AI workloads cannot run on legacy fibre”: Interview with STL’s Dr Badri Gomatam

As artificial intelligence (AI) workloads scale from pilots to production, data centre infrastructure is being pushed to its limits, with rack densities surging, latency tolerances tightening and the physical layer emerging as a decisive competitive differentiator. In an interview with tele.net, Dr. Badri Gomatam, Chief Technology Officer, STL, discussed how the company is positioning itself at the intersection of connectivity, compute and cloud, the advantages of its next-gen fibre technologies for AI interconnects, and his outlook for India’s data centre growth over the next two to three years. Edited excerpts…

How are enterprise and hyperscaler requirements for data centre infrastructure evolving? How is the relationship between telecom networks and data centres changing as AI workloads scale?

The requirements have shifted from storing data to processing intelligence at extreme speeds. According to JLL, AI workloads are set to account for 25-50 per cent of all data centre power demand by 2030, making infrastructure efficiency a survival metric, not just a goal. We are seeing three primary evolutions:

• The density explosion: Traditional racks used to pull 5-10 kW, but AI racks are now demanding 50-100 kW. This is forcing a complete redesign of the physical layer – more fibre in less space to handle massive east-west traffic.
• Zero latency: In AI training, a millisecond of latency can leave thousands of graphics processing units idling. Infrastructure is moving from standard 100G/400G towards 800G and 1.6T per wavelength.
• Sustainability as a constraint: It is no longer just about power. It is about green AI. Operators are seeking fibre solutions that reduce the carbon footprint of the data centre build and support liquid-cooling architectures.

How is STL positioning itself at the intersection of connectivity, compute and cloud?

We view ourselves as the physical architects of the AI era. With vertical integration, from glass to data centre infrastructure, we ensure that the physical layer – the fibre – is never the bottleneck for the intelligence layer. This positioning spans three core pillars:

• Vertical integration (glass to data centres): By controlling the glass chemistry, we are capable of engineering fibre that handles the tight bends and high densities of modern cloud environments better than anyone else.
• Neuralis data centre portfolio: We have moved beyond providing an integrated portfolio (Neuralis) specifically for the AI data centre infrastructure, covering everything from high-count intermittently bonded ribbon cables to plug-and-play interconnects.
• Co-creation: We do not just sell products – we work with hyperscalers to design solutions that fit their unique duct and power constraints.

How is STL’s work in advanced fibre and cabling technologies addressing the low-latency demands of AI interconnects? What measurable advantages do they deliver?

STL’s advances in fibre and cabling technologies are directly engineered to meet the low-latency and high-throughput demands of modern AI interconnects, with each innovation targeting a distinct bottleneck in data centre infrastructure. Its recently announced hollow core fibre cable, in which light travels through air rather than solid glass, reduces latency by approximately 30-40 per cent compared to standard fibre, a performance leap with direct implications for high-frequency trading and AI synchronisation workloads.

At the density frontier, the company’s Celesta cables accommodate up to 6,912 fibres within a single conduit, enabling massive parallel processing and meaningfully reducing the number of hops data must traverse across the network. Complementing this, STL’s 160-micron fibre achieves a thinner form factor without compromising tensile strength, allowing operators to pack greater capacity into existing cable trays – a critical advantage for space-constrained data centres seeking to scale bandwidth without infrastructure overhaul.

Perhaps the most consequential development for next-generation AI clusters is multi-core fibre (MCF). Where 160-micron fibre addresses density at the conduit level, MCF multiplies throughput at the strand level, embedding multiple data lanes within a single fibre to deliver fourfold, sevenfold or even greater increases in capacity, all without any corresponding increase in cable diameter. As AI architectures push towards 1.6T and 3.2T speeds, MCF is positioned to become the standard backbone of high-performance data centre connectivity, offering a scalable path to the extreme throughput thresholds that future AI clusters will demand.

How is STL building out its end-to-end manufacturing ecosystem for AI interconnect kits? What does its global manufacturing push mean for supply chain resilience and customer delivery timelines?

STL’s manufacturing strategy for AI interconnect kits is built around an integrated, geographically distributed model designed to compress delivery timelines while insulating customers from supply chain volatility. The company operates major manufacturing hubs across India, Italy and the US, with its South Carolina investment anchoring a localised supply chain presence in one of the world’s largest data centre markets. This multi-region footprint allows STL to serve customers with proximity-sensitive fulfilment, reducing both lead times and exposure to cross-border logistics disruptions.

At the product level, STL is shifting decisively towards factory-integrated optical connectivity kits, pre-terminated and fully tested solutions that arrive at the data centre ready for immediate deployment. By moving integration work upstream into the factory, the company reduces on-site labour and installation time by up to 50 per cent, a measurable efficiency gain for hyperscalers and co-location operators running compressed build cycles.

Underpinning the entire ecosystem is STL’s ownership of preform and glass supply, the raw material foundation of fibre manufacturing. By controlling this critical input layer, the company limits its customers’ exposure to global commodity shocks that have historically disrupted network build programmes, ensuring that AI data centre roll-outs can proceed on schedule regardless of broader materials market conditions.

With AI moving rapidly from pilots to production deployments, what are the most pressing infrastructure priorities that enterprises and operators need to address today?

As AI transitions from pilots to production, enterprises and operators face three infrastructure priorities that cannot be deferred. The first is fibre densification: running 2026 AI workloads on 2016 fibre infrastructure is simply not viable, making an audit of the physical layer for capacity bottlenecks the most immediate starting point.The second is thermal and space management. Data centres are already operating at the edge of their physical limits, and fitting the connectivity density that modern AI workloads demand into environments that are already hot and crowded requires a shift to high-density intermittently bonded ribbon, a step that has moved from optional to obligatory. The third priority is modular scalability, ensuring that infrastructure is architected to grow in pay-as-you-go fibre increments rather than requiring wholesale upgrades, allowing operators to match capex to deployment pace without creating stranded capacity or premature bottlenecks.

Data localisation and digital sovereignty are becoming central to infrastructure planning in India. How is STL helping enterprises and operators build resilient and future-ready national infrastructure? What is your outlook for AI-ready data centre growth in India over the next two to three years? What will be STL’s key priorities as the market moves into its next phase?

India’s data protection laws require data to stay within our borders.

• Resilient national backbone: We are helping Indian telcos build a secure, high-capacity national long-distance network that ensures data sovereignty is not just a legal requirement but also a high-performance reality.
• Domestic innovation: Because we develop our own intellectual property and manufacture locally, we provide a trusted source for India’s critical infrastructure, reducing reliance on imported, black-box technologies.
• Quantum readiness: Through our work in quantum key distribution-ready fibre, we are helping secure India’s infrastructure against the future threats of quantum computing.

Looking ahead, the next 24-36 months in India will be the most explosive in our digital history.

• The multi-GW era: India is on track to cross 1.5 GW of data centre capacity by 2026, with AI being the primary driver. We expect to see AI clusters emerging beyond Tier I cities into states like Gujarat and Uttar Pradesh.
• STL’s priorities: Our focus will be on scale, speed and sustainability. We will be doubling down on our Neuralis portfolio for the Indian market, ensuring that the IndiaAI Mission is supported by world-class and home-grown optical technology.