← Back to newsletters

AI Daily Brief — Thu Oct 29

2026-10-29

By Vadym · Generated with AI, curated by me


Listen to this issue

Today’s stories are about who controls the scarce inputs: a memory maker posting a margin that shouldn’t exist, OpenAI reportedly chip-shopping at a rival, and a reminder that AI can already overwhelm a process built for humans to use.


Headlines & News
Models

OpenAI Ships GPT-6.1 Terra at Half the Price of Its Predecessor

OpenAI released GPT-6.1 Terra on October 26, replacing GPT-5.6 Terra across ChatGPT Work, Codex, and the API. API pricing dropped to $1/$5 per million input/output tokens — half the previous Terra tier. OpenAI is pitching it for computer-use tasks like filling out forms and reading technical drawings, and says its internal error rate on a hard-prompt benchmark fell from 53.6% to 16.7% at low reasoning effort.

Source: BleepingComputer

Industry

OpenAI Reportedly Weighing a Deal to Buy Google’s AI Chips

Reuters, citing The Information, reported that OpenAI is in talks to buy Google’s custom TPUs for its own data centers. No volume, pricing, or timeline was disclosed, and neither company has confirmed the talks publicly.

Source: Reuters

Robotics

Google Cloud and NASA Test Open-Source Robot Model for Deep Space Missions

Google Cloud and NASA are testing an AI-controlled robot built on the open-source Pi0 model (about 4 billion parameters) at Johnson Space Center. The system hit an 85% success rate in initial lab tests, and fine-tuning it for a new task reportedly took about 2.5 hours.

Source: Google Cloud Blog

Hardware

SK Hynix Posts Record Q3 Profit as HBM Demand Outpaces Supply

SK Hynix reported Q3 revenue of 104.62 trillion won (about $75.5B) and operating profit of 83.33 trillion won (about $63B), up 632% year over year — a 79.7% operating margin. The company says its production is effectively committed through 2029 and plans roughly 100 trillion won in shareholder returns this year.

Source: The Korea Herald

Funding

General Compute Raises $100M to Challenge Nvidia on AI Inference

San Francisco startup General Compute raised a $100 million Series A at an roughly $800M post-money valuation, co-led by Blackbird Ventures and Abstract Ventures. It rents out a “heterogeneous inference cloud” built on Cerebras, SambaNova, and Positron AI chips, claiming up to 40x faster inference than conventional GPUs on some workloads — a company claim that hasn’t been independently verified.

Source: TechStartups

Policy

AI Bots Generated Most of the Responses to a UK Government Consultation

A study published in Tobacco Control found that AI bots generated 90,835 of 118,756 responses — about 76% — to a 2023 UK government consultation on banning disposable vapes. The Department of Health and Social Care excluded the bot submissions during review, and the ban became law anyway.

Source: The Guardian


Analysis

Takeaway

Zoom out and today isn’t about a smarter model — it’s about who owns the scarce inputs underneath one. SK Hynix is printing a margin that shouldn’t exist in memory chips, General Compute is betting it can out-engineer Nvidia on efficiency instead of scale, and OpenAI is reportedly willing to buy from a rival rather than wait in the GPU line. Even the robot story fits: the win there wasn’t a bigger brain, it was a smaller one that retrains in hours. And the UK consultation story is the quiet warning underneath all of it — the infrastructure race gets the headlines, but the systems built for humans are already being tested by machines, and most of them aren’t checking.

— Boba


Curated by Vadym