0

Equinix is a global leader in digital infrastructure, colocation data services, interconnection solutions, and multi-tenant data centers. In addition to building and managing massive physical facilities, the company has 500,000 physical and virtual connections across its fleet of “carrier-neutral” data centers, which serve more than 10,500 total customers —60% of which are Fortune 500 companies; 4,900 of which are enterprise customers; and thousands more that are IT and cloud providers.  Below is my wide-ranging discussion with Phil Read, senior director of colocation strategy for Equinix.

When talking about how Equinix is changing around AI and the surge around capacity demand, Phil Read, Equinix’s senior director of colocation strategy, compares AI’s momentum to what happened in the cloud years ago, with enterprises working to find the right synergies around Capex savings and moving to an Opex model. But, he admits, the scale and speed in AI are  “mind-boggling” and making fundamentals all the more important in ensuring integrity in how Equinix customers operate.

“When you think about businesses that went to the cloud in the past and maybe came back through ‘repatriation,’ AI is about rebalancing to get the right workloads in the right places…and the right economies, the right controls, and avoiding vendor lock in,” says Read, noting that Equinix is all about connecting and interconnecting digital ecosystems, with a business that goes beyond the physical colocation of services down into the connective tissue that bind together disparate clouds, networks, and enterprise IT infrastructures in a cohesive fabric.

With 280 data centers within its fleet of IBXs (international business exchanges), Read describes some of the elements of Equinix’s success and the ways in which the company is evolving to support its customers, “wherever they are in their journey, as well as where they want to be.”

Standardizing innovative design and construction

To Read, getting customers to their desired destination means offering the “right infrastructure” and doing so in under a year. “We are pushing toward 100 locations with 90-day turnaround times.” To achieve that, he assures Equinix isn’t “wildly deviating” from traditional colocation models in terms of how they market and sell colocation and fixed infrastructure, but rather evolving toward converting traditional Capex hardware costs into Opex subscription-based digital infrastructure models.

“It’s an interesting challenge because we have to do two things at once – new builds, while also retrofitting,” which he explains sometimes means putting newer technology into older buildings that have well-established ecosystems, even if there are physical limits and constraints. “This is the era of scale, where it’s not so much about ‘can you do it?’ but more about ‘can you do it in a scalable, standardized, repeatable way with pricing that makes sense, and with a brand promise and value proposition that resonates with customers – and can you do it globally, within a highly connected platform?'”

To check off all those requirements, Equinix is focusing on standardization around not only the original generation of infrastructure, but the next generation, and even the one after that. For example, for the past few years, the company has made 100 of its facilities liquid-cooling enabled, with standardized liquid cooling designs that integrate direct-to-chip capabilities natively across its core data center infrastructure.

Rather than treating liquid cooling as a completely custom, one-off engineering project for each customer, Equinix is using standardized “Smart Build” design packages that allow customers to seamlessly deploy AI and HPC hardware using uniform architecture layouts. Read says it’s the shift to inference that accelerated the move from testing and planning liquid cooling to a full-scale global production rollout across scores of IBX data centers across metropolitan areas. “All of our new locations have a native liquid cooling capability,” which Read says include a “spectrum” of density requirements. “Not everyone is jumping from 0 to direct to chip, liquid cooled environments operating at 120 to 200 kVA per cabinet, so we have offerings for every step along the way,” such as a single cabinet or a large cage, a rear-door heat exchanger or an in-row cooler. “We can offer mid-range density and serve customers in the 20-50 kVA per cabinet range, all the way up to the high end with Nvidia and the newest Vera Rubin generation, and whatever they have after that.”

Read emphasizes that “It’s all about steering the workloads to the right places with the infrastructure to support, but without over-rotating and putting them in a place that isn’t advanced enough or that is overkill what they need.”

To meet the full spectrum of customer needs, Equinix continues to evolve its offerings. Its specialized colocation offering, “Liquid-to-Cage,” and “Standard Smart Build” or “Liquid Cooling Smart Build” are all architectures engineered to support high-density AI and HPC workloads through standardized facility integration of major hardware setups, with key components, such as:

• Direct-to-chip integration
• Hybrid capabilities to support advanced liquid-cooled and standard air-cooled hardware alongside one another
• Rear-door heat exchangers to neutralize heat before it leaves server chassis.

For mid-tier enterprises that do not require massive HPC power or liquid cooling, there are production-ready server cabinets with standard power (typically under 5 to 10 kW), providing immediate physical rack space for basic compute and storage arrays. There are also scaled-down, secure mesh cages starting around 20 square feet per cabinet positions for custom configuration in private physical server layouts.

To continue improving its offerings, Read says Equinix is not only making physical investments, but it is also building new partnerships (like that with Nvidia) to build out more robust capabilities. “We are doing what we have to in order to bring distributed AI architecture to our customers, so they have the privacy they need, and the scalability, and distribution of the model on a hybrid, multi-cloud architecture.”

