The AI race is often framed around models, chips and whatever new chatbot appeared this week. But for Tan, the foundation powering all this progress is AI infrastructure. Julian Tan believes the bigger story sits underneath all of that.
Tan, Director of Corporate Strategy at BT Group, says companies and governments may be underestimating the infrastructure needed to keep artificial intelligence running at scale. Faster models are useful. They still depend on networks, data, computing capacity and systems that cannot afford to fail.
Speaking ahead of LEAP 2026 in Riyadh, Tan pointed to three areas that deserve far more attention: technological sovereignty, resilient connectivity and the changing location of computing power.
AI Sovereignty Is Becoming a Strategic Priority
Many countries want to build competitive AI industries, but relatively few control every layer of the technology stack.
The most important components often sit outside their direct reach. Advanced chips may come from foreign manufacturers. Cloud platforms can be operated by overseas companies. Training data, model access and computing infrastructure may also depend on external providers.
Tan argues that this lack of control creates a strategic weakness.
AI now influences government services, national security, healthcare, finance and business productivity. Depending heavily on another country or company for the chips, platforms and data behind those systems can leave organisations exposed to supply disruption, policy changes or commercial restrictions.
That does not mean every country needs to manufacture its own processors or build a domestic version of every major cloud platform. It does mean governments and businesses need a clearer view of which systems they control, which ones they rent and where the risks sit.
Sovereign AI has already moved beyond political language. It is becoming an infrastructure decision.
Reliable Networks Are Doing More Work Than People Notice
AI platforms do not operate in isolation. Every request sent to a cloud model relies on connectivity. Enterprise AI agents need secure access to business applications, databases and external services. Autonomous systems may exchange data continuously rather than waiting for a person to enter a prompt.
That activity places heavier demands on network capacity, cybersecurity and resilience. Tan says people often assume connectivity simply works. The investment required to maintain it remains largely invisible until an outage interrupts a service or exposes a weak point. As AI adoption expands, network failures could affect far more than video calls or website access. They could stop automated business processes, customer services and operational systems at the same time.
The infrastructure conversation, then, cannot stop at data centres and GPUs. Telecommunications networks will carry much of the AI economy. They need enough capacity to handle rising data volumes, enough security to protect sensitive information and enough redundancy to keep services online when something goes wrong.
AI Computing May Move Away From Mega Data Centres
The current AI boom has triggered enormous spending on centralised data centres. Technology companies are filling them with high-performance processors, cooling systems and power-hungry computing equipment. Tan does not expect that structure to remain fixed.
More efficient models and increasingly powerful chips could allow companies to run a larger share of AI workloads closer to the user. Instead of sending every request to a distant cloud facility, devices, offices, vehicles and local network infrastructure could process some tasks directly. This is known as edge computing, though the shift may eventually go further.
Tan suggested that future AI workloads could become highly distributed, with personal devices handling jobs that currently require large data centres. Improvements in chip performance, model efficiency and emerging computing technologies could accelerate that transition.
The data centre will not disappear. Training large frontier models will continue to demand industrial-scale computing power.
Inference may follow a different path. Running a trained model can happen in more places, particularly as developers create smaller systems designed for specific tasks. That could reduce latency, keep sensitive information closer to its source and ease some pressure on centralised infrastructure. It would also change where companies spend their infrastructure budgets.
Technology Companies Can Learn From Sports Fans
Tan’s background stretches across telecommunications, media, technology and sports. That mix has shaped his view of how technology companies build relationships with customers.
Sports organisations rarely forget who keeps the industry alive. Fans sit at the centre of ticketing, broadcasting, sponsorship and merchandise decisions. Their attention and loyalty have direct commercial value. Technology businesses sometimes work the other way around.
A company develops a capability, builds a product around it and only later asks how customers will fit into the picture. Tan believes this product-first mindset can limit long-term engagement.
Sports also treats the customer journey as an ecosystem rather than a single transaction. Teams, leagues, broadcasters and sponsors work together to turn casual viewers into committed supporters.
Tech companies often protect their place in the value chain instead. The desire to own the entire customer relationship can make partnerships harder, even when collaboration would expand the overall market.
AI companies may face this problem more often as the industry matures. No single provider controls every piece of the stack. Chipmakers, cloud platforms, telecommunications companies, software developers and enterprise customers all depend on one another. Building the market could matter more than winning every layer of it.
Big Companies Do Not Always Have to Innovate Slowly
Established businesses carry a difficult burden. They need to build new products without damaging the systems customers already rely on. That challenge can create caution. It can also turn ordinary decision-making into months of meetings. Tan argues that large organisations can move quickly when teams share a clear strategy. A common direction gives people enough context to act without waiting for approval at every step.
Execution matters more than simply producing ideas. Companies also need people with different experiences in the room. Engineers may see technical limitations. Commercial teams understand the customer. Security specialists notice risks that product teams might miss. Innovation becomes stronger when those perspectives collide early rather than after launch.
Trust completes the picture. Employees need enough autonomy to make decisions, but not so much freedom that they accidentally damage critical systems. When people understand the strategy, recognise the risks and trust one another’s judgement, established organisations can move faster than their size might suggest.
The AI Race Will Be Won Below the Surface
AI models attract the headlines because people can see what they produce. Infrastructure rarely gets the same attention. It still determines whether those models can operate securely, respond quickly and reach millions of users without collapsing under demand.
Countries will need to decide how much of the AI stack they want to control. Businesses will have to strengthen the networks connecting their systems. Hardware and model improvements may push more computing toward devices and the edge. The next stage of AI will not be shaped by software alone.
It will depend on who builds the infrastructure underneath it, who controls that infrastructure and whether it remains reliable when AI becomes part of everyday operations.
Source: LEAP:IN — Building for the AI era starts with infrastructure
