Companies that move more of their operations to AI trade reliance on a workforce for reliance on technology. AI makes it possible to shift operational responsibility from people to machines, and each company has to decide how far to go. Headcount savings don’t settle the resilience, financing, and data governance questions that follow.

Sachin Ohal is Chief Technology Officer at International Systems Technologies (IST) and has more than 22 years of enterprise technology experience. He previously served as Chief Architect for Distributed Cloud and Industry Solutions at Deloitte Consulting and Global Head of Data Services and Cloud at CSL Behring. He’s worked across biotechnology, insurance, and banking, and advised a major bank on financing a technology partnership, giving him a view of AI adoption from both the infrastructure and financing sides.

“Now you have significantly lower human costs but substantially higher technology costs,” Ohal says. Lower payroll is the visible half of that trade, and the technology bill is the half that grows.

Fewer people means the technology can’t be allowed to fail

Ohal sees enterprises choosing between two paths. One keeps humans in the loop alongside a sizable workforce. The other hands most responsibilities to AI, to the point where, in his example, five people could run an entire data center. In that model, “where you have fewer humans and greater dependence on technology, you need to ensure that the technology will be available when you need it,” he explains. Supply chains tied to geopolitics, rare minerals, and manufacturing capacity make that availability uncertain, so enterprises may buy GPUs, CPUs, and other hardware before they know they’ll need it.

He compares the resilience problem to an airport with two air traffic controllers overseeing 100 incoming aircraft because the organization trusts its automation. When the automation fails, those two people have to manage everything. “In the second model, where you’ve significantly reduced the human workforce, a major technical failure could bring the entire business to a halt,” he says. A chief risk officer’s likely answer is duplicate systems and extra infrastructure under business continuity planning, which pushes capital and operating costs higher still.

Hardware bought ahead of need also sits on the books with an uncertain useful life and depreciation schedule, which, Ohal notes, can force companies to revise shareholder guidance quarter after quarter. 

Capital spending and thinner margins strain the early math

“Previously, an organization’s technology spending might have represented roughly 10% to 15% of its budget. Now, with AI, that figure can move toward 15% to 30%,” he says. “A company that previously operated with net margins somewhere between 5% and 25% could find itself operating in a much narrower range, perhaps 5% to 15%,” Ohal adds. Restoring those margins takes either new sales, which cost money and time, or fewer employees.

Ohal sees companies laying off staff in the name of AI, but not because they no longer need those people. “In many cases, it’s because the financial equation doesn’t work during the early stages of AI adoption,” he says.

Depending on models leaves enterprises unsure who controls their data

A company that has cut people and committed capital is also betting on systems whose handling of its data it can’t fully see. Ohal says enterprises pass proprietary information into AI models without clear answers on where it’s processed or stored, and he worries that AI-driven systems increasingly make those placement decisions dynamically. A provider’s published safety framework doesn’t necessarily give an enterprise full visibility into data in motion, at rest, or during execution.

“If AI systems make decisions about data processing and movement without sufficient visibility and human oversight, enterprises risk losing meaningful control over their information,” he says.

Hybrid architecture puts a human-governed boundary around the model

Ohal points to cloud adoption, where early resistance gave way to hybrid approaches, and expects AI to need the same. In his model, enterprise data stays in a company-controlled environment, potentially on-premises, with the AI model isolated behind a separate firewall. “Between those environments, there should be a clearly defined layer of policies, controls, and governance,” he says, overseen by humans who decide what information reaches the model and for what purpose.

He expects three patterns to emerge within three to five years. The first delivers AI as a service on top of enterprise-controlled infrastructure, which suits large organizations with their own data centers. The second adds an AI layer to existing cloud infrastructure through “a highly restricted gateway between the two technology stacks.” The third embeds AI in devices and dedicated hardware, giving direct visibility at the data and application level. Fortune 500 companies may adopt all three, while smaller businesses will likely lean toward the second and third.

Data ownership may need an independent arbiter

Ohal expects ownership of enterprise data to become one of the industry’s biggest discussions, with the LLM provider, the cloud provider, and the enterprise each a possible claimant to control. In his view, major technology companies face a bind, because trying to establish complete control over the process could draw accusations that they’re monopolizing enterprise data. 

The above leaves room for an independent organization, such as a nonprofit or a body with government affiliation or oversight. “Whoever can develop a trusted solution to that problem could hold the keys to the kingdom,” he concludes.