When it comes to the customer, Read believes they will use the potential of today’s vast compute resources to do what’s “adjacent possible,” where one good thing will lead to another.  “There’s always risk and fear, as well as challenges, so it’s about being responsible as we go along that technology curve. The potential of these compute resources could vastly change what humans are capable of,” he said, noting that cancer research, fraud detection, and other important challenges could be resolved in a way that positively serves broader humanity. “We are proud of what our customers achieve outside of our four walls.”

To further enable its customers, Equinix will continue to seek out the next generation of partners and suppliers. “If we had done this interview 10 years ago, a lot of people would’ve been like “oh, I think I’ve heard of Nvidia,” he notes wryly. “The transformation means new entrants, like neoclouds, and massive Capex investment, such as the more than $1 trillion dollars in investment going into data center shell builds, with a lot of places being built as quickly as possible – some with liquid cooling, and some without.”

He believes neoclouds are a segment of AI infrastructure that will be important to making innovations a reality, solving latency, cost and structural bottlenecks that CPU-centric providers might struggle with. “Now, with the shift to inference, neoclouds are going to be an important part of what’s happening in the infrastructure space as well as the broader economy,” says Read, who looks to neoclouds with a lot of optimism and potential to drive AI innovations.

Partnerships form the ‘connective tissue’

In the most recent wave of innovation, Read looks to companies like Nvidia and AMD as very impactful to the data center landscape. “We have this core strategic pillar to build boulders to construct big data centers with these modern technologies in all the right places, and to support customers where they are and their growth ambitions to get to the places they need to be.” But connecting existing data and ecosystems with something like a Nvidia deployment requires a lot of what Read calls “connective tissue,” which are the “on ramps” over which public clouds can place their native physical hardware, with Equinix Fabric fostering connectivity among the different clouds.

Because Equinix has a very dense aggregation of digital ecosystems, it forms that connective tissue with a network of more than 2,000 telecom networks and ISPs; a huge network of IXPs; native public cloud on ramps and multi-cloud connectivity to all major cloud providers; and software-defined network automation through Equinix Fabric for instant API provisioning and on-demand pipelines.

“Our network services and interconnection team have incorporated AI intelligence into the Equinix Fabric and layer in a level of observability …it’s like putting a brain in the network to identify and manage either high latency and reroute or leverage unused network assets,” says Read. He believes the AI-enabled network is going to help Equinix take things a lot further: “When I think about the buildout of the colocation space to enable that networking capability, having that AI-enabled network will take customers to the next level of observability and management capabilities, on top of what’s already a very high-performing network.”

In that vein, Equinix has partnered with both Cisco and Nvidia in a core alliance working to build, automate, and secure hybrid, AI-enabled networks that move enterprise AI from experimental pilot phases into full, secure global production.

“The network is the AI control plane,” explaining that his network colleagues talk of “infrastructure as code” and the need for a software-defined control plane for all of their physical infrastructure. “We think about the network as the AI control plane, which has to be highly observable, highly capable technologies that align with the physical infrastructure so all things are in one place.”

Constructing at pace, but responsibly

With plans to build and grow to more than 3 GW in the next few years, Equinix promoting a strong sustainability posture, with a commitment to achieve 100% renewable energy coverage by 2030. “We are 96% of the way there,” says Read, adding, “we have to construct at pace, but do so responsibly.”

More than half of Equinix’s RFPs ask for detail about compliance with sustainability programs. “From a power perspective, we’ve operated in certain communities for more than two decades. We see ourselves as a partner to the community and that’s an important aspect to how we build and grow. We work with municipalities around how we think about power and our presence in that space as a tech provider so we remain a part of the community.”

To be efficient in how it builds and grows in different communities, Equinix works with not only local utilities, but a growing number of alternative energy providers and BTM technologies. “We have a GW of capacity under contract with Oklo in the SMR space, and Bloom microreactor technology,” which Read says gives customers certainty that Equinix can grow and support customers from the perspectives of site, space, power, and connectivity.

“From our perspective, it’s efficiency, we know that we drive the right sustainability programs for responsible use of that compute.”

Growing need for sovereignty

Because Equinix has always built around the concept of “neutrality,” the recent push for more sovereign solutions has become a differentiator. “ We don’t have exclusive partnerships or preferential treatment to one network over another,” posits Read, adding that it’s increasingly important to connect and route networks in compliance with sovereignty requirements.

He considers Equinix Fabric as a type of sovereignty-enforcement layer at the network level so that sensitive AI datasets never accidentally leave designated legal jurisdictions when in transit. Equinix Fabric’s Geo Zones virtualize the network perimeter, creating a “sovereign-by-design” backbone that logically blocks data from exiting a specific legal jurisdiction—even during automated cloud failovers or network rerouting events.

“This is a huge step for bringing AI to our enterprise customers in a useful and highly relevant way today,” says Read, emphasizing that it’s important to include an “intuitive customer experience” for managing critical infrastructure, with the network assets in place so customers can manage them the way they want to. “Some customers are very savvy and want to do everything themselves, and others prefer to use a partner, and others are seeking a whole managed services offering.” For that reason, Equinix continues to build out its platform to support distinct operational models that can be tailored to the technical maturity and internal resources of each customer.