- A startup is rethinking the CDN model for the AI era, with inference workloads—not video—at the edge
- Zero Latency’s new beta platform is already attracting attention from major telecom and infrastructure players
- The bet: future demand from robots, drones and enterprises that need low-latency AI closer to users
Michael Huerta readily admits the industry’s attempts at Edge 1.0 and 2.0 “whiffed at every turn.” But this time, he says, things are different.
Huerta is CEO of Zero Latency, a two-year-old startup that is building what he described as a content delivery network (CDN) for AI inferencing. Think what Akamai and Cloudflare do for media, only geared toward non-deterministic agentic flows that are on the horizon.
“If you believe that content won’t be Mike Tyson fights but it will be inferencing API calls or other types of commands that have to go to the edge, then probably the biggest consumer of content for CDNs will be machines, will be AI,” Huerta said.
That’s the bet Zero Latency is making and it’s building the infrastructure to prepare for that future. The company is both building physical infrastructure with its own facilities and GPU hardware as well as a software-based pipeline that will be able to dynamically pool and allocate compute resources to the inferencing workloads that flow through.
The software pipeline is what’s currently in beta trials with partners.
Zero Latency currently has three edge facilities online today in California and Florida and is aiming to build six more this year across Northern California, Texas, Illinois and the northeastern US. Like Akamai, Zero Latency is leveraging Nvidia’s AI Grid reference architecture.
While its footprint is small compared to someone like Akamai, which as of November had inferencing cloud compute available in 17 cities, Zero Latency is already drawing interest from some big-name partners. AT&T, Crown Castle, Zayo, Intel, Nvidia and Red Hat are notably listed among its customers and partners on its website.
“What we’re seeing with our partnerships with some of the guys that are named on the website and elsewhere is ‘how do I deliver point-to-point inferencing content? I don’t necessarily need to own it or host it on my real estate, but if I own fiber I’d love to make more money on the fiber,’” Huerta said.
Who are its target customers?
Huerta said Zero Latency is planning to sell to the companies behind physical AI robots and drones — a market that could be worth $5 trillion on the humanoid side alone by 2050, according to Morgan Stanley — but also to “anybody who needs to coordinate an AI stream at the edge.” That could be anyone from the Nokias and Ericssons of the world to a tower operator trying to serve tenant needs. It could also be the telcos themselves or even folks like Akamai who are already involved in the CDN ecosystem.
Pressed on what the use case might be beyond yet-to-be-realized physical AI proliferation, Huerta pointed to “bursting,” which is a sudden increase for inferencing capacity that can be served by edge locations when hyperscale availability is delayed. The need to serve enterprises with data gravity concerns is another, he said.
“I think the hardest part is you know where the puck is going, but it seems to always go a little slower than you might expect. We all know the AI age is here. But I think that I’d like to see it go further faster,” Huerta concluded.