Indoor 5G Coverage: Malaysia’s Billion-Dollar Blind Spot

• Malaysia’s agentic AI revolution faces a important infrastructure gap.
• Agents increase consumption from 33GB to 60GB per month by 2030, networks require redesign.

While Malaysia celebrates achieving 82% 5G outdoor coverage, a blind spot threatens the nation’s artificial intelligence ambitions. Agentic AI – where autonomous software agents handle complex tasks on behalf of users – demands robust indoor 5G infrastructure, yet 80% of Malaysian data traffic occurs indoors, where coverage remains patchy.

According to Gayan Koralage, Director of Malaysia Business and Director of Group Strategy at Edotco Group, this disconnect represents “the real battleground for AI adoption.”

The indoor coverage gap

Malaysia’s 5G rollout has progressed rapidly by outdoor coverage metrics. Digital Nasional Berhad (DNB), the government-owned wholesale 5G network operator, has achieved 82% outdoor population coverage – a significant milestone. The recent allocation of an additional 100 MHz spectrum (giving DNB both F0 and F1 in the 3.5 GHz 5G band) has strengthened the business case and improved rollout capacity and efficiency.

Edotco recently solidified its relationship with DNB through a five-year extension of its tenancy contract. Yet both DNB and U Mobile, the second network operator with approximately 3,500 sites, face the same fundamental challenge: limited indoor 5G rollout means outdoor beaming into buildings simply doesn’t work.

“The outside-in beaming doesn’t provide the consistent distribution needed for high-density, low-latency applications.” The average Malaysian cellular customer consumes 33GB monthly.

Yet 80% of this consumption happens indoors, creating a fundamental mismatch between where coverage is deployed and where it’s needed. Malaysia has approximately 1,000 in-building systems currently deployed, most converted to 4G in the early 2000s, but 5G upgrades remain limited.

The Agentic AI catalyst

The urgency around indoor 5G infrastructure intensifies when considering the trajectory of agentic AI adoption. Unlike generative AI, which creates content on demand, agentic AI refers to autonomous software agents that execute complex, multi-step tasks independently – from managing banking transactions via chat interfaces to coordinating health monitoring, entertainment, and transportation needs.

“The real end user of agentic AI is the individual, not necessarily the enterprise,” Koralage said. “You and I will have agents that convert text to commands for banking apps, healthcare applications, and entertainment. The agents will talk to each other and perform repetitive tasks on our behalf – that’s where the largest data growth will happen.”

By 2030, Koralage projects that average data consumption will nearly double from 33GB to 60GB per customer monthly, with the incremental 27GB driven almost entirely by machine-to-machine communication as AI agents operate continuously.

This shift challenges existing network architecture designed for human use patterns with distinct peak and off-peak hours. “AI agents don’t sleep and don’t commute to offices,” Koralage said. “They operate 24/7, which means networks will always be at peak capacity. The concept of off-peak hours disappears entirely.”

Rethinking network architecture for always-on AI

Meeting agentic AI demands requires shifts in multiple dimensions. Traditional networks prioritise downlink capacity – streaming content to users – but AI agents generate substantial uplink data through sensors, video streams, and telemetry. Networks must evolve toward symmetric bandwidth allocation.

More importantly, the current centralised cloud architecture proves inadequate for AI’s latency requirements. Processing data by routing it overseas to major data centres creates both performance bottlenecks and sovereignty concerns.

“When you go through multiple layers – access layer, transport layer via land and sea cables, then data centres – it takes longer and costs more,” Koralage explained. “The solution is deploying edge computing at tower aggregation points. This way, Malaysian data stays in Malaysia, processing happens locally, and the long transport route gets eliminated.”

Edge deployment also addresses energy consumption concerns. Rather than powering massive centralised facilities, distributed edge computing at cellular towers enables localised processing where AI demand concentrates – major cities and high-density venues.

The 20,000 tenancy challenge

Translating these requirements into infrastructure buildout reveals the scale of investment needed. Malaysia currently operates 60,000 tower tenancies, with 12,000 converted to 5G and 48,000 remaining on 4G.

Koralage estimates an additional 20,000 to 30,000 tenancies will be required to achieve near-universal coverage, upgrade in-building systems, and meet the government’s minimum speed mandate of 10 Mbps for 4G networks by January 2026.

The indoor component remains particularly challenging. While building owners increasingly recognise internet connectivity as essential infrastructure – similar to water and electricity – economic viability varies significantly.

Upgrading a major venue like a shopping mall can cost RM500,000 or more, creating questions around cost recovery when multiple operators share the same infrastructure. “We’re seeing a positive shift where Malaysian building owners behave more like those in Singapore and Australia, not necessarily demanding extra rental for 5G upgrades,” Koralage said. “Out of about 100 new sites this year, only 10 requested additional payments. The Internet is increasingly viewed as a basic building amenity.”

Yet only 15-20% of Malaysian data traffic currently runs on 5G networks despite 82% outdoor coverage availability. Device limitations, coverage gaps in the 3.5 GHz spectrum’s smaller footprint compared to 4G’s broader frequencies, and call quality issues during handoffs between technologies contribute to slow migration.

The data sovereignty question

As AI agents manage increasingly intimate aspects of daily life – from health monitoring to financial transactions – data sovereignty emerges as a important challenge. Current AI processing routes Malaysian user data through overseas servers, raising questions about long-term storage and access.

“The intimate data about our health, behaviour, and preferences – where should it be stored? In Okinawa or somewhere in Johor?” Koralage asked. Edge computing offers partial solutions through local processing, but comprehensive frameworks around data residency remain underdeveloped in Malaysian policy.

The path forward

Malaysia’s rapid 5G deployment represents a genuine achievement, moving from zero to 82% coverage faster than regional peers. Yet the agentic AI wave exposes important gaps between outdoor-focused metrics and indoor-heavy reality.

Success requires acknowledging that 5G infrastructure serves as a means, not an end. The true measure isn’t coverage percentages but whether networks can support AI agents operating continuously in hospitals, campuses, and malls, managing thousands of simultaneous users.

As Koralage says, “AI adoption isn’t just about generative content – it’s about creating digital versions of ourselves that work on our behalf. That requires network infrastructure designed for machines, not just humans. Malaysia’s AI revolution will be won or lost indoors.”

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.