The Cloud Cost Reset: Why Enterprises Are Scaling From “Cloud First” to Hybrid Infrastructure

Enterprises are becoming far more selective about what they place there.

The next infrastructure cycle appears to be hybrid. It features public cloud for elasticity, experimentation, and temporary capacity, private cloud or colocation for predictable, high-utilization workloads, and on-premises infrastructure for sensitive data, steady-state compute and economically sustainable AI.

A Broadcom-commissioned survey of 1,800 senior IT leaders found that 69% were considering moving workloads from public to private cloud, while more than one-third had already repatriated workloads. The direction is increasingly workload-first rather than cloud-first.

Public-Cloud Customers Are Already Beginning to Fund the AI Buildout

These companies will improve utilization and develop custom silicon, but the underlying economics remain unavoidable. The costs of GPUs, HBM, networking, power, cooling, construction, and financing must ultimately be recovered from customers.

Over the next one to three years, enterprises should expect that recovery to appear through some combination of:

• Higher compute and managed-service pricing

• More expensive premium AI capacity

• Longer reserved-capacity commitments

• Increased software and platform charges

• Egress, storage and networking economics

• Tighter consumption discounts

• Greater pressure to commit spending in advance

The impact may not arrive as one dramatic price increase. It could appear gradually across dozens of cloud services and then hit enterprise budgets abruptly at renewal.

Real-World Perspective

GEICO

GEICO disclosed that it was repatriating a majority of its cloud workloads after annual cloud spending reportedly exceeded $300 million. It’s infrastructure strategy uses Kubernetes, OpenStack, open hardware, and substantial on-premises storage rather than returning to traditional legacy operations.

GEICO subsequently reported reductions of approximately 50% per compute core and more than 60% per gigabyte of storage through its private-infrastructure approach.

The lesson is not that every company should copy GEICO. The lesson is that a sufficiently large, predictable workload can justify infrastructure ownership.

37signals

37signals moved Basecamp and HEY away from public-cloud infrastructure after determining that its mature, steady workloads did not require hyperscale elasticity. The company projected millions of dollars in savings while operating owned hardware in colocation facilities.

This demonstrates that modern private infrastructure does not necessarily mean building and staffing a traditional corporate data center. It may mean owned equipment, colocation, automation, and a deliberately simplified architecture.

Broader Enterprise Movement

Almost half of the organizations in Broadcom’s survey estimated that more than 25% of their public-cloud spending was wasted. Common causes included idle compute, oversized resources, unmanaged development environments and excessive storage consumption.

This waste becomes much harder to tolerate as enterprises simultaneously fund AI programs, cybersecurity initiatives, and broader infrastructure modernization.

The AI infrastructure boom will not be funded by hyperscale’s alone. As hundreds of billions of dollars are poured into next-generation GPUs, power, networking, and data centers, those investments will ultimately be recovered through cloud consumption. Enterprises are already feeling the impact in hardware refresh cycles; public-cloud customers should expect the same cost pressures to accelerate over the next 1 to 3 years.