The IOWN Global Forum will likely focus on datacenter interconnect use cases in the, to help diverse providers of AI infrastructure ply their trade.

The Forum develops and promotes Innovative Optical and Wireless Network technology, an effort to replace all wired networks with optical. IOWN’s backers hope to one day bring their tech all the way to the connections between dies on a chip, but for now have a high speed and low latency WAN technology that has demonstrated the ability to handle workloads like synchronous data replication over hundreds of kilometers of optical fiber that an IOWN-aligned carrier assembles into an all-photonic network.

The Forum held its annual meeting yesterday in Sydney, Australia, and during the event The Register met with steering committee chair Gonzalo Camarillo and Katsutoshi Itoh, who leads IOWN’s use case working group.

Both told us that recent consultations with potential users saw datacenter interconnect emerge as a use case that will likely drive IOWN adoption.

Camarillo said IOWN reps recently engaged with representatives of the financial services industry in London, and learned they see great potential for IOWN if it lets them use datacenters outside the city as such facilities will offer lower costs than those closer to town or in central business districts.

However, more distant datacenters will only be useful if latency remains low.

IOWN thinks hyperscalers will find their own ways to address that issue, but that smaller and newer datacenter operators – especially so-called “neoclouds” that offer hosted GPUs – will need fast links. IOWN wants to provide that link and feels its tech is fast enough that it can enable remote GPU access without creating a bottleneck – and perhaps enable access to more diverse sources of AI infrastructure.

The Forum also expects neoclouds will build many smaller datacenters in locations where land and energy is available, creating a need for interconnects between those bit barns.

Camarillo said recent board-level discussions at the Forum suggest IOWN will happily volunteer its tech as that interconnect.

The group also hopes to become an enabler of sovereign AI, suggesting that organizations keep data in their own infrastructure and use a fast all-photonic WAN to send it to a cloud or neocloud hosting AI accelerators. The cloud will never store the data other than to process it and send the results back to their customers over an IOWN network.

• Not all networks can handle AI traffic – and experts are sounding alarms
• Rent-a-GPU neoclouds need to adapt or die as the AI market evolves
• Open Compute taps IOWN to help design distributed datacenters and a 'computing continuum'
• The plan to make all networks optical is about to take two big steps forward

Another use case IOWN plans to target is remote content creation. Itoh used the example of a sporting event to explain why this matters because broadcasters typically deploy 30 or more cameras at sporting events and use an outside broadcast van to produce their broadcasts.

Conference venues typically appoint a single audio-visual company to handle events, and they are not shy of charging high fees. Itoh said the same happens at stadia. IOWN Forum wants sporting venues to instead adopt its fast WANs, so broadcasters can create a central production facility from which they can produce live broadcasts without needing a van on site. Itoh, whose day job is with Sony, admitted the company might profit from such an arrangement.

Beyond industry scenarios, IOWN imagines it can enable development of disaggregated datacenters that see different facilities devoted to hosting GPUs or CPUs. Fast IOWN connections between both can allow creation of wide-area composable compute clusters that use resources housed in multiple facilities without adding unworkable latency.

A who’s who of computing industry giants has backed IOWN, so these ideas aren’t entirely fanciful, although they do need carriers to get on board. Support from major networking vendors will help, too. ®

More about

• Fibre-optic
• Network

More like these