Good morning. It's Thursday, June 4, and we're covering how AI is minting new trillion-dollar winners in the chip industry, the growing debate over AI's impact on jobs, and Tesla's latest move to scale robotaxis.

Plus: new reader poll, and a guest essay arguing why running local AI models may become as essential as digital literacy.

Tesla expands its unsupervised robotaxi service across the entire Austin metro area even though it operates far fewer vehicles there than rival Waymo. The move marks the latest step in Elon Musk’s push to transform Tesla from an EV maker into an AI and robotics company. Tesla has about 50 robotaxis in Austin versus nearly 600 for Waymo, according to city officials. The company also launched robotaxi services in Dallas and Houston earlier this year and expects broader U.S. deployment in 2026.

Meta now allows employees to pause workplace activity tracking for up to 30 minutes at a time after internal backlash over a tool designed to train AI models. The Model Capability Initiative records keystrokes and mouse clicks to help build AI agents that can perform computer-based tasks. More than 1,500 employees signed a petition against the program, while Meta has cut roughly 2,000 jobs this year and plans to reduce its workforce by another 10%. Staff can also request exemptions as concerns over privacy and data usage persist.

Microsoft claims its new MAI AI models outperform larger rivals on key benchmarks while running more efficiently on the company's custom AI chips. The company unveiled seven models, including MAI-Thinking-1, a 35B-parameter reasoning model that scored 97% on AIME 2025 and 53% on SWE Bench Pro, alongside new image and coding systems. Microsoft says MAI-Thinking-1 delivers 30% better performance per dollar than NVIDIA GB200-based deployments and is already being customized for clients through its new Frontier Tuning platform.

OpenAI is preparing to bring Codex directly into ChatGPT as part of its broader push to turn the chatbot into an all-purpose AI platform. Executives say Codex increasingly handles tasks beyond software development, reflecting a wider shift toward specialized AI agents inside consumer products. The move revives one of OpenAI’s earliest coding brands and advances its “superapp” strategy, which aims to combine multiple AI tools and workflows within a single interface. Codex is already used by millions weekly and has expanded into research, analysis, and operational tasks.

Genspark went from zero to $250M ARR in 12 months—and it's easy to see why. It's an all-in-one AI Workspace powered by the models you already love (Claude, Gemini, GPT, Nano Banana Pro, GPT Image, Veo, etc) that actually does the work for you. One prompt builds your slides, sheets, docs, designs, websites, images, videos. And in 2026, all paid users get unlimited AI Chat and AI Image. Less prompting. More shipping.

Try Genspark free — sign up and get free credits today.

The following essay was written by guest contributor, Ahmad M. Osman, an AI researcher and systems engineer specializing in infrastructure and hardware.

“So run the models. Break them. Stress-test refusal boundaries. Compare them. Fine-tune one badly. Make one summarize your own notes. Make one argue both sides of a position you care about. Make one hallucinate in a domain where you know the truth. Make one write something beautiful and wrong.”

I first posted an earlier version of this argument on X/Twitter, January 17, 2026. This version is expanded as of May 18, 2026, because the local model ecosystem, open-weight releases, tooling, and real-world risks around model-mediated cognition have moved fast enough that the argument deserves more context.

Model names mentioned here will age. The core point will not.

In short: You should own, borrow, rent, or otherwise get direct access to enough local compute to understand the systems that are increasingly mediating your thinking. → Read the full article here.

Sam Altman and Dario Amodei are both walking back their AI jobs apocalypse predictions, companies are using AI as a scapegoat for bloat-era layoffs, and Jevons' paradox is playing out in real time as AI spending drives more hiring, not less.

Micron Technology reached a $1 trillion market capitalization in May 2026 after a dramatic shift from commodity memory chips to specialized high-bandwidth memory (HBM) used in AI systems. According to Reuters, NVIDIA CEO Jensen Huang encouraged Micron several years ago to align its product roadmap with the growing demands of AI infrastructure, helping the company secure higher-margin, long-term supply agreements.

The strategy transformed Micron’s financial performance, with the company reporting $14 billion in profit in its latest quarter after posting a $5.8 billion loss in 2023. While analysts say Micron’s historically cautious spending nearly left it behind in the AI race, that same cost discipline may help the company remain resilient when the semiconductor cycle eventually slows. → Read the full article here.

Microsoft unveiled its new Majorana 2 quantum computing chip on June 2, 2026, saying advances enabled by AI-assisted materials design have accelerated its path toward commercially useful quantum computers by 2029. The company replaced traditional aluminum-based superconducting components with lead-based materials, which it says delivered a 1,000-fold improvement in certain performance metrics and solved key manufacturing challenges.

The announcement places Microsoft on a similar timeline to IBM in the race to build practical quantum systems, alongside competitors including Google, Amazon, and several Chinese research efforts. → Read the full article here.

Altman Opposes AI Release Approvals: OpenAI will urge Congress to reject pre-release AI approvals while expanding federal model testing.

UK Forces Google AI Opt-Out: Publishers can block AI Overviews and AI Mode via Search Console, with attribution rules and a planned global rollout.

Morgan Stanley Opens Platforms to AI Agents: Client AI agents will access stock-plan systems directly, helping automate administration.

Perplexity Debuts Hybrid AI Routing: New system automatically keeps sensitive tasks local and sends complex work to cloud models.

Google Launches Gemma 4 12B: New model handles text, images, and audio without separate encoders, running locally on laptops with 16GB memory.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team