- Cirion moved Grokstream's Grok AIOps from pilot to production across 105,000 km of fiber in seven Latin American countries in six months
- In Cirion's testing the platform cut 90% to 92% of alarm noise, letting engineers act on root causes instead of thousands of alerts
- CTIO Paul Choiseul rates Cirion Level 3 on TM Forum's five-level autonomy scale and targets Level 4 and a self-healing network
Cirion Technologies, the Miami-based digital infrastructure operator that runs one of Latin America's largest fiber and subsea cable networks, has put an agentic AI platform into production to run its network operations, betting that machine-driven alarm correlation can hold service quality steady as its traffic and complexity grow.
The company is using Grok, an AIOps platform from Prosper, Texas-based vendor Grokstream (not affiliated with the similarly named Grok chatbot from Elon Musk's xAI), as "the intelligence layer that sits across the operational ecosystem," Cirion's CTIO Paul Choiseul said.
The software ingests telemetry from every piece of equipment on Cirion's network — from the transmission gear carrying terabits across submarine cables to the Metro Ethernet elements delivering enterprise services — and collapses the resulting flood of alerts into a short list of actionable incidents.
From dashboards to correlation
The problem Cirion is seeking to solve is familiar in network operations: traditional monitoring throws off far more data than humans can process. A single fiber cut affecting a large area can generate thousands of alarms, and network operations center staff can drown in the noise, Choiseul said.
"It doesn't matter how good your network operations folks are, their ability to parse through thousands of alerts and alarms that go through these platforms on a daily basis, it's overwhelming," Choiseul said. "So when you do that, you end up with a situation that turns you into more of a reactive player. You're going to be at the mercy of having to wait until a customer calls you and tells you that they're sensing a problem."
Adding more monitoring tools does not solve that, he said — it makes it worse. Hard-coded filters catch known problems but miss novel combinations of faults.
"A dashboard that's been hard coded is unable to tell you about a new thing or a new combination of factors that are basically working together to take your network out," Choiseul said. Grok, by contrast, clusters and correlates events using machine learning and Cirion's own knowledge of its circuits and their locations, aiming to surface the single root cause behind a wave of symptoms.
In Cirion's testing, the effect was steep. The company fed the platform six months of historical alarm data and saw what Choiseul described as "about 90, 92% compression" — meaning a network operations center (NOC) that once had to triage 100 alerts could focus on roughly eight. "It's impactful day one," he said. Dealing with eight alerts rather than 100 means frees staff to give a customer a clear diagnosis rather than a promise to investigate.
What the numbers do and don't show
For Choiseul, the Grokstream result was no surprise, because he'd seen it before. He joined Cirion in January from American Tower, where he held a similar role, and he says he deployed Grokstream twice in previous positions, including work in Nigeria and South Africa where he saw NOC efficiency gains "somewhere in the 50% range."
The previous experience allowed Cirion to move quickly from pilot to production. "That's the great thing about having done it before. I've done it twice before, so this was pretty much perfunctory," Choiseul said.
Brought in to move Cirion's transformation quickly, Choiseul said he deliberately reached for a known quantity. "The requirement was to steer us in the right direction as quickly as possible. So the best way to do that is to work with what you already know works," he said.
Cirion is not yet evaluating the open-weight or sovereign AI models drawing industry attention; that, Choiseul said, is for later. "Once you've been able to get to the operational effectiveness that you're looking for, then you can start getting creative and start looking at some of these platforms and start developing your own agents."
The predictive claim
"In a matter of six months, we've gone from being reactive to being proactive to being predictive," Choiseul said. He draws a distinction between proactive operations — detecting a fault and dispatching a crew before the customer notices — and predictive operations, in which the system flags the conditions that precede a failure. He offered the example of a transmission unit that overheats and takes down part of the network: before it fails, cooling fans may throw early warnings about reduced airflow.
Cirion is also building a two-way integration with its trouble-ticketing system so that Grok learns from the notes engineers enter when they resolve an incident, and can later tell operators not only which element to fix but which vendor they called last time — Infinera, in one example Choiseul cited.
The endgame Cirion describes is a self-healing network, which Choiseul frames as "a journey rather than a destination" that moves through visibility, correlation and automation.
On TM Forum's five-level autonomous networks scale, he places Cirion at Level 3 today and is targeting Level 4. That puts Cirion roughly where much of the industry sits: TM Forum has reported that operators are only now beginning to cross into Level 4 on live networks.
A crowded race toward autonomous operations
Cirion is far from alone. Its own former parent, Lumen Technologies — which sold its Latin American business to become Cirion in 2022 — is deploying agentic AI across its network through Blue Planet, using AI agents for data migration and network domains.
Zayo's CTO describes a similar strategy of mining minor alarms for predictive maintenance. Verizon says AI now autonomously resolves 85% of network problems before customers see them.
But Cirion's network presents a special challenge. Latin America is "one of the most dynamic and complex digital markets in the world," spanning multiple countries, regulators, languages and network architectures, Choiseul said. He cited a six-to-seven-month effort just to meet new Brazilian billing requirements as an example of the operational drag that comes with the region's fragmentation — the kind of complexity that, he contends, makes a correlation engine more valuable than another dashboard.
Cirion's footprint
Cirion operates roughly 105,000 km of fiber and a portfolio of subsea cables that form much of the backbone into and out of Latin America. Its submarine systems land on the U.S. East Coast and in California, and it owns a cross-Andes route between Chile and Argentina that Choiseul likens to "a Panama Canal of sorts, cable-wise," linking the Atlantic and Pacific. In late 2024 the company announced a new Atlantic subsea route, SAC-2, tied to Google's Firmina system, and it has continued expanding metro networks and data centers across the region. Its customers include hyperscalers, large enterprises, carriers and government agencies that need high-capacity connectivity — multi-gigabit up to 100- and 400-gigabit links — between South America and the U.S.
The company's competitive position is that of a large but not dominant regional player. It competes against far bigger balance sheets, including América Móvil, which holds a larger owned submarine cable, and Telefónica's infrastructure arm Telxius, along with neutral-host operators such as Ufinet. Cirion is privately held as a portfolio company of infrastructure investor Stonepeak, which acquired Lumen's Latin American operations for $2.7 billion in cash in a deal that closed in August 2022, rebranding the business as Cirion and retaining its executive team. Choiseul declined to discuss revenue.
On return on investment, Choiseul pointed to softer but operationally meaningful gains rather than a headline cost figure: reduced mean time to repair, fewer engineers "chasing noise alarms," and — the metric he stressed most — lower customer churn, since missed SLAs driven by slow repairs are a measurable driver of defection. "The biggest benefits come from improved reliability, better operational focus, faster decision making," he said.
A vendor win, and a bet on familiarity
For Grokstream, a small vendor competing against much larger AIOps and OSS players, the Cirion deployment is a marquee reference customer across an entire region. "Cirion is setting the standard for what modern, AI-ready operations look like in Latin America," Grokstream CEO Casey Kindiger said in the announcement.
For Cirion, the deployment is less a leap into frontier AI than a calculated, low-risk move: a proven tool, installed fast by a leader who had run the same playbook before, aimed squarely at a real operational problem. Whether it carries Cirion from Level 3 to the self-healing network Choiseul describes will depend on execution over years, not the six months it took to get to production. The compression numbers are promising; the autonomous-operations vision, for now, remains a roadmap the rest of the industry is chasing too.
Read more on Fierce Network about autonomous networks
Blue Planet accelerates Lumen’s agentic AI adoption