AI could raise enterprise IT costs by as much as 75% in less than a decade

Bain & Company’s latest warning about AI costs is blunt enough to change spreadsheets: AI could push enterprise IT costs up by as much as 75% in less than a decade, even when organizations try to invest carefully. CIO Dive, citing Bain’s analysis, ties that run-up to a familiar mix that is now arriving at the same time, more infrastructure, more security, and a scarcer talent pool to build and run it all.

What’s useful for operators is that the budget pressure is already visible in three places that procurement and IT operations can touch: how AI workloads are being architected across model types, how “agentic” tools are being governed and certified, and how long it can take to actually get hardware into racks. Developments at Deloitte, KPMG, the U.S. Department of Transportation, and Dell map directly onto those three levers.

The cost spike is less about the model, more about everything around it

Bain’s cost outlook, as reported by CIO Dive, is a reminder that the AI line item rarely stays contained. Storage and networking move with the data. Security expands with the new attack surface. And in many enterprises, the biggest hidden multiplier is people, platform engineering, governance, and incident response capacity that didn’t exist when AI programs were just pilots.

CIO Dive also reported Gartner data showing fewer than 25% of enterprises have scaled AI successfully. That matters because failed scale attempts still burn real dollars, GPU reservations, platform build-out, security reviews, and time from the teams that also keep core systems running. In practice, the “AI tax” can include the cost of deciding to stop.

Deloitte’s hybrid-model push turns model choice into an operating model

In that same cost environment, Deloitte is leaning into a hybrid strategy that treats proprietary and open models as a portfolio rather than a religion. CIO Dive reported Deloitte is launching an open model engineering practice aimed at helping enterprises find the right mix, a signal that “multi-model” is becoming a normal operating posture, not an advanced pattern reserved for a few.

For CIOs and procurement leads, the practical consequence is contract sprawl unless governance is designed up front. A hybrid model stack can mean separate terms for hosting, fine-tuning, data retention, evaluation, and safety tooling. It also changes integration work. The expensive part is often the plumbing that makes switching possible, model gateways, logging, policy enforcement, and repeatable evaluation, not the model license itself.

Agentic AI is pulling assurance into the buying checklist

As AI tools move from generating text to taking actions, buyers are asking for proof that the vendor has put guardrails into the product. CIO Dive reported KPMG got its agentic tool certified for security and reliability as enterprises confront growing risks from AI agents that can act autonomously.

Certification does not eliminate risk, but it gives procurement teams something concrete to negotiate around: defined control objectives, test artifacts, operational limits, and escalation paths. It also changes what “implementation” means. For agentic tools, the biggest integration work can be governance workflows, approvals, auditing, change control, and role-based permissions that map to how the business actually runs.

Public-sector continuity at DOT shows what ‘keep the lights on’ looks like in AI-era IT

Leadership changes can look like politics from the outside, but for operators they often signal whether a modernization program will keep moving. According to FedScoop, the U.S. Department of Transportation named Jack Albright acting chief digital and information officer and deputy CIO for IT shared services. FedScoop also reported Albright takes over for Pavan Pidugu, whose last day is Friday.

According to FedScoop, Albright has served as DOT’s deputy CIO since December 2020 and previously held an associate CIO role for IT shared services. Pidugu, FedScoop noted, migrated services to Google Workspace and restructured the technology department. The reporting does not say whether Albright’s move signals a change in direction. What it does show is how much AI-era planning depends on the “shared services” spine, identity, endpoint, collaboration, data center operations, and cloud foundations that determine whether AI can be deployed safely and repeatedly across agencies or business units.

Dell’s $95B backlog turns AI infrastructure into a scheduling constraint

Even if the contract is signed and the governance is written, projects still wait on physical reality. CIO reported Dell has a $95 billion AI backlog, framing it as evidence that the infrastructure crunch behind agentic AI demand is not easing yet. CIO’s reporting described shortages stretching across servers, storage, and other infrastructure components.

For enterprise IT and facilities teams, this is where the 75% cost projection becomes real. Backlogs and constrained supply don’t just raise prices, they lengthen timelines, complicate refresh schedules, and force interim architectures. Some organizations will rent capacity longer than planned. Others will standardize on fewer configurations to get priority allocation. In both cases, the operational win goes to the teams that treat infrastructure procurement like supply planning, with lead times, alternates, and pre-approved substitutions, rather than a one-off PO.

Questions to put into AI contracts and capacity plans now

• When a vendor sells an “agent” feature, what are the documented limits on actions, approvals, and audit logging, and are those controls included in the base price or priced as add-ons? (KPMG certification coverage via CIO Dive is a useful forcing function for this conversation.)
• For hybrid model strategies, what is the exit path if a model changes price, policy, or availability, and what integration layer will own routing, evaluation, and logging across vendors and open models? (Deloitte’s open model engineering practice, as reported by CIO Dive, is a signal that these layers are becoming standard work.)
• If hardware lead times slip, what is the approved fallback, burst to cloud, extend existing server life, or shift to smaller models, and which team owns the decision and budget? (Dell’s $95B backlog, per CIO, is a reminder that the fallback needs to be written down before the delay hits.)
• In IT financial planning, what portion of AI spend is being tracked outside “AI,” security controls, data movement, platform SRE headcount, and governance tooling, and does that tracking align to Bain’s cost drivers? (Bain via CIO Dive projects up to a 75% rise in less than a decade.)