In just four years AI has transformed the data center world from the ground up, with unprecedented compute needs driving new technology from chip to grid, with power densities on track to reach and then surpass 1MW in a single rack. The OCP Community has dramatically expanded, as hundreds of new companies from diverse industries have joined the OCP to tackle the challenges of deploying these technologies at scale and around the world.
The Emergence of the AI Data Center
To respond to this AI-driven transformation, the OCP Foundation, responding to urgent Community needs, announced in October of 2025 a new and broader vision to deliver on all aspects of AI data centers, supplementing the OCP Community’s ongoing efforts to standardize industry-wide and open common elements within the Cloud and AI data hall. The expanding of the Foundation vision included an enhanced focus on data center facilities, to cover all aspects of the data center from chip to grid.
To realize this expanded focus on the complete AI data center, a new Open Data Center for AI Strategic Initiative (SI) was launched, to increase efforts on key data center facilities challenges: power, cooling, mechanical, and management telemetry.
The addition of this new strategic initiative was, in part, a response to the open-letter call for collaboration initiated by Google, Meta and Microsoft and since then, they have been joined by many others.
This expanded vision does not alter the OCP Foundation’s mission to take innovations born in the hyperscale data center and cascade them to other market segments. Hyperscale is where new innovations are born and proven, to the benefit of everyone else.
A Year of Tremendous Progress for the OCP Community
We have had a tremendous year of progress on developing standards and best practices, both for AI hyperscale and regional data centers, and covering both the white and the grey space. Together we have secured a clear path to 1MW racks in power and cooling, advanced the state of the art in scale up networking, re-defined the data center as an integral and dynamic part of the grid, and nurtured OCP’s standards-based chiplet ecosystem.
Starting with Facilities we have had numerous contributions
(1) specifications covering a potential roadmap for open data center facilities for AI (2) a white paper providing an overview of proposed direct current power distribution architectures that covers current plans for 800VDC distribution and the evolution to 1,500 VDC (3) Community-developed guidelines for safe design, operation, and maintenance of energy storage systems, (4) new methods for gathering telemetry to enable data center real time IT and OT management, (5) a requirements specification for sold-state transformers that convert medium voltage to 800VDC, and (6) a specification for disturbance performance between hyperscale data centers and the utility grid that addresses voltage ride-through requirements, active-power recovery, and facility coordination.
To strengthen the OCP Community’s ability to deliver within data center facilities we launched a new OCP Energy Project focused on both energy generation and storage. Looking to the future, we also launched a Future Technologies Initiative Workstream covering high-temperature superconducting power infrastructure for power distribution.
Within the data hall we have seen the Community focus on key issues delivering contributions covering (1) building AI Clusters providing reference architectures for building air-cooled and liquid-cooled (in progress) AI Clusters, developed from Open PoD Groups with the OCP-published version 1.0 Deschutes CDU and guidance for TCS liquids, (2) new designs for Open Rack (ORW and HPR) doubling its width and growing its power profile, and (3) using Ethernet for AI scale-up networking (ESUN and SUE-T) , (4) Mult-path Reliable Connection (MRC) providing a method for reliable connections over Ethernet(5) a foundational specification for developing open chiplet ecosystems and (6) Bunch of Wires (BoW) Flexi 1.0 for non-high performance applications such as automotive.
The OCP Foundation and Community have not been idle in the hyperscale-adjacent space preparing for the next wave of AI, inference. Having (1) launched a new project, the AI Computing Continuum, (2) exploring new use cases, and (3) with a specification for a server targeting edge deployments.
To add new dimensions to OCP’s ecosystem in support of our new vision, we established new alliances with (1) EPRI and Current/OS for data center to power grid interaction, and (2) IOWN Global Forum to address hyperscale adjacent environments .
OCP Academy now holds a very large library of AI-related courses, covering many of the specifications and whitepapers just mentioned. Since its formal launch one year ago, it has become a very important resource for the OCP Community with 30 course series from over 100 modules. Over 6,000 learners already have registered and collectively have logged over 30,000 learner hours.
A Peak at the Future
Turning our attention to the future, the OCP Foundation and Community remains focused on its Open Data Center Ecosystem Vision, bringing its collaboration model to all aspects of the market from grid to chip.
The support for this is very strong, and is reflected in the new impact study just completed by IDC, which forecasts that for 2030, over $890B will change hands between buyer and seller on IT equipment that is based upon OCP recognized designs.
This is a dramatic increase from last year’s forecast for 2029 of $295B. This is due to two major factors: (1) the market itself grew, and (2) more importantly the penetration of OCP within the market increased dramatically to over 50%.
Key focus areas for the OCP Foundation as we look to 2027 and beyond include (1) standardizations to support AI silicon diversity, (2) development of co-optimized data centers, and (3) delivering Inference as AI moves to hyperscale adjacent locations.
AI Silicon Diversity
AI workloads are scaling faster within the hyperscale data center, and becoming more diverse as AI moves to hyperscale-adjacent deployments, demanding more computational performance. To meet this demand, the use of specialized computational architectures has multiplied to overcome the limitations of general-purpose compute and AI accelerators on specific workloads.
