DAILY INTELLIGENCE BRIEF

The Vault — AI Edition

Monday, June 29, 2026
15 stories
14,205 words
73 min read

Business

5
01

South Korea Bets $576 Billion on AI Chips, Tapping Samsung and SK Hynix to Cement Its HBM Lead

Standing alongside the chairmen of the world's two largest memory-chip makers, President Lee Jae Myung unveiled a $576 billion plan.

5 minRead →
02

'Tokenmaxxing' Is Over: Enterprises Pivot From Spend-at-All-Costs to Model-Routing Efficiency

Six months ago, the fastest way to look like an AI power user inside a big technology company was to burn tokens.

5 minRead →
03

Lindy Moved 100% of Its Traffic Off Claude to DeepSeek — and Watched Its AI Bill Collapse

When Flo Crivello, the founder and CEO of AI agent startup Lindy, pulled the trigger last week, he did not move a slice.

4 minRead →
04

IBM and Google Cloud Forge a Multi-Billion-Dollar Enterprise AI Alliance Built on Gemini

Big Blue just placed a very large bet on a model it does not own. On June 4, IBM and Google Cloud announced a new practice.

5 minRead →
05

Figure AI Enters 2026 as Humanoid Robotics' Best-Capitalized Bet — a $39B Valuation, Its Own AI Brain, and a Factory Race

When Figure AI told investors last autumn it had blown past $1 billion in committed capital at a $39 billion valuation.

5 minRead →

Llms Genai

4
06

OpenAI Previews GPT-5.6 as Three Tiers — Sol, Terra, and Luna — but You Cannot Use It Yet

OpenAI on Friday unveiled its most capable model family to date — GPT-5.6, split into three permanent tiers.

5 minRead →
07

Z.AI's GLM-5.2 Pushes Open-Weight Coding to the Frontier With a 1M-Token Context

For two years the unspoken rule of the AI race was that the most capable coding models stayed locked behind a closed API.

5 minRead →
08

Moonshot's Kimi K2.6 Bets on Open-Weight, Multimodal Agents for Coding and Visual Reasoning

When Moonshot AI open-sourced Kimi K2.6 on April 20, the Chinese lab did not just ship another large language model.

4 minRead →
09

Alibaba's Compact Qwen3.5 Aims to Put Frontier-Class Reasoning on a Single GPU

The smartest model you can run on a laptop this month was not built in California. It was built by Alibaba.

5 minRead →

Research

3
10

Anthropic's Economic Index Finds 35% of Workers Expect AI to Do Most of Their Job Within a Year

Ask the people closest to the frontier of AI how much of their job a machine will be doing a year from now.

5 minRead →
11

'RoboAlign' Teaches Robot Policies to Reason at Test Time, Tightening Language-Action Alignment

For the past two years, the dominant recipe for building robot brains has looked deceptively simple.

5 minRead →
12

A New Benchmark Tests Whether AI Agents Can Reason in Real Time as the World Changes Around Them

Remember your first highway drive: knuckles white on the wheel, eyes locked on the car ahead, when your exit suddenly nears.

5 minRead →

Policy

3
13

US Partially Restores Anthropic's Claude Mythos 5 for Critical-Infrastructure Defenders — Fable 5 Stays Banned

Fifteen days after the federal government forced Anthropic to pull the plug on its two most powerful models worldwide.

5 minRead →
14

The Government-Gated Era of Frontier AI Has Begun, and It Starts With GPT-5.6 and Mythos 5

For the past decade, the moment a frontier AI model shipped was the company's call alone.

5 minRead →
15

As Its June 30 Effective Date Arrives, Colorado's Landmark AI Act Is Already Hobbled by a Hold on Enforcement

On Tuesday, the first comprehensive 'high-risk' artificial intelligence statute in the United States is supposed to take effect.

5 minRead →