This post originally meant to go out on 21st June was delayed because I had taken some personal time off to participate in Ashara Mubaraka [from 16th to 24th June]
This Week in All Things AI covers key developments in models, agents, tools, infrastructure, and policy curated from discussions in the All Things AI Telegram group.
If you follow AI for work, research, investing, or just to understand where the technology is heading, this weekly brief is a concise way to scan the most important launches, risks, and resources in a few focused minutes.
The week of June 14–20, 2026 saw high-profile industry consolidation, with SpaceX acquiring Cursor AI in a $60 billion all-stock deal aimed at accelerating model development on its Colossus supercomputer. Model releases continued to push boundaries, headlined by GLM-5.2—an open-weight model with a 1-million-token context window and competitive pricing. On the talent front, Noam Shazeer, co-inventor of the Transformer architecture, departed Google for OpenAI, while Microsoft's Satya Nadella outlined a strategy centered on companies owning their own "learning loops" and building "token capital" from internal data. Market concerns also surfaced, including leaked reports questioning OpenAI's financial stability and broader commentary on unsustainable business models across the sector. On the product side, Midjourney announced a healthcare division, CozyBlanket Pro introduced real-time AI topology prediction for 3D modeling, and MLX-LoRA-Studio brought local LoRA fine-tuning to Apple Silicon Macs.
The sections that follow walk through these items day by day, with short context and links so you can dive deeper into the pieces most relevant to your work or interests.
via Sam O who asked the following question
What do you think about what is said in this video ?
Short summary of the video: The AI industry is facing a potential collapse due to unsustainable business models, over-reliance on hype, and financial struggles, which may ultimately lead to a significant downturn in the industry.
Satya Nadella Outlines AI's Future for Businesses in Detailed Post
Satya Nadella says companies must own their AI “learning loops” to compound human and token capital. He describes a 'cognitive loop' where human judgment guides AI, emphasizing two key assets: employees' ingenuity and 'token capital' from custom AI developed on private data. He warns against AI labs capturing all value, pushing for a 'frontier ecosystem' that spreads benefits across firms and nations while avoiding industry hollowing. Tech leaders praised the positive vision, though critics pointed to job risks from Microsoft's own layoffs and workflow automation.
Who captures the learning?
His argument is that companies are becoming a new kind of learning system.
People bring judgment, taste, relationships, context and ambition. AI brings
“the real opportunity is not in picking
via Alex in response to Sam O's question on 14th June
Ed Zitron just leaked financials of OpenAI 😬 (good read):
https://www.wheresyoured.at/exclusive-openai-financials/?ref=ed-zitrons-wheres-your-ed-at-newsletter
Numbers read almost as badly as the SPCX S1.
If the numbers turn out to be real, this is 30~40% less revenue & >800% higher losses than claimed/hinted at by Sam previously
Here the FT reporting, they wouldnt publish without proper vetting. Super bearish:
https://www.ft.com/content/e15b0d7e-ff6b-4f16-ba7a-4068feddb828?syn-25a6b1a6=1
Lol, the FT just updated the article. They now claim that 77% of the losses booked last year (30 out of 39B$) might be attributable to the non-profit restructuring, which would be a much better case.
21B$ costs from operations on 13B$ of revenue.
Sorry, last message from my side on this (I just find the economies of scale of frontier models fascinating and have to share).
Im in Ed's discord and he now claims that he expects the real costs of operations to be much higher given that the leaked numbers are non-GAAP and are artificially reduced by removing at least 30% by associating costs to listed stakeholders (e.g. Microsoft).
He expects 9$ spent on 1$ earned.
Time will tell.
SpaceX announced it exercised its option to acquire Cursor AI, the leading AI code editor by Anysphere, in an all-stock $60 billion transaction aimed at building the world's most useful AI models.
For the past few months, SpaceXAI has been jointly training a model with Cursor, which will be released in Cursor and Grok Build soon.
The combination of Cursor’s leading product and distribution to expert software engineers with SpaceX’s million H100 equivalent Colossus training supercomputer will
GLM-5.2 offers a 1 million token context window, two reasoning modes, and low pricing at $1.40 per million input tokens. It leads as the top open-weight model on Arena.ai's Code Arena: Frontend with a 1,595 Elo score, dominates subcategories like Gaming and Data & Analytics, and ranks #3 on the tough FrontierSWE benchmark. Developers highlight its 11-times cheaper cost over competitors and real-world coding skills
- #2 React and #4 HTML sub-leaderboards
- Ranks as the top model in
Noam Shazeer, co-inventor of the Transformer architecture and Mixture of Experts, announced he is leaving Google to join OpenAI. Shazeer joined Google in 2000, co-led Gemini after returning from founding Character.AI, which Google acquired for $2.7 billion in 2024.
It was a difficult decision to move on. I’m incredibly proud of the amazing team at Google and everything we’ve built together. It has been an honor and a pleasure to
via Anthony Estebe
Hello everyone,
I’m trying to better understand how companies are actually using AI in their day-to-day operations.
Not the hype, and not coding, more things like admin, reporting, sales, support, internal knowledge, workflows, etc.
If you run a business or are involved in operations, I’d love your feedback. It takes around 2 minutes:
I'm doing a small market study on how companies are actually using AI today. Not for coding or technical work, but to run the business: operations, admin, sales, reporting, support, internal... | Anthony ESTEBE
I'm doing a small market study on how companies are actually using AI today. Not for coding or technical work, but to run the business: operations, admin, sales, reporting, support, internal knowledge, documents, coordination, etc. I'd love to hear from founders, CEOs, COOs, operations leads, and business owners.
via Ben
This is pretty amazing. Never would’ve expected this
Midjourney made new scanning tech
“Announcing a new division of Midjourney called "Midjourney Medical"”
via Kevin Dent
This is WOW
via Robby Yung
via George Isichos
There's a new app that lets you post-train models on any Mac with enough RAM. You can experiment with most popular finetuning frameworks like LoRA/QLoRA, full fine-tuning, or Quantization-Aware Training (QAT) that Gemma 4 recently used. I find it amazing that after local inference, any user can also create a customized version of Gemma 4 or Qwen3.5/3.6 at home.
GitHub - Goekdeniz-Guelmez/MLX-LoRA-Studio: A native Mac App for LLM fine-tuning on Apple Silicon - fully on-device, fully open source.
A native Mac App for LLM fine-tuning on Apple Silicon - fully on-device, fully open source. - Goekdeniz-Guelmez/MLX-LoRA-Studio
Below is my personal website which aggregates links to many of my socials as well as the various content and community that I curate. Feel free to share this link to others who you think may find this content/community useful to them
The cover image of this newsletter via generated via the Nano Banana Pro model within the Krea tool via the following prompt
Overgrown blooming flower, garden, Italian ornate courtyard, Watercolor, trending on artstation, sharp focus, studio photo, intricate details, highly detailed