The waves of AI-related news were relentless in 2025 as the technology moved from experimentation to mainstream customer service and network operations, but did it live up to the endless hype?

Mobile World Live (MWL) surveyed several industry analysts for their perspectives on what worked in 2025, what did not and where AI is headed in 2026.

This AI panel of experts includes AvidThink founder and principal Roy Chua, Scott Raynovich, founder and principal analyst at research company Futuriom, Appledore founder and principal analyst Patrick Kelly, and Matt Walker, chief analyst at MTN Consulting.

MWL: Is the much talked about AI bubble real or imagined?

Walker: “Most definitely this is a bubble and has been for several quarters. It hasn’t popped because people keep raising the stakes and nobody is willing to back away and pursue a different strategy. Unless they are forced to, as Chinese companies are due to supply chain constraints. That may be a hidden blessing.”

Chua: “There’s definitely a valuation bubble for specific AI companies”.  He noted OpenAI went from $300 billion valuation in March to $500 billion in October this year. “Since the end of 2022, AI-related equities have driven 75 per cent to 80 per cent of the S&P 500’s earnings and total performance. The Bank of England is worried, and even Sam Altman admits investors are ‘overexcited’.

“Nevertheless, the underlying technology demand is very real. Hyperscalers can’t get GPUs fast enough, and the infrastructure buildout looks like the 1990s internet boom again. Over five to ten years, AI will likely have more impact than the internet or mobile but expect some painful corrections along the way.”

Raynovich: “Somewhat yes. The near-term value as expressed in financial markets is overexuberant, but its long-term impact will definitely be profound.”

MWL: 2025 was supposed to be the year of agentic AI, or, more granularly AI agents. Did it live up to the hype?

Kelly: “Early deployments demonstrate that AI agents can deliver measurable improvements when applied to well-defined domains such as RAN optimisation, transport anomaly detection, and fibre fault resolution.

“However, widescale adoption remains constrained. Appledore Research finds that the primary barrier is not model performance but data accessibility and quality. Fragmented data across OSS, network domains, and IT systems limits an agent’s ability to reason accurately and act confidently.”

Raynovich: “I give it a C+. It’s definitely going to be a long-lived technology but there are many things to work out on the compliance and security side.”

Chua: “Not quite, but that’s okay. The ‘year of the agent’ delivered more of a reality check than a revolution. While many organisations experimented with agents, only 17 per cent reached full deployment, per PwC’s Agent AI survey in 2025.

“However, AI agents are rapidly improving, and the state-of-the-art as represented by coding agents has progressed tremendously through 2025.  In general, agents worked great in bounded domains. And we’re seeing them across most telco OSS and BSS systems performing limited tasks well.”

MWL: What are the roadblocks for large-scale AI deployments?

Walker: “It depends on what kind of AI. With generative AI, the predictive, probabilistic nature of the models means that results are different every time they are run.”

“That is a big roadblock for an industry used to certainty and five-nines of reliability.”

Raynovich: “Security, data privacy, governance, and sovereignty.”

Chua: “Data, talent and culture. Few business leaders think their data maturity can support AI at scale, and many we talk to say data management can’t keep pace. We keep hearing that telcos are sitting on decades of siloed systems that don’t talk to each other.”

“The talent gap is brutal, and telcos have a hard time finding the ‘purple squirrels’ who understand both AI and telco. The hyperscalers and model builders are outbidding everyone for pure AI talent.

“Then there’s cultural resistance. At TM Forum’s DTW, industry leaders called it the primary hurdle to network automation. Fix the foundations first, or you’re just doing AI theatre.”

MWL: Are operators deploying AI to make money or save money?

Chua: “Save money. Telco survival demands it. As I shared, the GSMA confirms 75 percent to 80 per cent of deployments target cost reduction.

“But the pivot is starting. Deutsche Telekom is building a €1 billion ‘AI Factory’ with Nvidia. SK Telecom is pushing GPU-as-a-service, and T-Mobile US is using AI-powered switching tools to poach competitors’ customers. I think the path is clear: Save first to fund the revenue experiments.”

Walker: “So far it is mostly about cost and time saving. Hopefully next year we will start seeing more revenue growth examples.”

Raynovich: “The initial lift will come on the savings side. The operators that can efficiently connect data centres will see lift on the revenue side as well.”

MWL: Which US operator had the most success deploying AI this year?

Walker: “I would say that T-Mobile US deserves some recognition given the DT group leadership.”

Chua: “Probably T-Mobile, given their much lauded OpenAI-powered IntentCX platform. The T-Life app’s ‘Easy Switch’ feature completes carrier switching in 15 minutes using AI-recommended plans. These efforts are likely contributing to T-Mobile’s continued postpaid net adds that outpace the competition.”

MWL: Which vendor ruled the AI roost in 2025?

Chua: “Nvidia, obviously. They own 80 per cent to 90 per cent of the AI accelerator market. And they are spreading excess cash to give them visibility and influence over the data centre, application, and telco markets. For example, the $1 billion infusion into Nokia and the partnership for 6G AI-RAN help bolster their telecom position.

“Broadcom is also notable for its custom ASICs for Alphabet and Meta, as well as an OpenAI deal that represents 10GW of AI accelerators between 2026 and 2029.”

Raynovich: “Ciena had an epic year.”

Walker: “Ciena is a strong contender. Even if there is an AI bubble, all of these new AI data centres and factories need fibre connectivity, and Ciena is a leader in the space. It has a strong relationship with Meta, for example.

“Plus, Ciena continues to be a leading player in optics/IP in the telco market.”

MWL: What’s next for AI vendors, enterprises and operators?

Raynovich: “We believe the next phase of growth in AI is implementing AI at the edge, implementing new AI-focused data sovereignty and security solutions, and helping enterprises manage their data and AI applications.”

Chua: “Physical AI and edge intelligence. The next killer apps are robots, drones, and autonomous systems that need real-time network connectivity because they can’t process everything on-device.

“Telcos may own the beachfront property for supporting edge intelligence, but they should be cautious given past failings in MEC.

“Sovereign AI clouds are becoming real products. Deutsche Telekom, Orange, and others are wrapping Nvidia compute in national flags for regulated industries. Operators who figure this out could evolve into AI infrastructure providers, but the path will prove treacherous for all save the nimblest of telcos.”

Walker: “Not anytime soon, outside of attention-grabbing trials that don’t scale,” when asked about mass deployments of physical AI for robots or edge devices next year.