AI This Week: Offline Models, Nuclear Systems, and a Surge in Industrial Automation Google, Palantir, HCLTech, Apptronik, and Meta lead this week's AI moves

This week in artificial intelligence (AI), developments highlighted a shift towards edge computing, offline capability, and strategic investments across infrastructure, industry, and enterprise.

Google Pushes Offline AI with Gemma 3n and Gemini Robotics

Google formally launched Gemma 3n, a multimodal, open-source AI model capable of running entirely offline on devices with as little as 2GB RAM. Built on a new architecture called MatFormer, it allows scalable performance across hardware constraints and supports tasks involving text, video, image and audio. Key innovations such as KV Cache Sharing and Per-Layer Embeddings reduce memory load and speed up real-time applications like voice assistants and video analysis. Gemma 3n supports over 140 languages and processes content in 35, without requiring any cloud connection,"offering a significant advantage for privacy-sensitive and remote environments.

Meanwhile, Google DeepMind unveiled Gemini Robotics On-Device, a lightweight AI model designed to run locally on robots such as Franka FR3 and Apollo. Trained initially on ALOHA systems, Gemini can understand natural language, execute complex multi-step commands, and perform dexterous tasks like folding clothes and unzipping bags all without internet access. The model is built for latency-sensitive, low-compute environments, allowing robotic systems to operate independently in the field.

Palantir to Develop AI Operating System for Nuclear Construction

Palantir Technologies entered into a USD 100 million, five-year agreement with a Kentucky-based nuclear energy company to co-develop a Nuclear Operating System (NOS). The project follows recent executive orders in the United States aimed at accelerating domestic nuclear plant construction amid soaring demand from AI data centres and cryptocurrency mining operations. The NOS is intended to reduce costs, simplify planning, and accelerate construction timelines for new nuclear reactors. The initiative underscores how AI is now being applied to complex, regulated infrastructure sectors.

HCLTech Strengthens Enterprise AI Play through Salesforce and Global Alliances

HCLTech expanded its partnership with Salesforce to drive adoption of agentic AI solutions using the Salesforce Agentforce platform across sectors such as healthcare, manufacturing, retail, and financial services. The Indian IT major also announced a strategic alliance with AMD to accelerate digital transformation globally and signed a long-term agreement with European energy firm E.ON to modernise its cloud infrastructure using AI.

Additionally, HCLTech secured an engineering services contract with Volvo Group to support automotive innovation from its global centres. Despite these strategic moves, analysts maintain a cautious view on the stock, with a consensus rating of 'Hold' and an average price target of INR 1,670, suggesting a potential downside of approximately 3 per cent from current levels.

Apptronik Launches Elevate Robotics to Advance Industrial Automation

In a bid to go beyond humanoid AI systems, Apptronik launched Elevate Robotics, a wholly owned subsidiary focused on developing industrial-grade mobile manipulation systems. The spin-off aims to build "superhuman" robots designed for heavy-duty industrial tasks. Elevate will operate independently, leveraging Apptronik's decade of expertise and recent funding,"bringing the company's total Series A funding to USD 403 million, with backing from investors such as Google, Mercedes-Benz, and Japan Post Capital. The move reflects a broader trend of AI being integrated into large-scale, labour-intensive sectors.

Google Launches Doppl: An AI-Powered Virtual Try-On App

Google also released Doppl, an experimental AI app under Google Labs, which lets users virtually try on outfits by uploading a full-length photo. The app uses AI to simulate how clothing would look on a person's body, offering both static visuals and animated try-ons. Currently available in the United States on iOS and Android, the app has not yet been rolled out internationally.

Meta in Talks to Raise USD 29 Billion for AI Data Centres

According to media reports, Meta is in discussions to raise USD 29 billion,comprising USD 3 billion in equity and USD 26 billion in debts from private equity firms including Apollo Global Management, KKR, Brookfield, Carlyle, and PIMCO. The funding will support Meta's ongoing expansion of AI-focused data centres. The move is part of Meta's broader AI strategy, with CEO Mark Zuckerberg stating earlier this year that the company plans to spend up to USD 65 billion on AI infrastructure in 2025.