How Fortune 500 Infrastructure Leaders Are Navigating AI Readiness

Last week, the two of us appeared at a panel discussion at Cisco Live, alongside networking leaders from Home Depot and Disney. This session was an opportunity for network and infrastructure leaders to stop holding back and start being frank about what they really think about AI as a tool for managing network infrastructure.

In our conversations with network and infrastructure decision-makers, a few common themes come up repeatedly. They’re excited about the AI opportunity, but they’re also feeling overwhelmed. Many of them are worried that they may already be missing out on the latest and greatest AI innovations for networking. They’re also not sure what to do about that.

Despite this uncertainty, there’s one thing we know: AI has fundamentally changed enterprise networking, and that’s because it’s changed where data lives. A legacy hub-and-spoke architecture made sense back when the data primarily resided at the hub. Now, the data is everywhere. Enterprises need distributed infrastructure to reach data wherever it is and move it wherever it needs to go.

Let’s take a closer look at how Fortune 500 infrastructure leaders are adapting to this new AI landscape. Read on for four scenarios that are representative of our conversations with enterprise leaders, and what they might mean for your own AI readiness.

Scenario #1: Optimizing for flexibility

What we’ve heard:

We’re always working to ensure users have access to the newest, shiniest AI models at all times, only to find out that the “flavor of the week” has changed yet again. We struggle to keep up with this demand. We never know if we’re set up to support new models from different providers, with the right guardrails in place.

What it means:

When you don’t know what your future AI needs will look like, the best you can do is plan for infrastructure flexibility, so that you’ll be ready for whatever comes next. This is one reason that enterprises should use both private and public models hosted across private and public cloud environments. This is essential to avoid lock-in to a specific model or provider.

When you use a hybrid infrastructure with both public and private environments, it’s easier to balance performance, flexibility and security. If you build your hybrid AI infrastructure on a foundation that you control, you’ll be able to use vendor-neutral connectivity solutions that allow you to reconfigure or resize your cloud connections on demand.

Scenario #2: Getting to the root of the problem

What we’ve heard:

When our AI infrastructure isn’t working properly, it’s hard to know exactly how to respond. The problem often ends up being something completely different than what we originally thought. When we use AI tools to troubleshoot our infrastructure, it’s not always as helpful as we thought it might be.

What it means:

Infrastructure leaders deal with this kind of uncertainty every day of their lives. In some ways, AI has made the problem a hundred times worse. AI can help with troubleshooting enterprise infrastructure, but human leaders need to be careful about how they interpret and apply AI-driven insights. Human expertise is vital to interpreting AI results.

The challenge is that generative AI models are people-pleasers. When you’re trying to solve an infrastructure problem, the model will always give you what sounds like a plausible solution. You shouldn’t take these responses at face value, even when they align with your preconceived notions.

It’s up to you to determine whether the solution really makes sense, based on what you know about your infrastructure. You also need to carefully formulate prompts to give models the full context of the problems you’re trying to solve.

Scenario #3: Addressing shadow AI

What we’ve heard:

For years, we’ve had a “Wild West” mentality to AI. Even though we have a governance team that oversees AI policy, enforcing their guidelines is essentially impossible. People could do whatever they want, and that’s the thing that keeps us up at night.

What it means:

Users are accustomed to trying different AI models in their personal lives, so it’s no surprise when that mentality carries over into their work lives. They’re using the models they want to use, when and where they want, without asking for permission or waiting for guidance. This so-called “shadow AI” can lead to higher costs, introduce data privacy issues, and make it difficult to ensure consistent results.

Whoever oversees AI policy within an organization, one thing is clear: The network is the foundation that enables AI innovation while limiting risk. Because networking leaders have visibility into where AI workloads run and how data moves between them, they deserve a seat at the table when the time comes to formulate an end-to-end AI infrastructure strategy.

The organization should also set clear expectations around how network engineers can use AI in a responsible manner, and ensure they have tools available that align with those expectations.

Also, if network engineers are using AI to write code, they should do so as part of a collaborative development pipeline, rather than leaving the code sitting on their own laptops. This gives actual developers the chance to weigh in and address any issues they see in the code.

Scenario #4: Connecting with AI ecosystems quickly

What we’ve heard:

Recently, our executives decided that we need to acquire GPUs from a particular neocloud provider. We’ve never connected with this neocloud before, so we’re unsure about how to proceed. In fact, we’ve primarily worked with a specific cloud hyperscaler for years now, so connecting to anything outside their ecosystem feels new and weird. How do we make the right connections, quickly and securely?

What it means:

Ensuring access to the right ecosystem partners is an essential part of infrastructure flexibility. As mentioned previously, neutral connectivity solutions are ideal for AI, because they make it quick to connect with new partners whenever the need arises, without vendor lock-in standing in the way.

Even with the best possible connectivity, physical proximity still matters. Distance inevitably leads to latency. That’s why choosing who your ecosystem partners are is only half the battle. It’s equally important to determine where those ecosystem partners are, and then place your AI workloads accordingly.

Deploying private AI infrastructure at a global colocation provider can help. You can feel confident knowing that thousands of potential AI ecosystem partners are already deployed with that same provider. For instance, the Equinix ecosystem is home to neoclouds like CoreWeave, Crusoe, Denvr Dataworks, GroqCloud, Lambda Labs and Nebius, as well as all major cloud hyperscalers.

Get started with distributed AI infrastructure today

AI is not a panacea to all the infrastructure challenges facing enterprise IT teams. In fact, it often exacerbates those problems. But teams that focus on optimizing their infrastructure flexibility today will position themselves to overcome the challenges of AI, all while capitalizing on new AI-driven opportunities as they arise.

Equinix recognizes the key role the network plays in enabling AI-ready infrastructure. It starts with having interconnection hubs in all the right places—not just in a few core locations like a traditional WAN would. With distributed infrastructure at the edge, enterprises can get close to all the different places where they:

• Gather data from various sources
• Perform AI inference in near-real time
• Connect with various end users and partners

Learn more about why distributed interconnection hubs are so important to enterprise leaders trying to navigate AI readiness. Read the use case brief, “Optimize your global network for AI.”