What Equinix actually launched
Equinix has opened Enhanced Performance Innovation Centers, branded EPIC, at four of its data centers in Frankfurt, London, Osaka, and Paris. These are production-ready environments where enterprise customers can run liquid-cooled GPU and CPU platforms under real operating conditions, rather than relying on vendor demos or paper specifications to decide whether a given hardware configuration will actually work in their environment. The program builds directly on Equinix's 2024 Solution Validation Centers, extending that earlier testing concept specifically into liquid cooling, which has moved from a specialty requirement to a near-default expectation for high-density AI hardware.
The timing lines up with Equinix's own infrastructure moves. In September 2026 the company announced an inference platform built in partnership with Nvidia and Together AI, and EPIC reads as the natural next step: having built closer ties to the chip ecosystem, Equinix is now giving its colocation customers a lower-risk way to adopt the hardware that ecosystem produces. Equinix runs 282 data centers worldwide, so starting EPIC at just four locations signals a deliberate, controlled rollout rather than a blanket capability available everywhere at once.
Why liquid cooling stopped being optional
Equinix SVP Brian Stein put the company's position plainly, calling liquid cooling an essential component of modern enterprise infrastructure and a requirement for customers deploying AI. That is a stronger claim than most colocation providers made even a year or two ago, when liquid cooling was often framed as a premium option for the highest-density racks only. The shift reflects where GPU power density has landed: current and next-generation AI accelerators generate enough heat that air cooling alone is no longer a practical default for serious AI infrastructure, regardless of budget.
For enterprise buyers, that framing changes the decision from whether to invest in liquid cooling capability to how and where to validate it before committing capital. A test environment that behaves like a real production deployment, rather than a vendor showcase, is the difference between discovering a thermal or power integration problem during a pilot and discovering it after signing a multi-year colocation contract and installing racks. EPIC is explicitly positioned to move that discovery earlier in the buying process.
The build versus buy calculation this changes
Any enterprise considering liquid-cooled AI infrastructure faces a real build versus buy question: invest in in-house liquid cooling expertise and test facilities, or rely on a colocation partner's validation environment. EPIC pushes that calculation toward the latter for most organizations, because building an internal equivalent of a production-ready liquid cooling test environment requires capital, specialized facilities staff, and lead time that most enterprise IT organizations do not have sitting idle. Renting access to Equinix's validation infrastructure is a cheaper, faster way to de-risk a hardware decision than building that capability from scratch.
The tradeoff is that this ties the enterprise more closely to Equinix's own hardware ecosystem and its Nvidia and Together AI partnerships, which may not perfectly match every customer's eventual vendor choice. An organization planning to deploy AMD accelerators, for instance, should confirm EPIC's test environments actually cover that hardware path rather than assuming Nvidia-centric validation generalizes. The program reduces risk on the cooling and integration side, but it does not eliminate the need to verify that the specific hardware being tested matches what will actually be purchased and deployed.
What a four-site rollout means for availability
Starting EPIC at only four of Equinix's 282 global data centers means geographic coverage will be a real constraint for the near term. Enterprises headquartered or operating primarily outside Frankfurt, London, Osaka, and Paris will need to factor in travel, remote testing logistics, or simply waiting for Equinix's 2027 expansion before they can use EPIC directly. That is a meaningful planning consideration for any procurement timeline that assumes validation can happen on short notice close to home.
The sequencing also suggests Equinix is treating EPIC as a capability it wants to prove out carefully before scaling, rather than rushing a global rollout to capture demand immediately. That is a reasonable approach for a program tied to expensive, specialized liquid cooling infrastructure, but it means enterprises with urgent AI infrastructure decisions on the table this year may not get access in time to influence those specific purchases, and should plan their 2027 roadmap with EPIC's broader footprint in mind instead.
The roadmap implication for enterprise infrastructure teams
For any CIO or infrastructure lead weighing a move into liquid-cooled AI hardware, EPIC is a signal that the industry now expects validation before deployment, not after. Treat the availability of this kind of neutral testing ground as a factor in vendor and colocation selection going forward, the same way reference architectures and proof-of-concept labs became standard steps in enterprise networking and storage purchases in earlier technology cycles. Colocation providers without an equivalent validation offering should be asked directly how they expect customers to de-risk liquid cooling decisions without one.
The broader lesson is about sequencing capital commitments correctly. Liquid cooling infrastructure is expensive to install and difficult to retrofit, so any organization moving toward high-density AI deployments should build a validation step into its procurement process even if that step happens at a partner's facility rather than its own. Equinix just made that step easier to take for a subset of customers near its four EPIC sites, and the rest of the market should expect more colocation providers to follow with similar offerings over the next year.
