Once again, there was a split in the approach, buy existing and, or build new in new markets, but now there was also a verified split in the research data to support it. I now have the address list of every existing neutral interconnection site with an IX in it, and I also have a list of every city without an IX. This was valuable information. What to do?
Prior to the re-visiting of my MMR research I had already been pursuing the creation of a strategy to develop distributed, neutral MMR’s in all of the places where they did not exist, but had a market that would support one. In 2017 I was introduced to Connected Nation, a 501c3 non-profit dedicated to rural broadband and closing the Digital Divide.
In 2018 Connected Nation was contracted by the Governor of Iowa to “solve the broadband problem” in the Grant Wood Area Education Agency (AEA) between Iowa City and Cedar Rapids. Connected Nation asked me to help them analyze the problem and come up with a solution. After reviewing how schools use the e-rate Federal subsidy program to buy Internet access the issue became clear. The schools all buy Internet access from their respective school buildings. This keeps them locked in to a local loop and denies them any ability to benefit from economies of scale. This is a very old issue that I had addressed long ago within 60 Hudson St, so I applied the same logic, but across a much larger geographic area.
There are two primary e-rate eligible services, WAN Access, which as a service is known in the wholesale industry as transport, and transit, which is public Internet access. We asked if the schools could bid out WAN Access circuits to an address that they don’t own. We had to ask the governing authority of e-rate, the Universal Service Administration Company (USAC), if this was permissible. After some research they came back and told us that they could find no prohibition to what we asked. So, that is what we did.
Instead of each school bidding out 100 Mbps of transit to each school building they bid WAN Access circuits to one common address that we chose in Coralville, Iowa and aggregated their Internet transit bid to 10 Gbps. We had three major wholesale ISPs all bid $.35 / Mpbs. At the time the schools were paying on average $3.85 / Mbps. As a result of this approach, we succeeded in reducing the cost of Internet transit for the schools by 90%+.
From there we discussed how we could work together to scale this concept. Two key elements were to own the real estate and introduce an IX. The reasons were the same as before. Control and the ultimate reduction of latency, not just the reduction of the cost of Internet access.
By combining the Connected Nation research data on the rate per megabit each K-12 school district pays for Internet transit in the U.S., along with cities with a major State university and a minimum population of 150,000 people, and my data showing where IX’s were not, a map appeared. These were the top 125 cities in the U.S. that lacked a local IX, but had the potential to support one.
In 2022, we formed a Joint Venture, Connected Nation Internet Exchange Points, LLC (CN IXP) and pursued the NTIA Middle Mile Grant program to fund our first five sites. We were not successful with that particular program, but one of the five sites was Wichita State University (WSU). Subsequent to the NTIA program, the State of Kansas created a specific $5MM grant for an IXP building and IX switch to be built in Wichita on the campus of WSU. CN IXP applied for and was awarded the grant. We broke ground in May 2025 and are on track for a Spring 2026 project completion and Ready for Service.
The future of IXPs and AI interconnection will be a combination of the growth of the existing IXPs, through the growth of the IX’s within them and the need for more space and power to support them, as well as development of new IXP facilities in new markets. There are unique challenges with each path that naturally exist, but both paths now have to contend with the power required for the GPUs that drive real-time inference AI and the low latency that is required for those applications to perform optimally.
Within the major metros there is currently a push for GPUs to find their way in to support:
- Real-time inference AI latency requirements of sub-3 milliseconds for certain applications, such as fraud detection by major banks
- Hyperscale requirements for inference and agentic AI that are at 25-50 megawatts per node
- GPUaaS for latency specific RTDs to specific ASNs within specific metros for real-time inference AI applications
With new market developments there are all of the above requirements, albeit not exactly at the same scale yet as more populated metros, but with all of the usual, unique challenges with bringing networks to a neutral interconnection point that does not yet exist. In addition, there is the new issue of dealing with NIMBY pushback related to new "data center" developments due to the negative perception of power consumption and costs to consumers as well as land use, environmental impact and more. IXP’s are not necessarily data centers, so this has to be managed very closely.