Thus, the race is on for hyperscale operators, neocloud providers, and many startups to develop custom AI silicon as a market differentiator. We see many examples of this emerging in the marketplace, where even within one hyperscale operator there are several different AI accelerators in use.
This diversity in silicon is creating many bespoke design bottlenecks in on-chip power distribution, thermal management, memory architectures, and chiplet interconnect technologies. This is compounded by the complexities in advanced packaging needed to assemble these chips into a running system, while hitting reliability targets.
To move the market forward for all, reducing complexity and cost, and allowing many companies to innovate rapidly, industry-wide and open common practices in co-optimized design and advanced packaging will need to be developed.
A special focus is needed on translating system-level requirements into lower layer silicon technology and design requirements, which is often ignored in today’s silicon development cycle, causing expensive rework to be done as defects are found only at system- level test and validation. A prime example is warpage when mounting a chip on a circuit board, causing failures that could potentially be avoided during design.
This will become even more essential as co-packaged optics is deployed at scale, with additional complexities of including the optical engine with the chip introducing new optical coupling, electrical, thermal and mechanical requirements.
The OCP Foundation is placing effort on building an enhanced semiconductor ecosystem within the OCP Community and working to launch new projects that can help solve this problem.
Co-Optimized AI Data Centers
The need for co-optimization does not stop with developing specialized silicon for different AI workloads. In fact, this impacts the entire data center from chip to grid. The silicon used for the AI accelerators impacts the design of a data center from the circuit boards, the servers, in-rack power distribution and cooling, the rack itself, the cooling and power systems within the data hall, the facility systems servicing the data hall, and the interaction with the power grid.
Put simply: it is one large chip to grid design, deployment and operations problem.
Within the data hall, and the rack, the industry-wide and open common elements for optical interconnects are still mostly undefined; for example, the transition to two-phase direct to chip cooling is looming, the busway and layout designs for 800VDC distribution to the rack are still incomplete, and designs allowing for automation of rack movement in the data center is still to be finalized.
Building high performant and resilient AI clusters with xPUs connected across racks in the data hall, or white space, is still as much an art as a solved engineering problem. We have much work to complete on specifications and best practices for assembling AI clusters at scale for both air and liquid cooled systems.
Looking at the “grey space”, in accommodating different AI Clusters, the data center needs to balance between the complete flexibility of custom designs and standardizing comment elements, creating reusable modules that allow for scale.
Areas that need attention include standardized power conversion architectures to deliver 800VDC and eventually 1,500 VDC to the data hall, common practices on deploying energy storage systems will be critical to stabilizing the power profile of the data center as it connects to the electrical grid, and understanding the best alternatives for on-site energy generation is still an unsolved problem.
Coming back to the chip to grid mission of the OCP, we are only at the beginning of the journey to having end-to-end management of the data center bridging the current silos of the white and grey space allowing AI-capable management platforms.
Hyperscale Adjacent AI
The industry is transitioning rapidly to Agentic AI and Inference as the primary drivers of growth. AI at scale will move outside only a handful of centralized data centers to include smaller scale deployments at the regional neoclouds, colocation, on-premises enterprise data centers, telco points of presence, and industrial sites, where even a 100kW rack is unlikely to deploy.
With computational resources in many locations, new distributed computing use cases are also emerging, where AI workloads operate across locations that include shared remote GPU farms with data maintained on-premises for data sovereignty.
The goal within the OCP Community is to define open, modular infrastructure components that can work across these diverse environments, making sure the infrastructure connecting that entire distributed computing continuum is open and consistent, so a workload can scale from a hyperscale AI cluster to an edge node without being re-architected each time.
Our alliance with the IOWN Global Forum supports this broader vision. The alliance brings complementary organizations together: IOWN brings the communications architecture — photonics and all-optical networking to deliver the bandwidth and low latency a distributed system needs. OCP drives the open hardware specifications: accelerated compute servers, power, and cooling that can span the continuum.
A part of the broader OCP Foundation mission has always been and still remains to take innovations born in the hyperscale data center and cascade them to other market segments that will never build a gigawatt campus, but still need that class of capability in their operating environment. Hyperscale is where new innovations are born and proven; the distributed AI computing continuum will be where they get delivered to everyone else.
By standardizing industry-wide and open common core platform building blocks, bridging the gap between hyperscale and enterprise needs, and addressing real-world constraints in power, thermal, and physical footprints, the AI Computing Continuum initiative supports repeatable, cost-effective deployment of advanced AI compute systems wherever they are needed outside of the centralized hyperscale data centers
A Continually Expanding Community
The collaborative and intellectual property-safe model of the OCP Foundation will serve its Community well over the coming years, as we work through the challenges of building AI data centers and deploying AI out to the end user. We are already laying the groundwork for the data center to integrate future disruptions such as our expanding community in quantum computing and new FTI initiatives such as High Temperature Superconducting Power Infrastructure. We have already shown that the collaborative model of the OCP Foundation applies, extending the OCP Open Data Center Ecosystem Vision to enclose Quantum. This is especially clear when we understand that Quantum Computing is not likely to be deployed stand-alone, but as an adjunct to classical computing in the data center.