Sarmadi AI Digest June 24, 2026 Updated 7:30 AM CT Today Archive Topics Saved Subscribe RSS

OpenAI + Broadcom unveil Jalapeño chip; Qualcomm buys Modular for ~$4B; Hollywood bends to OpenAI

OpenAI and Broadcom announced Jalapeño, a custom LLM-optimized inference chip — the most concrete sign yet that OpenAI is moving beyond Nvidia for its own compute. Qualcomm acquired Modular for roughly $4B, ending speculation about one of the most-watched AI chip-software startups. The Verge documents Hollywood publicly bending to OpenAI — Netflix, A24, Focus, Warner Bros.' Clockwork all making deals. Corporate AI super PACs reportedly spent $27M on a single local New York election, marking the political-economy battle the Sanders $7T plan named. Anthropic shipped Claude Tag — an always-on AI teammate that learns your company through Slack. Oracle cut 21,000 jobs to fund its debt-driven AI investments. Midjourney's medical-AI ultrasound pivot from last week is now drawing skepticism over an apparent lack of clinical evidence.

7 papers 18 news 8 sources ← Latest

News

14 items

OpenAI's chip, Qualcomm's buy, Oracle's layoffs

OpenAI and Broadcom announced Jalapeño — an LLM-optimized inference chip — pushing OpenAI further off Nvidia's roadmap. Qualcomm bought Modular for ~$4B, locking in a chip-software stack for its consumer-AI push. Oracle laid off 21,000 to fund debt-driven AI infrastructure spend.

News OpenAI

OpenAI and Broadcom unveil LLM-optimized inference chip

OpenAI and Broadcom introduce Jalapeño — a custom inference chip designed for LLM workloads.

Why it matters
  • OpenAI's most concrete public commitment to non-Nvidia silicon for its own inference workloads.
  • Validates the per-workload custom-silicon thesis Amazon, Google, and now OpenAI all share.
  • Pressures Nvidia's pricing power on inference even as Nvidia raises $25B in bonds for capacity.

OpenAI's institutional power moves

Hollywood is publicly making peace with OpenAI — Netflix, A24, Focus, Warner Bros.' Clockwork all in deals. Corporate AI super PACs spent $27M on a single local NY race. OpenAI also published a framework for 'shared standards' on advanced AI, and disclosed GPT-5 Pro helping solve a 3-year-old immunology mystery.

News The Verge AI

Hollywood is bending the knee to OpenAI

Verge reports Netflix, A24, Focus Features, and Warner Bros.' Clockwork all reportedly deciding to partner with OpenAI on creative work.

Why it matters
  • Major-studio consolidation around OpenAI rewrites the year of Hollywood-AI hostility into a partnership story.
  • Removes a key narrative obstacle to OpenAI's IPO.
  • Sets the consumer-content template that other labs (Anthropic, Google) will be measured against.
News The Verge AI

Why corporate AI super PACs spent $27 million on a local election

Verge: corporate AI super PACs spent $27M on the NY-12 primary against Alex Bores — a real-money political response to state-level AI regulation.

super PAC spend $27M
Why it matters
  • Concrete evidence that AI lobbying is moving from federal-K-Street to local-election spending.
  • Validates the Sanders-framing of capital concentration in AI politics.
  • Sets the playbook other industry players will follow against state-AG-friendly candidates.

Anthropic ships Claude Tag for Slack; CUGA gets practical

Anthropic released Claude Tag — an always-on AI teammate that learns a company through Slack. IBM Research shipped CUGA, a lightweight agent harness with two dozen working example apps. The agent-surface arms race continues in the workspace layer.

News TechCrunch AI

Anthropic's Claude Tag is learning your company, one Slack message at a time

Anthropic ships Claude Tag — an always-on AI teammate that learns from a company's Slack history.

Why it matters
  • Direct shot at Microsoft Scout's in-Teams positioning from last month.
  • Continuous-learning-from-internal-corpus raises clear privacy and data-residency questions buyers will ask.
  • Pairs with the Anthropic-Stainless dev-tools acquisition: Anthropic owns more of the developer + workspace surface.

Consumer AI lands mixed: Google Home OK, Fitbit catastrophizes, Sony bad, Midjourney suspect

The Google Home Speaker is good but finicky; the Fitbit Air's AI health coach treats every user as in crisis; Sony's Xperia AI Camera Assistant is 'exactly as bad as it looks'; and Midjourney's medical-AI ultrasound pivot is now drawing scrutiny for thin evidence. Meta launched cheaper smart glasses without Ray-Ban.

Papers

4 items

Agent research: world value models, mobile-GUI memory, escape from self-confirmation

World Value Models extend value alignment to robotic manipulation. MemGUI-Agent ships an end-to-end long-horizon mobile GUI agent with proactive context management. An Execute-Distill-Verify paradigm escapes the self-confirmation trap that plagues agentic-experience learning.

Also today