BATIC 2025 Day Three: AI is Forcing a Network Revolution
Day Three of BATIC 2025 made clear what’s being acknowledged behind closed doors: AI isn’t just another workload riding on the network. It is beginning to shape the network itself, bending design priorities around its own gravitational pull. What once was the hidden, stable layer of the digital world, the pipes, the fibers, and the data centers are now the very bottlenecks AI is exposing.
During Day Three sessions, operators, vendors, and cloud players didn’t dance around it: the infrastructure we built for the internet age cannot support the demands of inference-heavy traffic, ultra-low latency, and continuous cyber-resilience. What’s coming is not a gentle upgrade cycle. It’s a break. A revolution in how networks are built, sold, and secured.
Legacy networks are not fit for purpose
Ragheed Al-Dabbagh, Head of Optical Networks Product Management and Network Architecture, APAC, at Nokia, framed it bluntly: “AI is the next transformational inflection point. It’s going to impact networks: the way we build them, the way we architect them, and the way we design them.”
That sounds abstract until you hear the details. Nokia is moving from C-band to stacked C/L bands, building multi-rail optical line systems, and preparing for hollow-core fiber that could cut latency by up to 40%.
But architecture change goes beyond optics. “Legacy meshed networks are not going to scale to interconnect AI-based data centers,” Al-Dabbagh said. “We may need to build an express overlay: A to Z, with the absolute lowest latency possible from a cable landing station to a DC.” Availability and resilience targets are being rewritten. “Availability won’t be good enough at 1+1, we’ll need 1+1+N to survive multiple concurrent fiber cuts,” he added.
Rodney Kinchington, Managing Director for Asia-Pacific, Japan, and Greater China at BT International, didn’t hedge either: “Legacy networks are not fit for purpose for AI. We’re decommissioning the old and building a brand-new global fabric to meet AI’s needs. This is a network revolution.”
Think about that metaphor he used: the automobile. When fast cars emerged a century ago, roads weren’t built to handle them. So we built motorways and autobahns. The same is happening here, except what’s moving isn’t people, it’s data.
Edge inference resets economics
If networks are being reshaped, so too are economics. What’s making this harder and stranger is that AI doesn’t just want more capacity. It wants it everywhere. Inference closer to the user promises both lower latency and lower cost.
Dmitri Domtchenko, Director of Network and Data Centre Strategy for Asia-Pacific and Japan at Akamai Technologies, said early pilots are eye-opening. Why? “Inference at the edge is cutting costs by about 80% in our pilots,” he revealed. “That’s huge for operators looking to balance performance and economics.”
That’s not incremental. That’s system-level economics flipping. It’s not just about shaving milliseconds. It's about re-shaping where workloads live and where capital gets spent. Suddenly, half-forgotten exchange sites are regaining strategic importance. Al-Dabbagh suggested they could become edge hubs, while Domtchenko pointed to the role of non-carrier-neutral facilities across Asia as hosts for AI deployments.
Security has to be built in, not bolted on
AI not only fuels new services; it’s also a tool for attackers that accelerates threats. Domtchenko warned: “We’ve seen a 43% increase in cyberattacks across networks in the last year. AI makes it cheaper and more accessible for attackers.”
Wilson Liu, who heads the CDN & Cloud Security Specialist Team at Tencent Cloud, put it even more starkly and said the only answer is embedding security everywhere: “People can do good things with AI. People can do bad things. We’re building an AI-powered CDN, with cyber defense built in and intelligence at the edge. Think about security from the very beginning and do not wait until something happens.”
This isn’t the old model of patch and pray. The consensus is that security must be baked into the system, from the optical fiber itself to content delivery networks and application layers. Paul Baptist, RVP Partner Success at Salesforce, tied it to governance and trust: “Trust has been our core value for 25 years, so we build preventative guardrails, active monitoring, and self-learning into the platform.”
From RFPs to partnerships
Technology is only half the shift. The commercial model is breaking too. Al-Dabbagh put it plainly: “Hyperscalers are not as patient as operators. If they need capacity, they need it tomorrow, not in four weeks.”
