OpenAI and Broadcom Unveil Jalapeño Inference Chip — CNBC Jalapeño is the first chip in a planned family of LLM inference accelerators optimized for performance per watt, designed end-to-end in just nine months with AI assistance. Built for gigawatt-scale data center deployments, the chip signals OpenAI’s ambition to own the entire stack — compute, model, and product — rather than remain dependent on Nvidia.

🚀 Frontier Models & Hardware

Introducing Computer Use on Gemini 3.5 Flash — Google Google launched native computer-use capabilities for Gemini 3.5 Flash, enabling the lightweight model to interact directly with desktop interfaces via continuous screenshots, executing clicks, scrolls, and keystrokes across arbitrary software. The move brings agentic desktop control beyond frontier-tier models to a fast, cost-efficient option.

Anthropic’s Claude is winning over paid consumers, a market owned by ChatGPT — TechCrunch New data shows Claude gaining meaningful share among consumers willing to pay for AI subscriptions, a segment long treated as ChatGPT’s uncontested territory. The shift correlates with Claude’s reputation for accuracy and coding quality among power users.

OpenAI Updates GPT-5.5 Instant — OpenAI OpenAI rolled out an upgraded GPT-5.5 Instant inside ChatGPT for both free and paid tiers, targeting more natural and useful day-to-day interactions.

GLM-5.2 is the step change for open agents — Interconnects.ai Despite modest benchmark improvements, GLM-5.2 has unlocked a wide range of agentic use cases in coding harnesses and earned widespread community praise after hands-on use.

Fable 5 has reportedly reappeared in Amazon Bedrock — Thread Reader String changes in Claude Code v2.2.190 hint at preparations for Fable 5’s return, with the model spotted surfacing in Bedrock.

🏭 Infrastructure & Energy

As AI Companies Race for Power, Amazon and Google Have the Lead — WSJ Amazon holds the incumbent advantage in the data center power race through 2030, built on two decades of infrastructure investment, but Google is closing the gap faster than expected. The article maps how hyperscalers are positioning to meet AI’s surging electricity demands.

Tesla, Sunrun team up on 16 GW virtual power plant for data centers — Electrek Tesla, Sunrun, and Renew Home will aggregate 16 GW of home batteries and smart thermostats into the largest distributed power plant in the US, with 300 MW in Virginia ready for immediate deployment to serve data center demand.

Amazon ups India bet with fresh $13B AI infrastructure investment — TechCrunch Amazon’s latest India commitment underscores the global race to plant AI infrastructure as demand continues to outpace supply.

Qualcomm lands Meta as first named customer for its Dragonfly data centre chips — The Next Web

🤝 Industry Moves & M&A

Gemini Researchers Join Anthropic — TechCrunch Jonas Adler and Alexander Pritzel have left Google for Anthropic, continuing an accelerating wave of top AI talent departures following Noam Shazeer and DeepMind director John Jumper. The pattern is emerging as a structural challenge for Google’s AI ambitions.

Anthropic and Alibaba Launch Joint AI Model Distillation Campaign — CNBC The two companies are collaborating on an open-source framework to distill frontier reasoning from large models into efficient edge deployments, combining Anthropic’s safety alignment with Alibaba’s cloud scale.

General Intuition’s $2.3B bet that video games can train AI agents for the real world — TechCrunch The startup raised $320M on the thesis that millions of hours of gameplay data can give AI agents something closer to human intuition for real-world decision-making.

Adobe acquires image and video enhancement tool maker Topaz Labs — TechCrunch

Cerebras stock plunges after earnings as CEO says margin outlook was misunderstood — TechCrunch

AI was supposed to kill engineering jobs, but new data suggests they’re the most resilient — TechCrunch

💻 Developer Tools

Kiro IDE 1.0: Agent Focus Mode, Permissions, Custom Agents, and More — Kiro Kiro IDE reached 1.0 with parallel agent orchestration via Agent Focus Mode, granular permission controls, custom agent profiles, natural language hook creation, and session export — a significant maturity milestone for AI-native development environments.

Claude Code 2.1.191 — Anthropic Adds /rewind support to resume conversations from before /clear was run, fixes scroll position jumping during streaming, and permanently stops background agents from restarting after being stopped.

NVIDIA NeMo AutoModel accelerates fine-tuning of MoE models — NVIDIA / Hugging Face Expert Parallelism and DeepEP fused communication kernels deliver 3.7x training throughput gains and 32% memory reduction for large MoE architectures like Qwen3 and DeepSeek V3.

Notes on Amazon v. Perplexity — Educated Guesswork Amazon is suing Perplexity for its Comet browser identifying as Chrome rather than as an agent — a case with broad implications for AI agent transparency and user control on the open web.

🏛️ Policy & Safety

Anthropic’s White House Negotiations Are Reportedly On Track After Dario Amodei Was Replaced — Gizmodo Talks between the White House and Anthropic stalled under Dario Amodei’s direct involvement but are now progressing with co-founder Tom Brown leading discussions alongside Head of Public Policy Sarah Heck. No details on the substance of the negotiations were disclosed.

$500 Million Intercept Fund Launched to Eliminate the Common Cold — Time Backed by Stripe and Anthropic, the Intercept project will fund air filtration research and multi-virus treatment development aimed at making respiratory infections rare — a notable example of AI companies investing in adjacent health infrastructure.

Europe is pushing back on Washington’s chip war — TechCrunch

claude-sonnet-4-6 · 2026-06-25T10:00:00Z