AI Trends — 12 August 2026

Anthropic builds its own infrastructure layer

On August 11, Anthropic launched Theseus Infrastructure, a joint venture with Macquarie Asset Management and Singapore's GIC sovereign wealth fund. Macquarie and GIC will fund and own the majority equity in purpose-built US data centres, with Anthropic signing on as anchor tenant under long-term leases. Notably, Anthropic has committed to cover 100% of grid-upgrade costs and any resulting increases in consumer electricity prices — the first such pledge from a frontier AI lab, aimed at easing local opposition to new data-centre construction.

This follows Anthropic's existing $100B+ commitment to AWS for Trainium compute capacity. Theseus adds an owned real-estate and physical-infrastructure layer on top, giving Anthropic predictable capacity without full reliance on hyperscaler leasing or spot-market pricing. No financial terms were disclosed.

EU forces Google to open Android to rival assistants

Separately, binding orders issued by the EU under the Digital Markets Act require Google to open 11 Android features — including voice invocation, on-device app context, autonomous app control, and on-device ML models — to competing AI assistants such as Claude and ChatGPT. The deadline is Android 18, no later than August 1, 2027. Google must also begin sharing search query, click, and ranking data with rival search and AI providers from January 2027, under FRAND pricing terms.

Fines for non-compliance can reach 10% of global turnover. The order applies only to the EU/EEA; Android elsewhere is unaffected. Google has disagreed publicly with the ruling and may appeal, though the orders remain binding unless overturned.

Also this week

The broader AI landscape continues to move fast on several fronts: model releases keep accelerating (recent entries include GPT-5.6, Claude 4.6, Gemini 3.1, DeepSeek-V4, and Llama 4), with competitive advantage increasingly determined by matching the right model to the right task at the right price. Industry leaders are also increasingly vocal about AI entering an "intelligence explosion" phase, pointing to Google DeepMind's leadership restructuring, new ventures focused on automating scientific discovery, and labs' claims that their own models are now contributing to AI research. Enterprise adoption is shifting from simple prompting toward autonomous, end-to-end agent workflows — what some are calling "the agent leap."

This newsletter covers AI developments from the past 48 hours. Sources: AIToolsRecap, Bloomberg, European Commission press releases.