Introduction
0:03 [Music]
0:26 Hi again. For this session, I’m joined by David Kiparos, the Principal AI Architect at Red Hat. David, thanks so much for taking part in the Telco AI Forum.
0:34 David: You’re welcome. Happy to be here.

⸻

What is the Optimal Adoption Pathway for Gen AI?

0:39 Moderator: Let’s kick off with the headlining question. What does an optimal adoption pathway for Gen AI look like for CSPs, and where should they start?
0:52 David: In general, it’s a very difficult pathway for CSPs. AI is a fast-moving object in the tech world, which makes it challenging. Use cases from just a year ago are already outdated. CSPs are getting smarter about applying predictive and generative AI, with some even building their own models—engaging in tasks like continued pre-training.

1:35 In mapping adoption, it’s about mapping for impact. There are many science projects CSPs can pursue, but the real question is: how do you make meaningful impact?

1:58 I’d break it into three large categories:
• Internal productivity for employees
• Operational efficiency for lifecycle management
• Monetization through new services and platform capabilities

⸻

Gen AI ROI for Telcos

2:22 Moderator: Let’s dig into ROI. Which Gen AI use cases are delivering measurable ROI for telcos today, and why?
2:43 David: What works today is keeping core infrastructure and services intact, then layering Gen AI on top to enhance value.

3:05 Example: Tier 1 telcos offer enterprise/government APIs into their core platforms. Gen AI can make those APIs easier to consume—accelerating services into production without disrupting existing systems.

⸻

When to Use Gen AI

3:43 Moderator: With the hype cycle, operators might overuse Gen AI. How should they decide where and when to apply it—across network ops, customer care, or product innovation?
4:06 David: The lines will be blurry. But again—map for impact. Start with low-hanging fruit where value is proven.

4:27 Customer care/call centers are the clearest case. Hyperscalers make it easy to add agentic AI into call flows. But telcos also need hybrid setups—keeping PII on-prem for security.

5:14 Large frontier models are easy to use but expensive. Fine-tuned smaller models, trained on telco-specific data, can be cheaper, faster, and sometimes perform better.

⸻

How Red Hat Guides Telcos

5:52 Moderator: Given the complexity, telcos need strong partnerships. How does Red Hat guide telcos in building the right Gen AI infrastructure?
6:10 David: Telcos rely on NEPs (Ericsson, Nokia, Mavenir, Fujitsu, Samsung) and accelerators (Nvidia, Intel). They face the burden of integrating multiple vendors.

6:57 Red Hat’s strength is providing a partner-enabled, AI-enabled platform. We already have relationships with these vendors. Telcos can plug in their chosen accelerators/NEPs, and Red Hat helps integrate into a cohesive solution (APIs, monetization, edge services).

⸻

KPIs for Gen AI Performance

7:56 Moderator: Beyond ROI, what KPIs should telcos prioritize?
8:21 David:
• Low-level KPIs: time-to-first-token, throughput, API performance.
• Strategic KPIs: value delivered to end customers.

9:41 Ultimately, success is whether telcos’ AI services drive their customers’ revenue and make telco services “sticky” vs. cloud or in-house alternatives. Integration with 5G core, RAN, and edge offerings is critical.

⸻

Open Source Small Language Models

11:09 Moderator: What is the value of open-source small language models for telcos?
11:15 David: Innovation in AI is happening in open source/Linux. Red Hat leverages this strength with:
• Red Hat OpenShift AI for scalable MLOps
• RHEL AI for small dev workloads

These work seamlessly across dev-to-scale environments. Technologies like LangChain, Llama Stack, MCP, etc., fit well on top of Red Hat’s platform.

⸻

Readiness Factors

13:05 Moderator: What readiness factors—skills, data—are needed before implementation?
13:28 David: Two big ones:
1. People – Empower employees with safe AI tools to boost productivity. It’s about enabling, not replacing.
2. Data – Current telco data architectures (old data lakes, ETL) aren’t Gen AI–friendly. We need new ways to make data usable and productive for employees and AI.

⸻

Gen AI as a Service

15:22 Moderator: What does Gen AI as a Service mean for telcos, and how can they balance customization vs. scale?
15:41 David: Many telcos consider GPU-as-a-service at the edge. But the value lies in integration with telco’s core services (e.g., IoT, bandwidth on demand, healthcare). Otherwise, enterprises will just buy cloud GPUs.

17:02 Telcos are uniquely positioned to keep AI workloads/data in-country, addressing data sovereignty concerns. This is a big opportunity for them to pivot and deliver secure, trusted AI services locally.

⸻

Closing
17:49 Moderator: Perfect place to end—Gen AI is an opportunity for telcos. Thank you, David.
17:56 David: You’re welcome, thanks for inviting me.
18:04 [Music]