There is no shortage of tools that estimate energy efficiency for data centre hardware. There is also no shortage of vendor configurators that produce a recommendation alongside a quote. What is missing, and what we built Interact to address, is the layer in between: an independent, vendor-agnostic source of measured truth that sits next to both and evidences the decision before it is made.
Interact began as a two-year Knowledge Transfer Partnership between Techbuyer and the University of East London, funded by Innovate UK. The brief was simple in principle: build a model that could answer, for any given server configuration, how much energy it would draw under a given workload and how that compared to the alternatives. The simplicity stopped there. Doing it required thousands of hours of hardware benchmarking, a sustained collaboration between academia and industry, and a willingness to publish the findings in peer-reviewed venues whether they confirmed or unsettled the existing wisdom.
What the work surfaced
The most consequential finding from that period was uncomfortable for parts of the industry. The dominant assumption in data centre procurement has been that each successive server generation roughly doubles in energy efficiency, and therefore that the cheapest carbon path is always to refresh to the newest available. Our measured data, against real workload profiles, did not bear that out. With the right component-level upgrades, the immediate past generation can out-perform the latest. Once Moore's Law slows, as it has, the refurbished option starts to win on both cost and energy.
That finding has implications well beyond a single procurement. It changes the carbon arithmetic of refresh cycles, the economic case for a circular approach to IT hardware, and the credibility of vendor-led efficiency claims.
The work was published in 2020 in IEEE Transactions on Sustainable Computing under the title Optimizing Server Refresh Cycles: The Case for Circular Economy With an Aging Moore's Law. It remains, as far as we can identify, the largest peer-reviewed study of measured server efficiency across vendor generations published to date.
The finding has since been corroborated by the silicon side of the industry. AMD, Google and NVIDIA have all publicly accepted the slowdown in Moore's Law and responded by redesigning chip architecture and application loads to extract efficiency from sources other than density gains. The data centre sector, on average, has been slower to absorb the same finding. The procurement habit of refreshing to new every cycle persists more often than it should.
What Interact does with it
We turned the research into a tool. It takes the configuration of a current estate as input, applies the measured dataset to model candidate refresh scenarios, and produces a defensible answer on cost, energy and carbon for an evidenced procurement decision. It is sold to procurement, finance and IT leadership. It sells no hardware itself, and it takes no vendor commission.
The point is not the recommendation. The point is the dataset underneath it.