The Great Infrastructure Miscalculation: AI Is Not Replacing Traditional Infrastructure

Goldman Sachs recently raised its AI server forecast through 2030 by 18 percent. That number drew the headlines. The more revealing figure sat right next to it, a 31 percent increase in the firm’s traditional server forecast. If AI represents the future of computing, traditional infrastructure should be losing ground, not beating expectations by a wider margin.

That contradiction points to one of the larger misconceptions in modern infrastructure planning. Organizations are not trading traditional infrastructure for AI infrastructure. They are stacking AI infrastructure on top of what they already run, and in most cases, they are building another AI infrastructure island in the process.

The Assumption That Shaped Modern IT

Infrastructure planning has leaned on a comfortable assumption for decades. New technology replaces old technology, new servers replace old servers, and new applications push their predecessors out the door. The industry built its architectures, budgets, and refresh cycles around that belief. Each new workload would displace the previous one, and organizations would consolidate onto fewer, newer platforms over time.

The practice never matched the theory. Workloads rarely get replaced, and data almost never disappears. Mainframe applications did not leave when client-server arrived. Client-server did not leave when virtualization arrived. Each wave added a layer, and the layer beneath it kept running. What the industry described as a cycle of replacement has been a steady process of accumulation throughout. Companies still run traditional workloads as they introduce containers, analytics platforms, AI assistants, inference engines, and GPU-accelerated environments. AI is not breaking that pattern. It is confirming it.

The Goldman Signal

The Goldman forecast deserves a closer read than the AI headline suggests. Raising the outlook for AI spending surprised no one. Raising the traditional server outlook by an even larger amount tells a different story. A real shift from legacy infrastructure to AI infrastructure would show up as softening demand for traditional servers. Goldman projects the opposite, with stronger demand across both categories at once.

That projection matches what IT teams describe in the field. Existing applications keep expanding, data volumes keep climbing, and AI initiatives depend on access to that same data and integration with those same applications. AI is not replacing traditional infrastructure. It feeds on it.

AI Becomes Another Infrastructure Island

Storage Switzerland hears a consistent message from organizations evaluating modernization projects and VMware alternatives. A few of them describe AI as a means of eliminating existing infrastructure. Most describe AI as something they will deploy alongside it. The plans involve dedicated GPU clusters, separate storage architectures, and separate management frameworks, each one running on its own operational model. Every decision adds another island to an already crowded data center.

The pattern is not new. The industry has responded to every major workload shift the same way for 20 years. Virtualization became its own platform. Hyperconverged infrastructure became its own platform. In most shops, containers became their own platform. AI now follows that same path. The workload changes, the response stays the same, and the data center gains one more silo to operate.

The Cost of Accumulation

This accumulation explains much of what Goldman measures. Organizations are not replacing infrastructure, they are adding to it. The old environment stays in production, the new environment is deployed alongside it, and the original management tools remain in place as new ones are layered on top. The teams that ran the prior platform stay busy, and the company hires or trains for new skills to run the next one.

Spending rises at every step, not from a single large purchase but from the steady habit of treating each new technology as an addition rather than a replacement. The hard part is no longer deploying the next workload. The hard part is operating all of them simultaneously, with separate tools, teams, and failure domains.

The Layer the Industry Skipped

The data center has always excelled at supporting new workloads. Its weakness has been absorbing those workloads into a shared architecture. Personal computing eventually standardized on operating systems that ran many applications on common hardware. Virtualization standardized on platforms that ran many virtual machines across a shared pool of resources. The modern data center still lacks an equivalent layer, a common operating environment that runs traditional applications, storage services, networking services, containers, and AI workloads under one operational model.

Without that layer, every innovation creates pressure to stand up another platform. Read in that light, the Goldman forecast measures more than infrastructure growth. It measures another cycle of infrastructure duplication.

Future-Proofing Needs a Sharper Definition

Most conversations about future-proofing center on capability, on whether the infrastructure can run AI, run containers, and run whatever arrives next. Those questions matter, and they miss the larger one. Can the infrastructure run those future workloads without spawning another island?

VergeIO made a version of this case in its argument that a VMware alternative has to be AI-ready, judged not only on whether it replaces the hypervisor but on whether it runs containers, GPUs, and private AI without standing up a second platform. The organizations that come out ahead over the next decade will not be the ones that deploy the most AI hardware. They will be the ones that fold AI into their existing operational model with the least disruption. That outcome calls for an architecture designed to absorb new workloads, not one that answers each new demand with a fresh silo.

Storage Switzerland Take

The headline from the Goldman forecast is not that AI spending is climbing. The sharper point is that traditional infrastructure spending is climbing faster than expected. The numbers indicate that organizations are running both environments at once rather than swapping one for the other. Customer conversations explain why. AI has become the latest island in an industry that has spent decades meeting innovation with new platforms, new management frameworks, and new operational silos.

The question worth asking is not how to build a better AI island. The question is whether the business needs another island at all. A handful of emerging architectures aim at exactly that problem by providing a common operating environment for both traditional and emerging workloads. Ultra-converged platforms such as VergeOS fit this category. Rather than treating virtualization, storage, networking, containers, and AI as separate projects, VergeOS combines these functions into a single codebase and a single operational framework.

Whether that model becomes dominant remains an open question. The economics of accumulation grow harder to defend with each cycle. Organizations that keep answering every new workload with a new island will find themselves managing more complexity than innovation. The real infrastructure challenge of the AI era is not deploying AI. It is deploying AI without building one more island to run it.

About George Crump

George Crump is the Chief Marketing Officer at VergeIO, the leader in Ultraconverged Infrastructure. Prior to VergeIO he was Chief Product Strategist at StorONE. Before assuming roles with innovative technology vendors, George spent almost 14 years as the founder and lead analyst at Storage Switzerland. In his spare time, he continues to write blogs on Storage Switzerland to educate IT professionals on all aspects of data center storage. He is the primary contributor to Storage Switzerland and is a heavily sought-after public speaker. With over 30 years of experience designing storage solutions for data centers across the US, he has seen the birth of such technologies as RAID, NAS, SAN, Virtualization, Cloud, and Enterprise Flash. Before founding Storage Switzerland, he was CTO at one of the nation's largest storage integrators, where he was in charge of technology testing, integration, and product selection.