By
James Deaton - Vice President of Network Services
Estimated reading time: 7 minutes
The AI conversation in research and higher education (R&E) is shifting. What was once “what chatbots and generative AI tools can we subscribe to?” is now refocused on “what infrastructure can we build and sustain together to ensure broader access and opportunity?”
That shift matters, and it’s one the Internet2 community is uniquely positioned to address. For decades, we’ve been building shared infrastructure to connect R&E institutions to the resources they need to fulfill their missions. Now, we’re applying that experience to support AI innovation.
Innovating to Support AI and Putting AI to Work
Last year, we teamed up with an Internet2 industry member to build a dedicated 138 Tbps AI research network that connects five U.S. data centers. It’s operationally separate from the Internet2 backbone and purpose-built for the distributed training and inference workloads that intensive AI research requires.
The engineering lessons from that design effort translate directly into how Internet2 can support AI research and instruction for the broader R&E community. That’s the promise of national hyperscale network connectivity built collectively: helping to close the access gap to AI resources across the U.S.
At the same time, AI is a capability we’re deploying inside our own operations and development efforts. We recently announced a new virtual front door for the Internet2 Network Operations Center, foreshadowing the AI operations capabilities being developed by Internet2 and others within the community. Additionally, AI coding tools are helping accelerate the delivery of a robust routing security toolset through the Routing Operations Observational Technology: Building to Enable Education and Research project, an NSF-funded collaboration between CAIDA and Internet2.
Internet2 also has major initiatives underway to bring AI-powered tools into how we detect and respond to network events, support cross-network troubleshooting, and deliver self-service capabilities that meet community expectations. What follows is a midyear update on where we stand: the intelligent capabilities we’re integrating into the network, the infrastructure refresh kicking off this year, and what it means for our team to participate in regional gatherings where meaningful discussions are setting the stage for community progress.
If there’s one thing I’d ask you to take away at the midyear mark, it’s that these aren’t separate efforts. The intelligence we’re adding to the network, the platform we’re refreshing beneath it, and the conversations we’re having across the community are a single strategy: shared infrastructure, built together, ready for what R&E needs next.
AI-Powered Networking and the Next Internet2 Console
When we outlined the Internet2 Trusted Infrastructure Program‘s objectives last year, advancing automation and AI-enabled operations was central to the community’s vision. These capabilities address a real operational pressure we see across R&E: the need to do much more with much less.
What the Internet2 Network Services team has been referring to as HAWAT, or Heuristic Analysis With Adaptive Troubleshooting, is that vision taking operational shape within the Internet2 network. It’s an intelligent operations system with AI agents that monitor the network 24/7, built to surface network issues faster and resolve them with minimal human intervention and escalation.
What excites me most about HAWAT is the shift in operational posture it represents. Today, a network event typically means an alert, a ticket, and an engineer starting from scratch. With HAWAT, agents have already gathered the diagnostics and correlated the telemetry, often with a proposed fix, by the time a person looks at it. The goal isn’t to remove people from operations; it’s to remove the toil. It’s designed from the outset for the multi-domain environment that R&E traffic actually traverses.
That multi-domain design is taking shape in our Agent-to-Agent (A2A) interoperability proof of concept: a HAWAT agent that can run diagnostics, like our Insight Console Looking Glass, and expose that capability to trusted agents at other networks. Along with our colleagues at SURF in the Netherlands, we’ve demonstrated cross-domain troubleshooting between agents within both the Internet2 and the SURF networks. Research traffic doesn’t stop at network boundaries, and agents that can securely query one another across domains point toward transatlantic issues being diagnosed end-to-end in minutes, not days.
Another core objective of the Trusted Infrastructure Program is to expand our network transparency and self-service capabilities. The next version of Internet2 Insight Console will give every Internet2-connected institution the ability to see, manage, and troubleshoot across the full Internet2 Network Services catalog.
