Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Kevin Roof, Global Director of Offer & Capture Management, LiquidStack
The world is expected to spend over $1.7 trillion dollars developing data centers between now and 2030. To meet this immense (and growing) demand for AI and high-powered computing, these increasingly massive facilities are being built out at breakneck speed, using innovative techniques and technologies to supercharge development and enhance efficiencies. And in the year ahead, this trend will only continue to accelerate, bringing about a variety of other significant changes to the industry and its surrounding ecosystem that will forever change the market. The following predictions represent just a couple of many such changes we are likely to see in 2026.
Data center investors and operators will trade in the classic PUE metric for “tokens per watt per dollar.”
Infrastructure buildout is beginning to shift the economics of AI, with data centers transitioning from cost centers to revenue generators. With this transition, metrics for success are shifting from focusing solely on conventional efficiency and toward how those efficiencies can drive revenue generation. The new, top-of-mind metric discussed in industry circles today is “tokens per watt per dollar,” which is a compound efficiency metric describing how many AI model tokens you can generate for each watt of power consumed per dollar spent. It works to capture computational efficiency (tokens per watt) and economic efficiency (tokens per dollar) in a single expression. In short, it’s an ROI metric that measures how much it costs to generate AI value. This shift in focus highlights the ways in which traditional goals of energy efficiency and sustainability can work side-by-side with aims of economic efficiency. This represents a transition away from a cost-reduction mindset and instead towards one of revenue generation, and it marks a paradigm shift in the way we think about AI data center operation. Since power constraints are the threshold preventing data center growth, organizations must use the power they have most effectively. Stranded power represents lost revenue, and operators will be working overtime to maximize their tokens per watt per dollar as much as possible.
Use of temporary buildings and modular solutions will be vital to achieve maximum “time to power” for data center deployments.
With the aforementioned changing economics of data center operation, we’ve reached a point where these facilities simply can’t be built fast enough. The race to innovate and generate revenues is fueling an explosion in data center development that would have been simply unimaginable just a few short years ago. To jumpstart deployments and ensure available power is utilized immediately, I believe we will see more organizations turn to modular data centers and temporary buildings-much like Meta has with the use of temporary tents in recent projects-for data center operations until a permanent on-site facility is completed. We are also likely to see a greater number of building retrofits, as developers seek out stranded, unutilized space and power in order to generate this all-valuable compute and realize revenues as quickly and efficiently as possible. These types of strategies will be used more and more as organizations seek to dramatically reduce their overall “time to power” (i.e. the period of time between initiating a project’s development and generating compute) in order to accelerate ROI timelines and generate revenues that can then be reinvested back into additional projects and development.
2026 Will Mark the Beginning of a New Era in Data Center Design
With so much change taking place in the world of computing, the underlying infrastructure of data centers is poised to undergo some dramatic changes in the months and years ahead. As AI workloads push the limits of what’s possible in existing facilities, we’ll see more and more greenfield and retrofit projects employing novel techniques and technologies (such as liquid cooling systems and modular design practices) in order to accelerate deployment timelines and maximize near and long-term return on investment. As AI and the associated hardware continue to attract historic levels of investment, data centers will continue to grow and evolve in lock-step-enabling the continuing AI revolution, well into 2026 and beyond.
##
ABOUT THE AUTHOR
Kevin Roof is the Global Director of Offer & Capture Management for LiquidStack. He brings deep experience in data center liquid cooling, with over a decade of experience in engineering and product management. A mechanical engineer and PMP, he advances LiquidStack’s technology roadmap and strengthens its offerings. Kevin’s unique blend of technical expertise and industry knowledge positions him as a rising thought leader in the liquid cooling space.
