Super Intelligence in Telecoms: 2026 State of the Industry

1. Policy: from AI to SI

On 22 September 2026, at the UN General Assembly, President Trump rejected "any attempt to construct a globalist scheme of control for the Artificial Intelligence being spoken of so much now — hereinafter officially called 'Super Intelligence'", arguing that "artificial" makes the technology "sound fake".

On 29 September, the White House published the executive order Inaugurating the Era of Super Intelligence. It directs executive branch agencies to replace "Artificial Intelligence" and "AI" with "Super Intelligence" and "SI" in official communications, websites, reports and policy documents. It does not require private companies to rename products, and statutory definitions are unchanged.

The same week, Anthropic, Google, Meta, OpenAI, Nvidia and xAI signed the White House Accord on Super Intelligence: a voluntary pledge to run four layers of oversight on frontier models (internal controls, oversight teams, independent external audits and board-level review), including monitoring for cyberattack and biological or chemical risks. Nvidia's Jensen Huang described the new data centres as "SI factories, super intelligence factories".

Why it matters for telecoms: telecom suppliers to US federal agencies will increasingly meet "SI" in procurement and policy language, and the Accord's focus on cyber risk lands directly on critical infrastructure such as communications networks.

2. Autonomous networks: the Level 4 race

TM Forum's Autonomous Networks framework runs from Level 0 (manual) to Level 5 (fully autonomous). Level 4 is a "highly autonomous" network that makes decisions across complex, multi-domain environments through prediction and closed-loop control.

• TM Forum's 2026 report Scaling to AN Level 4 to unlock network value found 81% of operators aim to reach Level 4 or beyond by 2030, 20% expect it by 2027, and 75% are increasing autonomous network investment this year.
• In August 2026, the NGMN Alliance published a report on how agentic SI can take operators from Level 2/3 to Level 4: agents interpret intent, plan actions, delegate to other agents, call network functions and validate the results.

3. Agentic SI in operations and customer care

Telecoms is one of the most advanced sectors for agentic SI. Industry surveys put telecom at the top of cross-sector agentic adoption, with security and fraud prevention and customer service the leading priorities.

• AT&T reports more than 410 generative SI agents across the business, from proactive network incident response to fraud detection and software development.
• Verizon's Gemini-powered assistant in the My Verizon app handles upgrades, billing, line additions and account management, linked to a reported ~40% increase in service-team sales.
• Vodafone is working with Google Cloud on an autonomous network framework.

4. AI-RAN: the radio network becomes SI infrastructure

AI-RAN runs radio processing and SI workloads on shared, GPU-accelerated hardware, so base stations can also serve SI inference.

• Nokia and Nvidia validated Nokia's anyRAN software on the NVIDIA AI Aerial platform and completed functional tests with T-Mobile US (at its Seattle AI-RAN innovation centre), Indosat and SoftBank.
• Nokia's timeline: commercial trials in 2026, first commercial release in 2027.
• BT, Elisa, NTT DOCOMO and Vodafone are engaging on AI-RAN as a path to SI-native 6G.
• Not everyone is convinced: IEEE ComSoc's March 2026 "reality check" pointed to an unproven business case and no agreed definition or standards yet.

5. AI grids: SI at the network edge

At NVIDIA GTC in March 2026, major operators announced "AI grids": distributed SI infrastructure across their network footprint that runs inference close to users.

• AT&T, with Cisco and Nvidia: an AI grid for IoT on a dedicated IoT core, for real-time uses such as public safety.
• Spectrum: infrastructure spanning more than 1,000 edge data centres.
• T-Mobile (with Nokia), Comcast and Indosat also announced edge SI plans; Akamai is orchestrating inference across 4,400 edge locations.

6. Telco AI factories and sovereign SI

• GSMA Intelligence counts 36 telco AI factory deployments to date, mainly driven by sovereign SI demand, and forecasts just under 3.5GW of telco-operated AI factory capacity by 2030.
• Telus's first sovereign AI factory in Rimouski, Quebec (opened September 2025) is fully sold out; Telus is planning three more SI-focused facilities in British Columbia.
• Deutsche Telekom is building an industrial AI cloud in Germany with Nvidia for European manufacturers.

7. Telecom-specific models

• GSMA Open-Telco LLM Benchmarks, backed by Hugging Face, the Linux Foundation, Deutsche Telekom and SK Telecom, measure how well models understand telecoms.
• SoftBank's Large Telecom Model (LTM) achieved a top-tier overall ranking on those benchmarks (March 2026).
• The Global Telco AI Alliance (SK Telecom, Deutsche Telekom, e&, Singtel, SoftBank) has a joint venture building a multilingual telco LLM for operators serving around 1.3 billion customers in 50 countries.

8. 6G: SI-native by design

3GPP froze Release 19 in March 2026 and is now focused on Release 20 6G studies, with SI-native architecture, integrated sensing, new spectrum and non-terrestrial networks on the agenda. First complete 6G specifications are targeted for around 2029, with commercial launches near 2030.

9. Market size

Forecasts differ in scope, but all point to growth of around 40% a year.

What to watch next

• Whether vendors and operators adopt "SI" in their own product names and marketing.
• Whether Accord provisions move into law or regulation, with critical infrastructure likely first.
• AI-RAN moving from 2026 trials to 2027 commercial products.
• Edge inference demand as SI agents multiply.