And while the enthusiasm for AI adoption is understandable, IT decision-makers may not know where to start — especially with competing budget constraints and cuts. That’s why starting with a defined use case matters so much, and why CDW and partners such as TD Synnex ground these conversations with AI ideation workshops.
These touchpoints include the team sitting down with the institution’s AI team or AI center of excellence to identify concrete scenarios in which AI can make an impact across the schools of business, engineering, nursing law or liberal arts. From there, institutions can prioritize the use cases most likely to deliver measurable value and build toward them with a realistic implementation roadmap.
Modeling AI Consumption Before It Becomes a Runaway Line Item
Many colleges and universities have a team and an idea of what they want to do with AI, but they don’t have a goal that they are building toward. When I talk with IT leaders, I tell them that once we define the use case, we need to explore what the top two or three priorities will be and define what “good” looks like for them.
From there, we map out the hardware and bill of materials that will handle 80% of what they need, because they’re never going to get it 100% correct.
READ MORE: FinOps strategies can help higher ed IT navigate cloud-related costs.
Whoever is paying the bill needs to understand the consumption of the different users that have access to their AI model: faculty, students, researchers and administrative staff.
If something goes off the rails, there is also a line of sight into when adjustments are necessary, and IT teams can shut things off before the bill cycle is over. Some institutions leverage their visibility through AI and cloud observability through tools such as DataRobot or Datadog for model performance, or Terraform-based assessments that show where cloud resources — and by extension, cloud costs — are being wasted.
How To Approach Your AI Infrastructure and Consumption
Nine times out of 10, it’s going to be quicker, easier and cheaper to get in the cloud first. So, your use case determines your budget: your capital expenditure which represents investments that can be annual or go beyond a given fiscal year, and/or your operating expense, which can account for more flexible ongoing funding.
The challenge with not doing this planning upfront is that you can end up signing up for a subscription that you can’t regulate. You won’t have control over what professors or research scientists do. They could run up many excess token bills on an obscure data set. You won’t know what's going on until the bill hits. This is where the Zoltar machine stops being fun and novel.
LEARN MORE: Scaling token awareness now can mitigate scaling costs later.
For example, I’ve seen universities pay $200,000 annually for access to a large language model. That only pays for the right to be in the game. On top of it, the institution receives a separate token bill once usage crosses the allotted threshold — driven largely by unfenced research activity from faculty and graduate students.
Advice is free until you don’t take it. And for many institutions, the difference between a sustainable AI program and a runaway token bill is whether they invite partners such as CDW into the room before they start dropping quarters into the machine.