We’re calling it Periscope, an open-source platform the community can inspect, extend, and improve. And it’s federated, meaning regional and international networks will be able to run their own instances and connect them, so visibility follows the path of the traffic. We’ve kicked off a pilot program, and regional networks such as the Utah Education and Telehealth Network and OSHEAN are running initial versions of the looking glass functions.
What does that mean in practice? Today, a researcher’s slow data transfer to an international collaborator can mean tickets with the campus, the regional network, Internet2, and an overseas partner, each with only partial visibility. With federated Periscope instances and agents that can talk to their counterparts, that whole path becomes visible and testable.
Evolving the Physical Infrastructure
The intelligent capabilities we’re building depend on a solid foundation. That’s why we’re embarking on an Internet2 Packet Platform refresh, with a request for proposals released earlier this month to select vendors. This next major evolution of the Internet2 network infrastructure centers around three core capabilities: 800 gigabit Ethernet for high-density capacity, Post Quantum Cryptography for top-tier line-rate encryption, and a highly scalable control plane supporting traffic engineering and Ethernet VPN (EVPN).
Community input will continue to define how the Internet2 network evolves. Representatives from Florida LambdaRail, KanREN, Link Oregon, Mid-Atlantic Crossroads, OneNet, OSHEAN, and the University of Nebraska are serving on the Community Packet Refresh Panel, evaluating how vendor offerings align with community needs. We’ll have more to share as the panel’s evaluation progresses and we move toward a selection later this year.
Meeting the Community Where They Are
Internet2 staff across the organization have been enthusiastically out in the community in 2026, participating in regional events such as the Merit Member Conference, Florida LambdaRail Annual Meeting, and UEN Tech Summit, as well as GÉANT’s TNC. My colleagues and I will continue that momentum at the Great Plains Network and ARE-ON Joint Conference from June 22-24, the Link Oregon Annual Member Meeting on June 23, and PEARC26 from July 26-30.
These are opportunities to share updates on new initiatives and to hear directly from the community about what’s working, what isn’t, and what’s next. There’s no substitute for being in the room. The hallway conversation about a stubborn performance problem, or the regional network engineer who tells us exactly where our tooling falls short, shapes our roadmaps with a deeper layer of nuance that complements everything else we hear from the community.
So much of what we do at Internet2 and in Network Services, specifically, is a collaborative effort centered on shared goals and guided by community input. Special interest groups (SIG) and working groups are where the priorities we’ve outlined in the Trusted Infrastructure Program get tested against community expertise and experience.
For the past few years, the Network Automation SIG has been doing substantive work on network automation tooling and orchestration best practices. The Performance SIG, relaunched last October, is developing inter-domain performance targets and troubleshooting guides for the whole community to use. The Time and Frequency Working Group is brand new, formed to address inter-domain precision time synchronization across R&E network infrastructure. Add to that the IPv6 Working Group and the Optical SIG, which round out a set of forums where community conversations continue to shape what we build next. If you’re not yet participating in one of these groups, I’d encourage you to reach out to networkdevelopment@internet2.edu to get involved.
Continue the Conversation at Technology Exchange
Lastly, I’m genuinely excited for the Peer-to-Peer Advanced Networking Unconference, which we’re hosting for the first time at the 2026 Internet2 Technology Exchange (TechEX26) this fall. The format is based on the premise that some of the most valuable conversations at any conference are the ones that can’t be planned in advance, with participants pitching topics and building the agenda together in real time. If you want to be in a room full of people grappling with the same challenges at the intersection of advanced networking, cloud, AI, automation, and security, this unconference is for you.
In fact, TechEX26 is where you can get the latest on all the efforts discussed here: HAWAT, Periscope, the Packet Platform Refresh, and more. Hear directly from the teams building these capabilities, share your feedback, and engage with the community in person. We hope to see you in Minneapolis.
ICYMI
About the Author(s)
James E. Deaton is the vice president of Network Services. James leads the strategic direction and ongoing operations of Internet2’s world-class network infrastructure and advanced services in support of research, education, and global collaboration.
James is a seasoned leader in the research and education community with nearly 30 years of experience driving strategic cyberinfrastructure initiatives across state, regional, and national networks.