AI Industry Daily News
A roundup of the AI industry's day, centered on Codex Windows support, grok-build-0.1, Claude Opus 4.8, Command A+, and Rosalind Biodefense.
Today's highlights
Key topics and reactions
SpaceX Completes IPO in 74 Days as OpenAI and Anthropic Near Public Filings
SpaceX completed its public offering in roughly 74 days, filing a confidential S-1 with the SEC on April 1, 2026, releasing a public version around May 20, and listing on Nasdaq under the ticker SPCX around June 12. Shares were priced at $135, with about 555 million shares sold to raise roughly $75 billion, exceeding $85 billion including the overallotment. The stock opened near $150 and rose past $192, briefly carrying a market capitalization above $2 trillion.
The speed and demand are seen as setting the stage for the two largest generative AI companies. Anthropic submitted a confidential draft S-1 on June 1 and carries a $965 billion valuation from its Series H round, with quarterly operating profitability reportedly in sight. OpenAI filed its S-1 confidentially around June 8 ahead of a Wall Street debut, after a $852 billion valuation as of March 2026.
The report from The Information suggests both companies could make their prospectuses public within weeks, following years of large private fundraising.
Some US Electricians Question Working on AI Data Center Projects, WIRED Reports
WIRED reported on June 22, 2026 that the boom in AI data center construction is prompting ethical hesitation among some US electricians, even as Meta, Google, Microsoft and Amazon execute multibillion-dollar projects. Electrical work accounts for 45–70% of construction costs, and the IBEW, the largest US electrical workers union, has framed the demand as a new generation of jobs supporting the AI revolution.
McKinsey estimates that about 130,000 additional electricians and 240,000 construction workers will be needed by 2030, with single data center projects sometimes requiring three to four times the usual workforce. At the same time, opposition over power and water use, noise and tax incentives has led to permit denials and moratoriums, stalling at least $15.6 billion in projects in 2025.
Views among workers diverge. Some treat the work pragmatically and value the high wages, while others cite AI-enabled fraud, bubble concerns and community harm as reasons to avoid the projects. The IBEW published its 'Data Center Principles' in March 2026 to assert the role of union labor.
xAI Begins Internal Private Beta of Grok 4.5 at SpaceX and Tesla
Grok 4.5 has entered a private beta limited to SpaceX and Tesla, according to an announcement from Elon Musk. The model is reportedly built on SpaceX's new '1.5T V9' foundation model, with Cursor data added to supplementary training and continued reinforcement learning.
Early evaluations circulating online have suggested the model could match or exceed Opus-class performance, raising expectations. There is also a claim that SpaceX plans to release a new model each month.
Because access remains restricted to internal teams, there are effectively no external user reports, and some observers have noted the gap between outside expectations and the model's internal-only status, with reinforcement learning still described as in progress.
Open-Weight Coding Model GLM-5.2 Becomes Available via Together
The open-weight, coding-specialized model GLM-5.2 has become available via Together, with developers describing it as competitive on coding benchmarks and usable in production. In practice it is used for agentic coding, sometimes paired with fallbacks such as DeepSeek v4 or Llama Maverick 4 on complex problems, and run seamlessly through OpenRouter.
One developer said GLM-5.2 is strong enough at agentic coding that they hit their weekly usage limit in a single day. Strict rate limits are cited as a recurring friction point, leading heavy users to combine it with other models.
The model is also viewed positively in the context of lower cost and avoiding certain Western regulatory constraints.
Category highlights
claude-reflect Plugin Turns Claude Code Corrections Into Persistent Memory
Developer BayramAnnakov released claude-reflect, an open-source plugin that automatically detects corrections, positive feedback and preferences given to Claude Code, queues them for human review, and syncs them to CLAUDE.md, AGENTS.md and skill files. Detection combines real-time regex with semantic AI verification scored from 0.60 to 0.95. A v2 update analyzes session history to auto-generate reusable skills and commands from recurring patterns. It is positioned alongside tools like claude-mem, which has over 65,800 GitHub stars.
Grok Imagine Video 1.5 Adds Image Quality, Speed Gains and API Preview
xAI released Grok Imagine Video 1.5 with improved image quality, faster generation and an API preview. The update is part of a broader push in video generation, where vendors are competing on quality and latency.
LumeFlow, Kling and AI Short Dramas Advance Video Generation
LumeFlow AI announced a major update adding Seedance 2.0 Mini, 4K video and a Marketing Studio. Kling is selecting its next character to animate via community voting. AI short dramas have reportedly surpassed 35 million downloads and 1.7 billion views and are moving toward full-scale production.
Together and MiniMax to Present on Large-Scale Agent Operations at AI Engineer World's Fair
Next week's AI Engineer World's Fair brings vendors together, with Together and MiniMax scheduled to lead workshops and a discussion on operating agents at scale, focusing on the gap between single model calls and production agents. HeyGen produced the official opening video using HyperFrames.
MiniMax Plans to Share Training and Inference Insights From M3
MiniMax plans to publish findings from its M3 model covering both training and inference. The disclosure feeds into a broader trend of vendors sharing implementation challenges around large-scale agent operations.