVIDIA profits whether Uniphore succeeds or not, as long as its models consume GPU time; Snowflake and Databricks sell the data plumbing underneath. These are ecosystem investments that pull Uniphore's workloads onto each investor's stack — not an independent market signaling a mark-up. The sales pattern fits the same read: in the past thirteen months Uniphore has announced partnerships with KPMG, Cognizant, Rackspace, and now Tech Mahindra — each a services or cloud operator, each a way to borrow somebody else's enterprise shelf rather than win it itself. The "agentic AI factory" phrase is even recycled: in 2025, a customer called its own Uniphore rollout an "agentic AI factory." Same template, new partner.

A private challenger selling access to its platform through one services firm after another is a distribution problem, not a product story.

The public side is a different, cleaner story on a longer ledger. Tech Mahindra's turnaround is real and audited: $6.4 billion of revenue for the year ended March, EBIT up 39% to about ₹7,150 crore at a 12.6% margin, and in the quarter that ended in June, EBIT rose 53% year over year on $1.08 billion of new deal wins. Notice how the recovery is built. Tech Mahindra has stacked third-party AI platforms into its delivery practice — ServiceNow (expanded in August), Microsoft, NVIDIA's Orion, StackGen, UKG, Cisco, now Uniphore — most within a year. It bundles the platform, sells the integration hours, and keeps the margin on the hours. That is how this turnaround is coming back: servitize, don't build.

The uncomfortable part is that the price for that integration work is deflating, and the deflation is being caused by the same AI. Reuters reported this month that Indian IT clients demand the same work for 25% to 30% less; that TCS's business-process contracts are now roughly 80% outcome-based, about double what they were when AI went mainstream in late 2023; and that providers are signing zero-payment first years in hope of sharing future efficiency gains. The Nifty IT index lost about a fifth of its value over the past year — roughly $73 billion of market value. The market is pricing the services layer for what happens when the delivered product gets this good: the price per unit of work falls even as the work gets faster. Tech Mahindra's own CEO is on record calling competitors' bidding "irrational," citing rivals that guarantee prices while assuming 70% to 80% productivity gains over five to seven years.

So where does this leave a US retail investor reading the headline? Hold the two stakes separately. Directly, the partnership changes nothing for a US account: no dollar figure has surfaced in public coverage of it, its own precedent with the same partner — a 2020 contact-center agreement — was described only as "multi-year, multi-million dollar," Uniphore is private and not buyable, and Tech Mahindra trades on Indian exchanges as TECHM, not on the NYSE or Nasdaq.

Indirectly, the deal is a usable map of where this wave pays. The value sits with the companies that own the shelf — the platforms enterprises already buy, and the silicon that runs the agents. NVIDIA, which owns the compute and sits on Uniphore's cap table, is a roughly $5-trillion company trading near 17 times trailing revenue; ServiceNow trades near 78 times earnings. The services layer that installs all of this is cheap — Accenture near 15 times earnings with a 3.5% dividend yield, Wipro near 14 times — and there is a reason: its future revenue is contingent, contract by contract, on outcomes the client defines, at prices the client is driving down.

That yields a test for the next "agentic factory" headline. Does it attach real numbers? Is the platform owner an incumbent, or an independent vendor borrowing distribution? And is the price per unit of delivered work rising or falling? This announcement answers none of those questions, which is the honest thing to say about it — and the useful one. The factory name will keep rotating; the price per unit of delivered work is the number that actually moves.