Confluent dials into the millisecond economy: AI-native telecoms in the IQ era

Plug into live data

As AI in telecoms now evolves to its next stage, the natural progression is for us to grasp live real-time data feeds in order to gain a more accurate, more timely and more manageable approach to uncovering faults and misconfiguration, to aid us in the fight against fraud and enable us to be able to tweak capacity in real-time.

“At the recent Mobile World Congress (MWC) event in Barcelona, we were able to celebrate what we might call the ‘IQ Era’, in which the strategic application of intelligence to solve global challenges will see companies move beyond automation for automation’s sake; it’s telecoms networks that will enable the lightning-quick performance that businesses now need,” said Matias Cascallares, OEM technologist at Confluent, in conversation with the Computer Weekly Developer Network.

Where Cascallares tells us this gets us to is the arrival of the real-time AI-native telecoms network, a substrate function that is perhaps less focused on increasing traditionally core considerations such as bandwidth or coverage.

Beyond batch

Today, the AI-powered telecom network is defined by its ability to detect, analyse and assess the elements passing through it – and, crucially, then know how to act upon those events.

“Batch processing has long been a standard part of telecom operations. For decades, it was sufficient for handling call detail records and periodic updates. However, this approach is becoming increasingly incompatible with autonomous telecom systems,” explained Cascallares. “The old ways of the past saw us relatively ‘happy’ to analyse snapshots of network activity… but today we need to realise that’s rather like deciding when to cross a road based on a photograph of a crossing taken last week. We need to know where the cars and trucks are, now.”

Confluent’s Quick Thinking 2026 report suggests that this whole evolution is impacting the decisions of business leaders across the UK. Some 82% have to choose between making a quick decision and an informed decision, while 71% say the data they get is out of date by the time it reaches the C-suite.

The advice from Cascallares is that we can not manage a modern telecom network if we’re basing our actions on information about network activity that has changed and gone stale.

“This type of delay creates serious inefficiencies (not to mention risks) for large enterprises. MWC this year enabled us to think about how we now build the AI-native telco and how these operations use real-time data as their lifeblood. At the risk of jumping from traffic analogies to the body… we need continuous data inputs and autonomous systems to be functioning as efficiently as the central nervous system in our own bodies. What’s more, this is now, this is a prerequisite for the 5G era… and we’re just about to look forward to the 6G era where latency requirements will approach the sub-millisecond level,” said Cascallares.

The next era of communications sees us use cyber-physical systems that might range from drones to remote medical robotics, so it’s edge computing 2.0 with specific AI-driven decision making enabled, because delays of a few hundred milliseconds can be too late.

Milliseconds matter

Some compute will occur on the edge devices themselves (where milliseconds matter even more) and placing processing power and intelligence near those “device events” on the edge of the network is the answer. We might even suggest that poor old embedded computing has just got a whole lot sexier and more fashionable.

“The future telecoms network will be architected to be hybrid (edge and datacentre compute and analytics) by design,” said Cascallares. “One of the factors driving this reality is the fact that we will see model training and data aggregation carried out at the backend in the datacentre in one central location, but mission-critical milliseconds matter decisions will need to be executed on the device. Agentic AI will play an important role here i.e. where decisions need to be made at a speed too fast for human intervention, the systems that make them will be under intense scrutiny.”

Looking ahead (and deeper at the same time) Cascallares insists that human-in-the-loop oversight will remain essential, but there’s a subtle shift going on where humans move from concerns over intervention towards managing governance and compliance with a sophisticated, constant flow of data acting as context and evidence for any decisions made.

“Again at MWC, we saw SK Telecom exhibit its ‘AI-native’ infrastructure, using real-time monitoring to manage GPU customers for AI-as-a-Service (AIaaS). This is not hypothetical; it is happening right now… and it is changing the ways in which telecoms providers operate at a foundational level,” concluded Cascallares.

In wrapping up, Cascallares covered some other key food for thought areas to consider.

Many more network connections (and many more API connections therein) create new cyber attack vectors, so we need to think about everything from SIM cloning to deepfake calls designed to trick humans into compromising their identity credentials and more. We can no longer run potential phishing analysis scans at the end of a telecoms billing cycle; everything has to happen in real-time.

Modernising the incumbent stack

But modernisation of the incumbent stack of world telecoms assets will not be simple. Around the world, we know that the plugs, wires, switches (and perhaps the odd token ring adapter) were largely put in place before the rise of AI and agentic services in general.

Dialling into the new reality will take some planning and we need to spend serious time analysing our architectures and look at what needs to be augmented and what needs to be refactored.

In the meantime, hold on please caller.