Cognizant Launches AI Nexus to Meter AI Spend

Cognizant has launched AI Nexus, a set of five software modules that it says will let companies track what they spend on AI, check the quality of AI work and send each task to a suitable model. The technology services company describes it as “a modular, multi-model operating layer” in its launch announcement on 22 October. It gave no pricing and named no customers.

The pitch is aimed at finance chiefs rather than engineers. Cognizant argues that the choice of model now matters less than proving what AI spending returns, and it is selling AI Nexus as the place where that proof is collected. Everything it says the product does is, in the company’s own wording, “expected functionality and potential benefits”.

Five modules split between doing the work and policing it

Cognizant splits the five modules into two groups, and customers can adopt them separately or together. Two change where work happens. Engage is a shared workspace grounded in a company’s own knowledge, where staff work with persistent “AI specialists”, which Cognizant defines as AI agents with their own context and memory. Orchestrate rebuilds business processes as agentic workflows, designed and run by two staff roles Cognizant created in June.

The other three govern how AI performs. Reveal “maps every token, tool and dollar across AI vendors” to the people, projects and outcomes behind them. Coach evaluates AI work as it happens and turns better practice into company standards. Route sends each request to a frontier, open-weight, small or self-hosted model, depending on quality, cost and control, and is designed to switch to other models when a provider slows or fails. On its product page, Cognizant says routing works “through your AI gateway of choice” and promises to use frontier models “only where it earns its premium”.

Cognizant says the layer is designed to run inside the customer’s own environment. Each AI specialist has its own identity, permissions and memory, every workflow keeps a full audit trail, and human approval stays in place for critical decisions, compliance checks and customer-facing actions. Cognizant says it already runs AI Nexus “at scale” across its own operations.

Cognizant’s own survey frames the cost problem

The launch leans on a Cognizant survey of 250 FinOps leaders at companies with more than $500m in annual revenue, 50 of them in the UK. In it, 72% said their organisation had exceeded its AI budget in the past 12 months, and 16% had overshot by more than 25%. The company also says 97% of organisations had seen an AI agent session escalate beyond expected cost or activity, with 36% calling the impact material. These are Cognizant’s figures, from a company selling the remedy.

“What boards and finance leaders need now is evidence: what that spend returns, who is accountable for it and how to get more from every dollar,” chief executive Ravi Kumar S said. Arvind Pal Singh, Cognizant’s EMEA head of consulting and AI, put the argument more bluntly in a blog post a day earlier: “The models themselves have more or less converged: similar capability, similar pricing, open to whoever’s willing to pay.”

The people behind Orchestrate are part of a bigger Cognizant bet. In July it committed to scaling to 5,000 Frontier Certified Engineers and 10,000 Frontier Business Operators, with the first deployment-ready cohort due by the fourth quarter of 2026.

Runaway agent bills are a governance problem, not a model problem

The most credible part of Cognizant’s case is the one it shares with other vendors. If Cognizant’s survey is even roughly right, agent sessions that run past their expected cost are routine, and the risk is not only internal: when Railway’s breach exposed customer secrets this week, its guidance warned that a leaked AI key spends prepaid credits directly. A layer that ties each token to an owner addresses both problems in principle.

Cognizant is also not alone in selling the routing idea. Notion says its Auto router cut costs by about 90% in its own tests, and cloud platforms are widening the menu, as Google Cloud did by adding OpenAI’s GPT-6 models. Microsoft has taken a stricter line on control, treating agents as potential insider risks and adding Guardian Agents to watch them. Where Cognizant differs is that it sells the governance layer alongside its own people to run it.

That makes the missing details important. Without pricing, named customers or an independent test, the evidence for AI Nexus is so far Cognizant’s own. The first outside customer to publish what it saved, or what it spent, will say more than the launch did.