That kills off the lengthy RFP cycle. The world of six-month procurement, locked contracts, and slow deployments doesn’t match AI’s clock speed. “Customers want flexibility to scale up or down instantly,” BT’s Kinchington explained. “That’s why network-as-a-service is essential. Procurement teams aren’t fully ready, but we’re educating them with new models, true-ups, and quarterly resets so they can manage within budgets but still move fast.”
The RFP replacement is partnerships, pre-built capacity, and risk-sharing. This is a cultural change as much as a technical one. Telcos that can’t shift from a transactional to a collaborative approach will find themselves left behind.
Beyond connectivity
If networks are being rearchitected, business models are evolving as well. There’s the deeper question: what business are telcos in? Andrijanto Muljono, Director & Chief Enterprise and Strategic Relationship Officer at XL Smart, offered an answer highlighting the need to move up the stack: “Our strategy is to go beyond connectivity. That’s where differentiation and growth are. Collaboration, collaboration, collaboration. This is how we scale new services and monetization.”
Telkom Indonesia’s Director of Digital IT, Faizal Rochmad Djoemadi, was just as clear: “We apply AI on both sides—our own operations and our customers. If you just sell connectivity without value on top, customers won’t buy.”
Norioki Sekiguchi, Global Business Executive at SoftBank and CEO of S&BTS Global, went further, mandating an AI-first culture: “All employees must use AI every day. Like smartphones in 2007. No reason needed, just use it.”
GPU-as-a-Service in Indonesia is already live. The first deployment by Lintasarta, using Nokia’s technology, is fully consumed. It shows how quickly demand can outstrip supply.
Governance and the agent problem
As AI systems talk to each other, governance becomes murkier. Salesforce’s Baptist flagged a coming crisis, mentioning a new consortium working on agent interoperability: “Now a Salesforce agent may need to interact with an Akamai agent, which then calls on a Tencent agent. We need frameworks to ensure governance.”
Their solution is zero-copy: leave data sovereign where it is, act on it without moving it. It builds trust but also drives more network traffic. It’s a governance model and a business model rolled into one.
The talent choke point
And then there’s talent. Every speaker hit the same theme: we don’t have enough AI engineers, operators, and data scientists. “We’ve empowered over 1.2 million Indonesians with basic AI skills,” said Dharma Simorangkir, Microsoft’s Indonesia head. “We work with communities and universities because training people in a vacuum doesn’t work.”
Telkom Indonesia is trying free certifications, university partnerships, internships, and nine AI Centers of Excellence nationwide. “Talent is number one,” Djoemadi admitted. “We cannot serve government requirements if we only have eight AI experts when 78 are needed.”
This isn’t a side issue. Without people to design, run, and govern these systems, none of the architecture or partnerships matter.
Power, capacity, and carbon
Even if the talents show up, the power might not. Al-Dabbagh said global data center demand could triple from 60 GW today to 220 GW by 2030. Domtchenko warned of a “real shortage of quality data centers near metros.”
Sustainability targets hang over it all. Akamai has pledged carbon neutrality by 2030. “That seems very problematic to achieve in the current environment,” Domtchenko admitted. “We seem to have a real shortage of quality data centers near metros. Sustainability targets, like carbon neutrality by 2030, are very problematic to achieve.”
AI as constraint, AI as catalyst
The themes from Day Three were clear. AI isn’t just changing what we do with networks. It’s changing what networks are. It forces architecture to bend, economics to shift, security to become structural, procurement to collapse into partnerships, and business models to rise above connectivity. It is pulling telcos beyond connectivity into curated marketplaces, vertical AI services and sovereign solutions.
It reveals shortages. Talent is scarce. Power and sustainability are hard limits. Governance and standards are still forming. But it shows, in Indonesia’s rapid experiments, how quickly new models can take hold when demand is there.
Discussions repeatedly circled back to where it is playing out fastest: Indonesia. From Telin’s subsea build-out to Telkom’s AI platforms, from Lintasarta’s GPUaaS to SoftBank’s AI-first mandate, the country is positioning itself as a testbed for the region.
The winners will be the telcos that act like system architects, not just bandwidth providers: re-designing networks, embedding trust, and co-creating with partners. Those who just keep turning up the bandwidth will watch the value flow past them.
AI has become the constraint. And constraints, history shows, are where revolutions begin.