The AI race: from model capability to capital

This reporting period provided strong evidence that AI competition is now a contest of operating models and capital allocation, not only model capability. Reports described a Google DeepMind reorganisation and the increased authority of chief AI architect Koray Kavukcuoglu. South Korea announced its 'Seven Major SEED' programme, covering quantum, lunar missions, space communications, AI-bio infrastructure, autonomous laboratories, energy and critical raw-material supply chains. These are policy goals with deadlines, not achieved capabilities.

Financial results gave harder evidence of demand. CoreWeave raised its 2026 capital-expenditure plan to US$35-39 billion, reporting quarterly revenue of US$2.58 billion and a backlog of US$104.2 billion. Nebius reported US$582.3 million in revenue, more than US$40 billion in customer commitments and a 5 GW contracted-power target for 2026. Tencent said AI improvements are lifting advertising and cloud demand, and Foxconn's AI server-centred cloud and networking division passed half of quarterly revenue. These confirm demand while exposing debt, power, advanced-packaging, customer-concentration and execution risks.

Regulatory pressure spread from internet content to embodied AI. A German authority filed a complaint over Meta's smart glasses, and a French press alliance asked the competition authority to act over Google's AI article summaries. Both are complaints only, not rulings. The key framework update for builders is LangChain 1.3.15, whose middleware changes fix history loss, stale structured output and the risk of a human-approval gate failing open.

The digest's sections cover the top 10 stories, sector news, new models and tools, LangGraph and LangChain monitoring, robotics, space and drones, defence AI country monitoring, funding and partnerships, government and regulation, and five key takeaways.