\n<\/p>\n
Mark Morgan is President of Commercial Operations at <\/em>Kinaxis<\/em><\/a>, helping global enterprises transform supply chains with AI.<\/em><\/p>\n \u200bThe AI gold rush is no longer a metaphor. Global AI spending is projected to reach $2.5 trillion in 2026<\/u><\/a>, growing rapidly year over year, as hyperscalers pour hundreds of billions into data centers and compute capacity.<\/p>\n Capital is flowing. Campuses are rising. Compute capacity is scaling at historic speed.\u200b\u200b But beneath the headlines lies a harder operational truth:<\/p>\n AI infrastructure is scaling faster than enterprise decision-making. And that gap is becoming the real bottleneck.<\/p>\n AI data center expansion is the industrial revolution of our era. Steel and rail have been replaced by silicon and fiber\u2014but the constraints are just as real.<\/p>\n The scale alone is staggering. North America is expected to require approximately 92 gigawatts of additional power capacity over the next five years to support AI data centers. Energy is no longer a background utility; it is a gating factor.<\/p>\n Supply chains are tightening in unexpected ways. Projections suggest that data centers could consume up to 70% of global memory chip production by 2026<\/u><\/a>, creating ripple effects across industries, from consumer electronics to automotive and medical devices.<\/p>\n Against this backdrop:<\/p>\n \u2022 Energy grids are under strain<\/p>\n \u2022 Semiconductor supply remains concentrated and volatile<\/p>\n \u2022 Lead times for critical components stretch months<\/p>\n \u2022 Demand for AI workloads is nonlinear and unpredictable<\/p>\n \u2022 Severe scarcity of supply chain experts<\/p>\n Meanwhile, the infrastructure itself has become exponentially more complex.<\/p>\n AI servers are not incremental upgrades. They are dense, GPU-centric systems with intricate thermal, power and compatibility requirements. A single rack may contain thousands of valid configurations, each with implications for performance per watt, cost, cooling and deployment timing.<\/p>\n Deployments that once scaled in thousands of racks per quarter now scale in thousands per month.<\/p>\n Each rack represents revenue potential. A delay in standing up capacity can mean lost product velocity, deferred customer commitments and missed market opportunities.<\/p>\n In this environment, planning is no longer a back-office function. It is a strategic growth lever.<\/p>\n AI infrastructure creates a high-stakes balancing act.<\/p>\n Stock too little, and you miss deployment windows. Stock too much, and you strand capital in mismatched or obsolete inventory.<\/p>\n Consider GPUs. They remain the economic engine of AI infrastructure. Scarce, allocation-driven and rapidly evolving. But they are only one piece of a tightly coupled system.<\/p>\n Now layer in memory constraints, networking equipment, power systems and cooling infrastructure. Each component is interdependent. A shortage in one area\u2014such as memory\u2014can stall entire clusters, even when GPUs are available.<\/p>\n Global data center spending is projected to exceed $500 billion<\/u><\/a>, and individual racks can cost tens or hundreds of thousands of dollars. At this scale, even small misalignments matter.<\/p>\n A modest 5% mismatch between supply and demand translates into billions in capital friction.<\/p>\n And increasingly, those mismatches are not driven by lack of investment, but by lack of coordinated decision-making.<\/p>\n Most enterprises rely on ERP and MRP systems to manage this complexity.<\/p>\n These systems are essential. They track transactions, manage bills of materials and ensure financial integrity.<\/p>\n But they were not designed for probabilistic, constraint-driven optimization at hyperscale.<\/p>\n AI infrastructure planning is inherently dynamic. Demand signals shift. Supplier performance fluctuates. Constraints, especially power and component availability, change in real time.<\/p>\n For example, should you deploy fewer next-generation GPU clusters constrained by power availability, or scale out with older architectures that consume more energy but are more readily available? The answer depends on a multidimensional trade-off across cost, performance, energy, timing and risk.<\/p>\n That is not a static calculation. It is a continuous optimization problem.<\/p>\n When assumptions break, traditional systems amplify the gap. Teams revert to spreadsheets. Manual overrides increase. Planning becomes reactive and person-dependent.<\/p>\n At this scale, that approach does not hold.<\/p>\n In the AI era, agility, moving quickly, is necessary but insufficient.<\/p>\n The defining capability is adaptability: the ability to continuously recalculate decisions as new constraints emerge.<\/p>\n This is where decision orchestration, driven by an agentic architecture, becomes critical.<\/p>\n Decision orchestration connects demand, supply, inventory and financial plans into a unified, continuously updating model. If ERP records what has happened, AI orchestration determines what should happen next.<\/p>\n Demand orchestration integrates inputs across product, finance and customers into a unified forecast. Supply orchestration evaluates constraints\u2014from chip availability to power capacity\u2014and recommends mitigation strategies. Inventory orchestration optimizes buffers based on variability and risk tolerance.<\/p>\n Integrated planning enables leaders to ask: This is not about dashboards. It is about synchronized decision-making across capital allocation, supply chain execution and infrastructure deployment.<\/p>\n Industry benchmarks<\/u><\/a> suggest advanced planning can reduce inventory by 20% to 30%, improve forecast accuracy by 15% to 25%, improve on-time deliveries and significantly reduce expedited costs. At hyperscale, that translates into tens or hundreds of millions in impact.<\/p>\n More importantly, it accelerates strategic execution.<\/p>\n AI infrastructure is now a primary growth engine. Data center capacity determines how quickly organizations can scale innovation, enter new markets and compete globally.<\/p>\n But as the scale of investment and constraint intensifies, the limiting factor is no longer just capital, compute or even chips.<\/p>\n It is decision-making.<\/p>\n Planning systems must now be treated as strategic infrastructure\u2014as critical as power, cooling and networking. Boards are scrutinizing capital intensity. Regulators are examining energy consumption. Investors are evaluating return on invested capital. Customers expect reliability at scale.<\/p>\n In this environment, disconnected decision-making introduces risk where continuous precision is required. The organizations that will lead in the AI economy will be those that not only scale infrastructure, but orchestrate the interconnected decisions required to build, allocate and deploy it effectively.<\/p>\n Those that replace reactive planning with continuous orchestration.<\/p>\n And those who recognize that, in the race to scale AI, the real advantage is not just compute but the coordinated, intelligent decision-making behind the infrastructure.\u200b<\/p>\n Forbes Technology Council<\/u><\/a> is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?<\/u><\/em><\/a><\/p>\nThe New Industrial Revolution Runs On Constraints<\/h2>\n
The Inventory Balancing Act: Shortages Or Stranded Capital<\/h2>\n
Why Traditional Systems Fall Short<\/h2>\n
From Agility To Adaptability<\/h2>\n
\u2022 What happens if memory supply tightens further?
\u2022 What if power availability delays deployment in a key region?
\u2022 Which workloads should be prioritized and what is the revenue impact?<\/p>\nPlanning As Strategic Infrastructure<\/h2>\n
\n
<\/div>\n