A Control Layer for the AI Era
Digital Realty on June 17 announced the availability of ServiceFabric MCP, a move that quietly reframes what a data center company sells. Rather than pitching square footage and megawatts, the operator is now offering programmable control over its physical estate through the Model Context Protocol, the emerging open standard that lets AI systems and agents interact with infrastructure through standardized interfaces. The launch extends ServiceFabric, the company's global interconnection platform, with controls designed specifically for Private AI environments where enterprises run sensitive workloads on dedicated hardware rather than in shared public clouds.
The thesis behind the product is blunt. Digital Realty argues that the next era of enterprise AI will be defined not by models but by the physical substrate underneath them: power density, advanced cooling, and sovereign placement. What changes with ServiceFabric MCP is that those physical attributes become programmable. An agent orchestrating a training run or an inference pipeline can now reach into the fabric to discover capacity, provision connectivity, and enforce policy, instead of waiting on a human to file a request and a technician to fulfill it days later.
What ServiceFabric MCP Actually Does
The capability is built on AI Private Exchange, or AIPx, the underlying architecture that carries Digital Realty's patented policy and orchestration technology. ServiceFabric MCP organizes its functionality into four areas. Design and provisioning lets agents stand up interconnections and environments. Discovery and telemetry exposes real time state about power, capacity, and connectivity. Identity and security governs who and what can act. Operations integration ties the fabric into existing enterprise tooling. Together these turn a sprawling physical network into something an automated system can query and command through a single, standardized protocol.
The reach is the headline. Digital Realty says the controls extend across more than 800 data centers, a figure that blends its own facilities with third party sites reachable through its platform. Its directly operated footprint spans more than 300 facilities across 55 plus metros in 30 plus countries on six continents. That scale matters because AI deployments rarely sit in one place. They straddle public cloud, neocloud GPU providers, on premises systems, and colocation, and the value of a programmable fabric grows with the number of endpoints it can stitch together.
The Voices Behind the Launch
Chris Sharp, Digital Realty's chief technology officer, kept the framing pragmatic. "Our strategy is simple: provide the foundational infrastructure enterprises need for sustained AI workloads, while enabling flexible scale as demand grows," he said. The emphasis on sustained workloads is deliberate. The early AI buildout was dominated by training, a bursty, capital intensive activity. The next wave is production inference, which runs continuously and rewards operators who can give customers fine grained, ongoing control rather than one time capacity allocations.
Customers echoed the point. T. Michael Thornton, chief executive of See All AI, said Digital Realty's Borton campus and ServiceFabric provide the high bandwidth, low latency connectivity required to support his company's NVIDIA DGX B200 environment. IDC research vice president Mary Johnston Turner offered the analyst view: "Enterprise adoption of Private AI infrastructure has reached an inflection point. Production AI workloads now demand control." That word, control, recurs throughout the announcement, and it is the crux of the pitch.
Why This Matters to Enterprise Buyers
For CIOs and infrastructure leaders, the significance is less about a single product and more about a shift in how data center capacity gets consumed. The colocation layer is becoming software defined. When physical placement, power, and interconnection can be driven by an API or an agent, the procurement model starts to resemble cloud, with its self service provisioning and policy driven governance, while retaining the isolation, residency, and density advantages that pull regulated and AI heavy workloads out of hyperscale public clouds in the first place.
There is also a sovereignty angle that enterprises should not miss. Digital Realty explicitly ties programmable control to sovereign placement, meaning agents can be constrained to provision only within approved jurisdictions. As data residency rules tighten across Europe and Asia, the ability to encode those constraints into the fabric, rather than relying on manual review, becomes a compliance feature as much as an operational one. ServiceFabric MCP is an early bet that the agentic future of infrastructure runs on open protocols, and Digital Realty wants its estate to be the place those agents reach first.
The Competitive Stakes
Digital Realty is not alone in chasing this territory. Equinix, its closest rival, has been building out its own distributed AI and programmable interconnection story, and neocloud providers are racing to make GPU capacity consumable through clean APIs. By embracing MCP specifically, an open standard rather than a proprietary interface, Digital Realty is betting that interoperability will win, and that enterprises will prefer a control plane that does not lock them into one vendor's tooling. That is a credible read of where the market is heading, even if the standard itself is young.
The open question is adoption. Programmable infrastructure only delivers value if enterprises actually wire their agents into it, and most organizations are still early in operationalizing agentic systems against production infrastructure. The launch is best understood as Digital Realty planting a flag ahead of demand, ensuring that when enterprises do hand provisioning decisions to agents, the largest interconnection fabric on the planet is already speaking their language. For a sector that has long sold space and power, selling programmability is a notable pivot, and one worth watching as the rest of the industry responds.
