Interact's Rich Kenny on simulating servers before you buy them

Interact, the IT performance and simulation consultancy, has launched Interact Verify, a tool that simulates the cost, performance, carbon and water impact of server hardware before an organisation commits to it. It draws on ten years of research and measured data from more than 500 data centres.

Rich Kenny, Interact's co-founder and managing director, is also a visiting research fellow in the School of Engineering at London South Bank University and chairs the IEEE Water Resource Technologies Standards Working Group. He told The Datatech Times what operators get wrong without that evidence, why water has been ignored for so long, and whether hardware efficiency can keep pace with AI demand.

Newer is not always better

The most common mistake Kenny sees is judging hardware on the assumption that newer always wins, “without a wider understanding of what each configuration actually delivers. A spec sheet tells you what a machine can do in a few variations, not what it does under your workloads or in the actual set up you will be ordering.” The same applies to the refresh cycle, where kit is replaced because it has reached a certain age. “Neither holds up in real life. We regularly see older servers outperforming newer ones. Plus, each configuration and variance has its own outcomes. Generalising leads to a lot of missed opportunities.”

The cost of getting it wrong compounds. “You overspend on the purchase, then carry higher energy and carbon costs for years. On a large estate that gap is over seven figures every time.” One financial services firm Interact worked with measured first and tripled its compute for the same spend, with savings running to millions on one small section of the estate.

From 500 data centres to one estate

The dataset behind Verify records how specific hardware behaves under real workloads, “measured across thousands of hardware configurations and benchmarks, rather than estimated.” Every server is graded on what it actually delivers, in transactions per watt and maximum performance, and an operator's configuration is matched against that measured reference. “What they get back is grounded in evidence from comparable kit doing comparable work.”

Kenny is candid about the limit. “The honest limit is that a simulation points to the right decision, it does not make it for you. The evidence tells you what to buy, keep or retire, but the value only delivers once that is interpreted for your estate and actually implemented.” That is why the tooling is paired with senior advisory on the bigger calls.

Water, finally on the balance sheet

Verify puts water alongside carbon, cost and performance, and Kenny's explanation for why it was secondary for so long is that it was invisible. “Energy has a clear cost and a clear carbon number attached to it, so it gets attention. Water consumption, particularly the water used in cooling of the facility and embedded in the creation of the power being consumed has rarely been measured at the hardware level, so it fell off the decision entirely.”

Measured on the same footing, water becomes something operators can act on “rather than a line in a sustainability report written after the fact. Once water sits next to cost, performance and carbon in the same decision, operators can see the trade offs before they commit, not a year later.” As pressure grows on water as a shared resource and regulators start asking, he said, “the operators who have been measuring it will be the ones able to answer.”

The reporting bar is rising

The EU Energy Efficiency Directive already requires data centres above a certain size to report energy performance, and a proposed common rating scheme would push that towards a standardised, comparable measure. “The bar is moving from ‘tell us roughly how your facility performs' to ‘show us, in a form we can compare against everyone else'.”

Most operators, in his view, cannot evidence performance at the level of the hardware itself. “They can report facility level numbers like total energy use and PUE, but they cannot yet show what the IT equipment inside is actually delivering for that energy, which is where most of the real efficiency story sits.” When the standard asks what an estate delivers per unit of energy, and increasingly per unit of water, “you need measured evidence per configuration, and that is very hard to produce retrospectively if you have not been capturing it.”

Efficiency slows the curve, it does not reverse it

With the IEA projecting data centre electricity use roughly doubling by 2030, towards around 945 terawatt hours, and the AI-driven share growing fastest, Kenny does not oversell what efficiency can do. “On its own, efficiency slows the curve, it does not reverse it. Efficiency per task is improving quickly, but demand is growing faster, so the total keeps climbing. Anyone promising that efficiency alone will solve this is overselling it.”

What it does do is make every watt of installed capacity count, which matters more when capacity is scarce and expensive. “If demand is going to double, the worst outcome is doubling it on top of an estate that is already running hardware it does not need. The lever operators actually control is what they buy and what they retire, and getting that right, repeatedly, is how you keep the estate ahead of the curve rather than simply adding to the problem.”

Finance and IT, looking at the same numbers

Kenny's final point is about the conversation inside the organisation. “Finance and IT are usually arguing from different evidence. IT talks in performance and capacity, finance talks in cost and payback, and because they are not looking at the same numbers they often end up negotiating rather than deciding.” A simulation puts cost, performance, carbon and water in one view.

Take a bank weighing consolidation against a move to cloud. “Instead of each side defending a position, they can model both options against measured behaviour and see the real figures side by side, the spend, the performance they would get, and the environmental cost of each path. It turns a judgement call into an evidenced one. That is usually the moment the conversation shifts from whose instinct wins to what the evidence supports, which is a much better basis for a decision of that size.”