Podcast

Ep 139 – Can telco build an AI grid? (NVIDIA’s Kanika Atri)

This week’s guest

Telco in 20 Guest Kanika Atri NVIDIA View Linkedin profile

Kanika Atri

Senior Director, Telecoms NVIDIA

AI is reshaping telco. At NVIDIA’s GTC conference, AT&T, T-Mobile, Comcast, and Spectrum announced they’re building AI grids, turning millions of existing cell sites and central offices into distributed AI compute infrastructure. NVIDIA’s survey of 1,000 telco professionals found more than 90% reporting that AI is already delivering cost savings or revenue growth, with the biggest ROI coming from agentic AI for autonomous networks.

For this episode, I’m joined by Kanika Atri, Senior Director of Telecoms at NVIDIA. We dig into where real AI ROI lives in telco, why the network may be the world’s most underutilized compute asset, and the big unanswered question: who’s going to build the software that makes all of this work?

Listen now to hear:

  • How NVIDIA’s survey reveals where AI is already paying off [03:06];
  • Why telcos’ existing infrastructure could become the backbone of distributed AI [05:06];
  • The critical software gap nobody in the industry owns [11:19]; and
  • Why countries are turning to operators, not hyperscalers, to build their national AI [12:27].

Links and resources

Wanna talk AI and public cloud? Telco execs, set up a meeting with our team to learn how to tap the immense business value it can bring.

Guest bio

Kanika Atri is senior director of telecom products and marketing at NVIDIA, driving AI adoption across the telecom ecosystem. She works closely with hundreds of customers, developers, and partners to embed AI into networks, operations, and services—accelerating the shift to AI‑native wireless networks, autonomous agentic-AI-powered operations, sovereign AI factories, and 6G‑ready infrastructure. Through these engagements, she shapes product roadmaps, go‑to‑market strategies, and commercial partnerships that translate advanced AI capabilities into measurable business impact for the telecom industry. 

Follow DR

The Telco in 20 podcast is ranked in the top 5% of all podcasts globally by Listen Notes! 🎉 We’ve also won 2024 and 2025 MarCom Awards, 2024 and 2025 Hermes Creative Awards, and are recognized as a TeckNexus Top 12 Telco and Tech Podcast, Forrester Top 100 Channel Podcast and Feedspot Top 10 Telecom Podcast.

If you enjoy the podcast, would you leave us a review? It takes you seconds to do in your app and it really makes a difference in helping to convince hard-to-get guests. And I love reading your feedback and reviews!

Podcast credits

  • Executive Producer and Host: Danielle Rios, TelcoDR
  • Senior Producer: Lindsay Grubb, TillCo Media
  • Senior Editor/Brand Manager: Alisa Jenkins, Springboard Marketing
  • Audio Editor: Andrew Condell
  • Supervising Producer: Amanda Avery
  • Associate Producer: Kriselda Dionisio
  • Music: Dyami Wilson

Most popular podcasts

  1. Ep 108 – ☕ Wake up and smell the BSS with Ray Le Maistre from TelecomTV ☕
  2. Ep 107 – 🧣Wrapping your head around Responsible AI🧣 (Ferry Grijpink)
  3. Ep 105 – NVIDIA’s vision for AI and the RAN (Chris Penrose)
  4. Ep 100 – The SPECIAL 100th episode of Telco in 20

Follow now

Get my FREE newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.

Wanna talk public cloud?

Set up a meeting with our team to learn how to tap the immense business value it can bring.

Get in touch

More episodes from Telco in 20

Telco in 20 Guest Chris Wade Itential
Ep 138 – From programmable networks to agentic AI (Chris Wade)

When AI agents start operating your network, the real challenge isn’t the technology—it’s building enough trust to let go.

Telco in 20 Guest Charles Fitzgerald Platformonomics
Ep 137 –  Hyperscalers’ $2T bet (Charles Fitzgerald)

Charles Fitzgerald returns with his tenth annual Follow the CAPEX report—and its implications for every major decision in telco.

MWC26 Wrap Up
Ep 136 – MWC26 Wrap Up (Mobile World Live)

DR joins the Mobile World Live podcast to unpack the biggest takeaways from MWC26.

Frequently Asked Questions

1. What is an AI grid, and why are telcos building them?

An AI grid is a distributed AI compute architecture that transforms existing telco infrastructure—cell sites, central offices, and switching centers—into a unified intelligence layer capable of running AI workloads. Rather than routing everything to centralized hyperscaler data centers, an AI grid routes workloads to the most appropriate location: latency-sensitive tasks go to cell sites, token-intensive ones go to larger facilities. AT&T, T-Mobile, Comcast, and Spectrum have all announced they’re building AI grids. Read NVIDIA’s Kanika Atri’s blog on AI grids for more details.

2. What does NVIDIA’s survey of 1,000 telco professionals reveal about AI ROI?

Kanika Atri, Senior Director of Telecoms at NVIDIA, highlights the key findings from NVIDIA’s State of AI in Telecommunications report: more than 90% of respondents said AI is already delivering cost savings or revenue growth, and nearly everyone reported improved employee productivity. The biggest ROI is coming from generative and agentic AI applied to autonomous network operations. The survey signals that telcos have moved past the experimentation phase—they’re actively investing.

3. Why are telcos the natural choice for building sovereign AI infrastructure?

Telcos already operate sovereign, in-country infrastructure and have deep, trusted relationships with governments and enterprises,  something hyperscalers can’t always claim. That’s why 24 operators worldwide are now building sovereign AI factories with NVIDIA. Indosat in Indonesia is a standout example: it has built the full five-layer AI stack, including locally developed LLMs that support 270 million Indonesian speakers and 20+ AI applications spanning healthcare, education, and agriculture. Read more about Indosat’s sovereign AI stack and SoftBank’s AI-RAN breakthrough in Japan.

4. What does it mean for 6G networks to be “AI-native” and “software-defined”?

Unlike 5G, which was designed before AI existed, 6G is being built in the AI era. AI-native means the network is designed from the ground up to serve AI-driven applications, including physical AI like robots and drones. Software-defined means breaking the traditional 10-year hardware cycle and enabling continuous updates as cloud software does. A major coalition including BT, Deutsche Telekom, Ericsson, and Nokia is aligned on building 6G on these principles. NVIDIA’s $1 billion investment in Nokia to build CUDA-accelerated 5G software supports this direction. Watch Jensen Huang and Nokia CEO Justin Hotard explain the partnership.

5. Who is responsible for building the software that makes AI grids actually work?

This is the biggest open question in telco AI, and Danielle Rios calls it out directly. Telcos buy and operate technology; they don’t build software. Ericsson and Nokia are primarily hardware companies. NVIDIA’s answer is open-sourcing parts of the stack to invite new players in. But as DR points out, telcos have historically struggled to trust unproven software vendors. The software layer is the missing piece of the five-layer cake, and until someone credibly owns it, the grand vision of distributed AI infrastructure remains out of reach.

6. How is Totogi approaching the software gap in telco AI?

DR’s takeaway from the episode is clear: the future of telco runs on software, not hardware—and that’s exactly the bet Totogi is making. The Totogi Ontology gives Tier-1 operators the telecom-specific context that AI agents need to work at scale. Without that software foundation, operators can’t turn AI grid infrastructure into actual business results. Read the Appledore Research report on the Totogi Ontology to understand why context—not just data—is what AI agents need to operate in telco environments.