Reimagining Supercomputing with the Distributed Bring-Your-Own-Resources Model

As society becomes more reliant on large-scale supercomputing centers for compute and storage, the costs associated with them have become a topic of discussion. One innovative hypercluster model—a distributed one—provides an alternative approach.

An example of this is the CENIC AI Resource (CENIC AIR), the California portion of the National Research Platform (NRP), which uses a “bring-your-own-resources” (BYOR) model.  CENIC members among California’s research and education institutions contribute processing and storage nodes in their campus machine rooms and racks throughout the state. These nodes are already connected to each other through the campuses’ California Research and Education Network (CalREN) connections and are labeled as available for use by other CENIC members participating in CENIC AIR.

By implementing this distributed model, the enormous challenge of creating, operating, and maintaining a hypercluster suitable for bleeding-edge research and education is turned into a set of many smaller, less intrusive, and more manageable challenges, with CENIC as the central enabling hub.

Moreover, this distributed model offers real solutions to many of the space, power, expense, and overhead challenges associated with large supercomputing centers, as well as with research funding. It also simplifies grant writing, perhaps just as important for a harried researcher or educator.

Aggregating Unused Rack Space

With the decommissioning of legacy computing systems and the migration of administrative functions to the cloud, empty rack space has become common in many campus machine rooms. Also, campus border router racks often have empty space where additional equipment can be placed.

The BYOR model takes all of these already-networked empty spaces—large and small—throughout California and aggregates them. In effect, this model turns one large infrastructure problem into many small, easily managed ones, and then brings those small contributions back together.

Making Grant Writing Easier While Focusing on Research and Education

A comment made by San Diego Supercomputer Center Director Frank Wuerthwein during his presentation at the 2026 CENIC conference reveals two major advantages of the BYOR model.

“The strategy,” Wuerthwein explained, “is to centralize what gets cheaper with scale and to decentralize when costs scale linearly.”

In other words, at the scale at which institutions purchase equipment, two processors or drives will cost twice as much as one, so each institution should buy its own equipment. However, operational, security, and maintenance costs decrease with scale and are best aggregated into a single centralized operations team.

This model also aligns beautifully with most grant programs, which are likelier to cover equipment purchases, as opposed to the ongoing operational costs of maintenance and security—i.e., participation in a BYOR hypercluster gives you convenient access to exactly the services a grant will not cover.

It also frees faculty and students to focus on research and education as opposed to tending equipment, to say nothing of reassuring a campus IT department worried about how safe a researcher’s multi-GPU setup is when maintained by someone who is not a cybersecurity expert.

Avoiding Overprovisioning and Informing Your Equipment Choice

Another major advantage is that an institution’s or researcher’s equipment needs may be lessened overall in this model, given that they need not purchase all the computer hardware that would otherwise be needed to handle their institution’s usage at peak scale. Typically, researchers can only estimate peak usage and then purchase equipment sufficient for that, which then often sits idle. However, BYOR hyperclusters can absorb bursts of peak usage and enable a researcher to purchase only what they will need rather than what they might need.

An individual campus or researcher can also save money while still purchasing higher-end equipment since they will need a smaller amount of it. Given that CalREN peers with most major cloud providers, participants can even use their commercial cloud credits at times of heavy use.

Moreover, the BYOR model helps inform equipment decisions. Connecting to CENIC AIR or NRP does not require the contribution of compute/storage nodes for the use of others, although institutions throughout the country do so. Thus, a campus or faculty member can “taste-test” from the wide variety of available compute and storage resources and decide exactly what they need before making their own purchase, which they may then choose to make available to other participants to help absorb their periods of peak use.

“Bring Your Own Brain” to a Community of Experience

Connecting to CENIC AIR and NRP also means joining a large member-created, member-owned, and member-led community rather than building expertise from scratch. CENIC was created by its California research and education members in 1996. We have been at the forefront of bespoke advanced networking and services for that community, which is capable and even eager to push any advanced network to its absolute limits.

As we are also governed by our member community, we offer a deep awareness of its needs and how to exceed them, no matter the economic, technical, or geographic obstacles. This is the level of knowledge that first created the Pacific Research Platform (PRP, which grew into NRP) and CENIC AIR, both built on the foundations of CalREN and the CENIC community.

Connecting to CENIC AIR and NRP opens the door to a national community of wisdom, not just equipment. Lessons learned may include everything from Science DMZ design, Kubernetes namespace setup, GPU scheduling, and more. These represent direct mentorship and collaboration opportunities between CENIC members, including live events, and CENIC AIR Champions at institutions that have already connected.

Best of all, participation in CENIC AIR (and NRP) already comes with CENIC membership and the implementation of a Science DMZ, which the CENIC engineering team can help architect and implement specifically for your campus. (For the purposes of CENIC AIR, a Science DMZ is a big-data network path that allows very large data flows to bypass the campus firewall while remaining secure, so research traffic does not overwhelm everyday campus traffic. You can learn more about how a Science DMZ achieves this at the CENIC Blog).

To learn how your institution can participate, please contact the CENIC Network Operations Center or your segment’s representative at the CENIC Program Management Office.

Related blog posts