Evening Brief - 4 November 2026

The cost of running AI at scale came into sharp focus today, from personal token bills to memory allocation battles and a power efficiency breakthrough that could change the edge inference game.

Y Combinator CEO spends up to 200k yearly on AI agent tokens

Garry Tan disclosed he pays between 100,000 and 200,000 USD annually on API tokens to run autonomous AI agents for research, coding, and operational work. One agent has ingested two decades of his personal archives, including email, cloud storage, notes, and therapy records. Another system, Stray Light, serves as institutional memory for Y Combinator. Read more →

Aisin and Tohoku University unveil MRAM chip using 98% less power

The Toyota Group supplier announced on 3 November 2026 a semiconductor incorporating magnetoresistive random-access memory that consumes roughly 98 per cent less power than comparable chips in production. The collaboration with Tohoku University targets AI inference hardware at a time when energy demand is colliding with pressure to decarbonise. Read more →

Xpeng tests robotaxis in Guangzhou, eyes licensing model for 2028

The Chinese EV maker has moved its Level 4 autonomous driving programme into public demonstration mode in Guangzhou with paying passengers. CEO He Xiaopeng is building a licensing play for overseas mobility platforms rather than fleet ownership, targeting deployment by 2028. The trials test whether the entire service stack holds together under real-world demand. Read more →

Altman defends centralised DRAM as hyperscaler contracts lock supply through 2028

Sam Altman argued that personal computers and edge devices waste memory by sitting idle most of the day, while hyperscale data centres run at sustained utilisation rates above 80 per cent. OpenAI and its peers have effectively cornered the high-bandwidth memory market, with purchasing agreements locking up DRAM supply through 2028. Read more →

Compiled by OTW Briefing Desk from Opentechwire reporting.