SK Telecom’s AI That Never Forgets
Let me start with the one that actually made me sit up straighter. On August 4, SK Telecom (South Korea’s largest wireless mobile carrier) released A.X K2, a 688-billion-parameter AI model that shifts from reactive to proactive AI.
Here’s why this matters. Most AI answers questions. A.X K2 plans and executes tasks on its own. It can handle documents over 300 pages long with 67.7% better throughput than its predecessor.
The real-world impact? SK Biopharmaceuticals is using it to develop targeted cancer therapies. The early-stage drug research phase went from one to two years down to five months. That’s not incremental improvement. That’s a rewrite of how drug discovery works.
SKT is also testing it at steelmaker KG Steel and automotive parts company Conec, where it learns from production line defects to analyze causes and propose solutions. This is AI that actually works in factories, not just in demos.
And here’s what caught my attention: SKT plans to release it under a free license to support commercial use by developers and startups. That’s how you build an ecosystem.
Adtran’s Open Foundation for Telco AI
On August 6, Adtran (American fiber networking and telecommunications company) launched Mosaic One Fabric—an open platform that lets telecom operators build their own AI agents and workflows without vendor lock-in.
Here’s the problem it solves. Most telco AI solutions force operators into predefined use cases and vendor-controlled ecosystems. Mosaic One Fabric flips that. It exposes the network, operational data, and telecom expertise as reusable, agent-ready tools.
Operators can build AI that reflects how their organization actually works. As Adtran’s CTO put it: “The telecom industry doesn’t need another closed AI ecosystem”.
This is how AI gets deployed at scale—not through vendor roadmaps, but through operator autonomy.
Airtel’s Edge AI That Cuts Cloud Costs to Zero
Here’s the one that made me think about the economics of AI differently. Airtel deployed an on-device AI model across 30,000 field engineers.
Instead of sending data to the cloud for processing, the model runs directly on the engineer’s handset. Images are analyzed in real time at the site.
The result? The company was spending roughly Rs 30-45 crore on cloud workloads for these operations. Now that it’s done on the device, the cost has gone down to zero.
This is what edge AI actually looks like in practice. Not a demo. Not a pilot. 30,000 engineers, real-time inference, zero cloud cost.
Alibaba’s 16-Day Autonomous Coding Marathon
On August 3, Alibaba released Qwen3.8-Max—2.4 trillion parameters, 1 million token context, and ranked fifth in Text Arena and second in Vision Arena.
Here’s the part that genuinely surprised me. In internal testing, the model autonomously executed a real-world software engineering project over a 16-day period. Tasked with creating a self-evolving agent framework from scratch, it established an engineering loop, synthesized feedback, and produced “oh-my-cli”—now fully open-sourced on GitHub.
This isn’t a model that assists. It’s a model that builds. On its own. For over two weeks.
The $2 Billion Bank Bet and the $1.1 Billion Security Play
While the model launches grabbed headlines, the infrastructure behind them told a different story.
Rabobank (Rabobank is a Dutch multinational banking and financial services company) announced it would invest €2 billion ($2.3 billion) over the next three years in data, tech, and AI. Banks are stepping up their AI investments, and this is one of the largest commitments yet.
Meanwhile, Obsidian Security raised $85 million at a $1.1 billion valuation to secure AI agents accessing sensitive enterprise data. Nearly 70% of their customers already allow agents to interact with business data. The CEO’s warning? “Six to twelve months down the road, we are going to see a significant disruption in the agentic economy”.
And AMD deepened its AI inference bet by acquiring Taalas, a chip startup specializing in reducing computing and memory bottlenecks. Specialized inference chips have become critical as AI shifts from training to real-time, high-volume deployment.
Not just experimental. Not just conversational. Operational. In factories, in field engineering, in drug discovery, in banking, in security.
This was the week AI stopped being a toy and started being the infrastructure businesses actually run on.
AI, #SKTelecom, #EdgeAI, #Alibaba, #Qwen3.8, #Adtran, #Rabobank, #ObsidianSecurity, #AMD, #AIDeployment, #OperationalAI, #Innovation, #TechNews, #TelecomAI, #AgenticAI