bernat farrero

Apple's unified memory architecture on M1-M4 chips allows running compressed open-source models locally with sufficient performance for bounded tasks (voice-to-text, scheduling, email triage), offering better privacy, lower latency, and no cloud dependency; this validates a hybrid model where frontier models handle novel tasks while specialized local models handle repetitive workflows.

Cloudflare confirmed Anthropic's unreleased model performs chained attacks (vulnerability discovery → lateral movement → credential reuse) impossible for humans; all public models are capped but open-source equivalents will match capabilities within quarters without safety controls; human remains weakest link (GitHub hack via compromised VS Code extension).

Anthropic CEO Dario Amodei forecasts 2027-28 as inflection for AI-driven biology where nearly all diseases can be simulated and cured in silico, but FDA clinical trial requirements (human testing, static regulators rejecting synthetic data) delay real-world deployment to 2035-37; cancer research centers just now building AI-native teams.

Industry consensus at Microsoft event: 200 GW of new power capacity required for AI data centers by 2030, equivalent to combined peak demand of four major European economies; 90% of US data center permits rejected due to local opposition over electricity costs and job displacement fears, creating structural bottleneck.

Google is the only hyperscaler with custom inference chips (TPUs), owned global fiber/data center network, massive cash reserves, and a consumer application layer (Search, Workspace, Gemini); this full stack lets it monetize both infrastructure demand and model leadership while competitors face compute constraints.

Elon Musk is committing to owning the entire AI compute stack — from lithography and chip design (Tesla Dojo) through robotics assembly (Optimus) to data center operations (Terafab) and ultimately space-based compute — to compress iteration cycles from years to weeks and achieve 20x current global AI compute capacity. This represents the most ambitious vertical integration play in tech history.

Generative video reduces production cost from ~1000-person crews to single creators, making fantasy, religious, and historical content economically viable. Contrary to displacement fears, this expands the total addressable market and creates more net jobs in audiovisual production as new formats emerge.

Personal agents (Idem, Hermes, OpenCloud) are evolving from chat interfaces to proactive assistants with explicit memory structures, cron jobs, 'dreaming' (memory consolidation), and heartbeat-driven proactivity; they act as digital chiefs of staff that learn user context over time to execute tasks autonomously.