Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
OpenAI and Broadcom revealed Jalapeño, a reticle-sized inference ASIC co-developed from design to tape-out in just nine months — possibly the fastest high-end ASIC cycle ever. Broadcom CEO Hock Tan says it delivers roughly 50% cost savings versus typical AI GPUs, with initial deployment targeted for the end of 2026. It's OpenAI's clearest move yet to "build the full stack" and chip away at Nvidia's grip on inference compute.
TechCrunch reports that Jonas Adler and Alexander Pritzel, both key contributors to Gemini, are leaving Google for Anthropic — fresh departures stacking on top of John Jumper's Nobel-winning AlphaFold exit and Noam Shazeer's jump to OpenAI. The drumbeat of foundational researchers walking out the door, as OpenAI and Anthropic prep for public offerings with rich equity packages, is fueling real questions about whether DeepMind can hold the front of the race. The concentration of losses in a single window is unprecedented for Google.
Internal documents show Meta is building a standalone prediction-market app, codenamed "Antwerp" and "FBForecast," to take on Kalshi and Polymarket in a sector some peg at $1 trillion. Llama would auto-generate markets from trending topics, make personalized recommendations, and — most striking — resolve markets in near-real-time, giving the AI final say over what did or didn't happen. Letting a model adjudicate real-money wagers is a bold and legally fraught use of generative AI.
DeepSeek sealed its first external funding — more than 50 billion yuan ($7.4 billion) — from Tencent, CATL, NetEase and JD, valuing the lab at an estimated $52B-$59B. The deal carries an unusual structure: a state-backed fund holds the only voting rights, while founder Liang Wenfeng personally put in 20 billion yuan. It's a six-fold valuation jump that cements China's open-weight champion just as US export curbs tighten.
Days after its $6.3B SpaceX compute deal, US open-source bet Reflection AI is in talks to raise roughly $2.5 billion at a $25 billion valuation, with JPMorgan's Security and Resiliency Initiative and existing backer Disruptive expected to join. Founded by ex-DeepMind researchers Misha Laskin and Ioannis Antonoglou, the company was worth just $545 million a year ago — a near-46x climb. Washington and Silicon Valley clearly want an American open-weight answer to DeepSeek.
Z.ai's GLM-5.2, a 744-billion-parameter MIT-licensed model, has taken the top open-weight spot on the Artificial Analysis Intelligence Index and is besting GPT-5.5 on several long-horizon coding benchmarks at roughly one-sixth the price. Developers locked out of Fable 5 have flocked to it, drawn by API pricing dramatically cheaper than the closed frontier. The gap between open and closed models keeps narrowing — fast.
Nine of ten venture rounds reported on June 24 carried a direct AI angle, led by Assort Health's $120M Series C (Menlo Ventures) for healthcare agents and Taktile's $110M Series C (Goldman Sachs Alternatives) for financial decision automation. The theme is unmistakable: capital is pouring into production AI systems that make real decisions, move money, and govern other agents. Governance startups like Runlayer raising alongside the agents they police shows the market maturing.
The Humanoid Robot Forum at Automate 2026 in Chicago capped a milestone stretch: Figure's BotQ factory is now producing the Figure 03 at one robot per hour, and Boston Dynamics' electric Atlas is shipping its first 2026 units to Hyundai and DeepMind. Physical AI is moving from demo reels to real factory floors. The bottleneck is shifting from "can it walk" to "can you build them fast enough."
A new WEF report released today at the Annual Meeting of the New Champions finds banks and insurers shifting from AI experimentation to broad deployment, with trust, governance, and human oversight now the central concern. The next phase, the report argues, will be won or lost on accountability frameworks, not raw capability. It's a reminder that the hardest part of enterprise AI isn't the model — it's the controls around it.
A former Infosys chief executive has unveiled a new venture aiming to upend the traditional IT-services model with AI-native delivery. The pitch directly targets the labor-arbitrage business that built the outsourcing giants — using agents to do what armies of engineers once did. If it works, it's an existential nudge to a multi-hundred-billion-dollar industry.
