Cloud Isn’t Right for Everything; You Need a Mindful Approach to Infrastructure

The public cloud delivers immense value for certain workloads, such as proofs of concept and other use cases that require highly elastic infrastructure. However, when businesses default to the public cloud for all their infrastructure needs, including for production workloads, it raises common challenges:

• Maintaining control over workload performance and data residency
• Avoiding high costs driven by networking egress fees and pay-per-use pricing models for established, stable workloads

Despite the enduring benefits of public cloud, it’s simply not right for every workload. Given the rapid acceleration of enterprise AI adoption, getting the right mix of public and private infrastructure is more important than ever. It’s also a delicate balancing act, which requires a mindful approach.

In some cases, public cloud is also the more cost-effective option. For instance, if you use cloud services to deploy a database, those services come with built-in database expertise. This could save $200,000 a year that would otherwise go toward hiring a full-time database admin.

On the other hand, if you’re hosting stable-state compute workloads that don`t require rapid scalability, then the unit-based price model would be less cost-efficient than running the same capacity on private infrastructure.

Today, enterprise IT leaders are increasingly pursuing hybrid multicloud to optimize their IT estate. They still rely on public cloud for experimenting with new use cases like AI. With cloud, they can access GPU capacity to get started quickly.

Equinix has gained a nuanced, front-line perspective on how enterprises have mitigated the risks of growing up and scaling in the cloud. Let’s take a closer look at what we’ve observed.

Diversifying your cloud mix

Typically, the shift away from cloud-first occurs when businesses are ready to:

• Advance from experimentation to production
• Standardize their environments
• Scale their operations
• Commercialize their services

Once they’ve reached these thresholds, enterprises can rarely achieve long-term value using public cloud infrastructure alone. Using a hybrid multicloud architecture can help offset the potential drawbacks in key areas. According to the 2026 State of the Cloud report from Flexera, organizations have already repatriated 23% of the workloads and data they previously hosted in the cloud.[1]

Networking costs

Beyond common challenges like data egress fees and workload optimization, one commonly overlooked issue in the public cloud is intra-cloud networking costs.

Connecting resources within a cloud provider’s network requires data transit services such as:

• Network address translation (NAT) gateways to connect to the internet
• Transit gateways to connect virtual private clouds (VPCs)

The cost of these services typically starts out small, but as the business scales, it can add up dramatically. This can make public cloud economics less attractive, particularly for production workloads.

This emphasizes the importance of avoiding cloud provider lock-in. With a multicloud strategy, businesses can use vendor-neutral connectivity solutions that are more cost-effective.

Control

In addition to spending more than you need to, scaling up in the public cloud can also mean giving up control.

Hosting sensitive data on public servers can raise serious security concerns. And, given that leading cloud providers are now truly global in nature, it’s difficult to know exactly where your data is physically stored at any given time. In turn, this makes it practically impossible to ensure data sovereignty. While public cloud providers have begun to develop sovereign cloud solutions, these offerings are not yet sufficient to meet an enterprise’s complete infrastructure needs.

By investing in private infrastructure, businesses can not only maintain complete control over their own data, but also ensure close proximity to multiple cloud providers. This minimizes latency for workloads and data that do belong in the public cloud and allows enterprises to easily trial different cloud LLMs  without having to place data in cloud native storage.

How to apply mindfulness to AI infrastructure

Perhaps nowhere is this public-versus-private balancing act more relevant than in AI.

Businesses are learning how cloud-first strategies impact the transition to bigger AI workloads, and the unexpected implications of putting those workloads in the cloud.

Once again, the question isn’t public infrastructure or private infrastructure; it’s which workloads go where.

Public AI

For AI, the benefits of public cloud include the ability to experiment, easy access to the latest services and models, and minimal setup requirements. These benefits make public cloud a particularly good fit for businesses that are just beginning their AI journeys.

On the downside, the pay-per-token cost model means that every single character of text that enters or exits an AI model costs you in some way. This quickly becomes unwieldy at scale, and it’s further compounded by agentic AI. Every interaction an AI agent has, with human users or other agents, costs additional tokens. For production workloads, there’s no limit to how high these costs can get.

There are also security and compliance concerns associated with sending your AI data to outside servers, because it means giving up control over how that data is handled and where it’s stored.

Public AI: Hyperscalers, neoclouds or Model as a Service (MaaS) providers

Private AI

Private infrastructure is a better fit for AI production and commercialization, where consistently large workloads are a given. It does require higher up-front investment and in-house infrastructure expertise, but the potential benefits often outweigh these concerns.

Also, your data never has to leave your private infrastructure, allowing you to maintain complete control. This benefit is especially important for regulated industries like financial services.

Perhaps most critically, you only pay one fixed cost for infrastructure. The more you use it, the lower your per-token cost will be. In contrast to public infrastructure, there’s no limit to how low this per-token cost could get for production workloads.

Private AI: On-premises, colocation or managed private cloud

What does this mean for leaders deploying AI?

A hybrid multicloud mix of public and private infrastructure helps businesses achieve the balance that’s right for them. As we enter the AI era, getting workload placement right has never been more important. No enterprise AI strategy can succeed without the right mix of infrastructure to support it.

Of course, we can’t account for every enterprise’s unique situation. But we can speak generally about what tends to work well:

• Target private infrastructure first, with cloud as a complement: We estimate that roughly 80% of enterprise workloads would benefit from private infrastructure, while 20% of workloads are better on public cloud.
• Private infrastructure is worth the effort: Many IT leaders find the cost and complexity of standing up private infrastructure to be off-putting. But for production AI workloads that drive consistently high demand, it’s the only logical option.
• How you connect matters: It’s not just about which infrastructure you pick to host your workloads. You need networking that’s formulated for the unique needs of hybrid multicloud, so that workloads and data can move between different environments whenever the need arises.

Adopt the right hybrid multicloud approach with Equinix today

Equinix can help you take a mindful approach to building and connecting your hybrid multicloud infrastructure, while also helping you navigate the unique challenges of the AI era:

• With Equinix Fabric®, you can create secure, private connections to the cloud providers of your choice.
• You can build your private infrastructure in the same places where the cloud providers already are, and take advantage of the industry’s largest portfolio of low-latency cloud on-ramps.
• This makes hybrid multicloud networking much quicker and easier.

The numbers speak for themselves: Our customers have already created more than 69,000 virtual connections worldwide using Equinix Fabric.

Learn more about how leading organizations are using the right connectivity solutions to bring together their different public and private infrastructure environments: Read the Equinix research paper on the global state of hybrid multicloud networking.

[1] 2026 State of the Cloud Report, Flexera.