The Open Compute Project Foundation (OCP) is spearheading a radical redesign of data center power architecture to support AI's explosive growth, including the concept of "1 Megawatt racks" that could reduce energy losses from 40% to just 7%. This transformation would eventually move power supplies outside computing floors and potentially integrate on-site renewable energy generation, including small nuclear reactors, according to Chief Innovation Officer Cliff Grossner, Ph.D.
With hyperscale operators encountering unprecedented challenges in compute density, power distribution, and cooling, OCP's collaborative community of over 400 corporate members and 6,000 active engineers is developing open standards to address bottlenecks that threaten to constrain AI infrastructure growth.
Credit: OCP
"The industry is at an inflection point," said Grossner. "We no longer assume that all processors can be generic. We need to create custom processors for specific workloads."
Radical power architecture redesign
Power delivery has emerged as a critical constraint for AI infrastructure. Current power conversion processes result in approximately 40% energy loss when converting from high voltage to chip-level voltage—a significant inefficiency in facilities that increasingly consume hundreds of megawatts.
The OCP community is exploring radical redesigns of data center power architecture, including the concept of "1 Megawatt racks" that would move power supplies out of server racks into separate rack units. Eventually, power generation capabilities could move entirely outside the computing floor to become integrated with the data center facility.
"We need to change the design of data center facilities to be able to supply 400 or 800 (-400, +400) volt DC," Grossner said, describing how facilities will need to be re-architected as power systems evolve.
Transitioning from AC to DC power conversion and UPS functionality from inside the IT Rack to outside of the IT Rack to make room for more. (Credit: OCP)
The organization predicts future data centers will incorporate on-site renewable energy generation through solar, wind, and potentially even small nuclear reactors as part of an ecosystem approach to power management.
Environmental sustainability push
As data centers grow in scale and power consumption, projected to reach 3% of global electricity by 2030 according to the International Energy Agency, minimizing environmental impact has become a priority for the OCP community. The organization is developing metrics beyond traditional Power Usage Effectiveness (PUE) to address water consumption, IT equipment efficiency, and is also looking at heat reuse.
"A data center is not a standalone object in the future," Grossner said. "It's going to be part of an ecosystem that includes potentially a wind farm next to it, and proximity to buildings or other industrial environments that can use the heat produced."
OCP Community Projects
Credit: OCP
One surprising area of focus is concrete production, which Grossner identified as a significant source of carbon emissions in data center construction due to the cement contained in concrete, and the manufacture of cement is very carbon-intensive. The OCP community funded by hyperscale data center operators is investigating alternative concrete formulations that use less cement, making them less carbon-intensive.
The organization is also collaborating with Infrastructure Masons (iMasons) on standardized carbon accounting for IT equipment, intending to eventually label hardware with information about its manufacturing carbon footprint to drive purchasing decisions toward lower-carbon options.
While experiments like underwater data centers have shown limited viability, OCP members are considering underground facilities to minimize environmental disruption and enable heat reuse in urban settings.
Liquid cooling transition
As computational density increases with each generation of AI accelerators, thermal management requirements are forcing a transition from air to liquid cooling systems.
Credit: OCP
"The thermal management needs of current and next-generation silicon make it pretty clear that chips will be cooled either by single or two-phase liquid cooling," Grossner said. "Air cooling will remain, but its use is becoming increasingly constrained."
The OCP community is working to standardize liquid cooling delivery systems, including specifications for connectors, acceptable pressures and pressure drops within the system, fluid properties, and other requirements to ensure interoperability across vendors. This standardization is crucial for the industry to scale up liquid cooling deployments cost-effectively.
Silicon diversity through chiplets
The OCP has emerged as the "front door to the open chiplet economy," establishing a marketplace with over 25 suppliers offering modular semiconductor components and related services. This approach aims to transform chip design by enabling an ecosystem where specialized silicon can be combined for optimal performance.
"The next inflection point for the silicon supply chain is open, with innovation driven by a collaborative community, just as we've seen with the computing platforms of the cloud era," Grossner explained.
OCP open chiplet economy marketplace
Credit: OCP
This transition mirrors the industry's earlier shift from closed proprietary systems to open platforms. However, it requires rethinking semiconductor supply chains, with design decisions by chiplet creators impacting packaging, testing, verification, and software work performed by separate organizations.
The organization envisions distinct roles in this ecosystem: companies creating chiplets, businesses building ASICs using both external and internal chiplets, and firms providing support through design tools, test facilities, and professional services.
Since 2018, OCP has worked to standardize chiplet technologies, addressing the complexities of a fragmented supply chain. "Standardization typically triggers a large acceleration in the marketplace," Grossner noted, pointing to the OCP marketplace as a catalyst for this shift.
Optical communications revolution
Interconnect technologies have become a focal point as AI clusters require higher bandwidth and lower latency connections between compute nodes. The OCP community sees optical communications playing a crucial role, with a transition from copper to co-packaged optics enabling next-generation interconnects.
"AI clusters cannot scale without critical interconnect technologies," Grossner emphasized. Interconnect bandwidth and latency are especially important for providing resilience during AI training runs that can take days or weeks to complete.
Global expansion with APAC focus
With 30% of its membership in the Asia-Pacific region (primarily Taiwan), 50% from North America, and 20% from EMEA, the OCP is increasing its regional presence in Asia. The organization will hold its APAC Summit in Taipei on August 5-6, 2025, to better serve its Taiwanese membership base.
"Taiwan's semiconductor industry is a key partner," Grossner emphasized, highlighting OCP's efforts to integrate local ODMs and chipmakers into its ecosystem. The summit aims to showcase how OCP's trusted IP model can drive semiconductor innovation for AI.
AI: Predominant OCP data center use case
Credit: OCP
Education and adoption barriers
To support the adoption of its community-driven standardizations, the OCP recently launched the OCP Academy, offering training courses for both members and the broader industry. The marketplace provides access to hundreds of OCP-recognized products based on community-developed specifications.
Despite its success, OCP can still face challenges in promoting an open ecosystem. "People often ask if they have to give away their IP," Grossner said. "That's not true. Collaboration means solving shared problems, not sacrificing proprietary assets."
By demonstrating the value of collective innovation, OCP has attracted participation from major technology companies, including Nvidia, Intel, AMD, and ARM, and many others, alongside numerous ODMs implementing OCP specifications in server designs.
Market impact by revenue
Credit: OCP
Future technology focus
Looking ahead, OCP aims to remain the premier organization for AI infrastructure by focusing on three pillars: standardizing silicon, power, cooling, and interconnects; supporting complete system development; and providing education through technical workshops and the OCP Marketplace and Academy.
The organization is also seeding communities to explore future disruptions through its Future Technology Symposium and Future Technology Initiative. These include quantum computing technologies and cogeneration for data center facilities.
"The OCP intends to be the premier organization that everyone turns to for open infrastructure for AI," Grossner stated, positioning the foundation at the center of the industry's response to unprecedented computational demands.
As AI and HPC continue to redefine computing requirements, OCP's role in fostering the development of open, sustainable, and scalable infrastructure appears increasingly vital to the industry's ability to deliver on AI's transformative potential while managing its environmental impact.
Article edited by Jerry Chen
