OpenAI's first silicon arrives. OpenAI and Broadcom unveiled Jalapeño on June 24 — OpenAI's first "Intelligence Processor," an accelerator built specifically for LLM inference. It's purpose-built for inference — running models in response to user commands — rather than training, which still leans on Nvidia. Why it matters: even small cuts to inference cost reshape OpenAI's brutal unit economics ahead of a rumored IPO. TechCrunch · VentureBeat BroadcomTechCrunch

Figma absorbs the workflow. At Config 2026, Figma announced code layers, native animation and motion support, AI-generated shader effects, and custom AI skills — teams can now clone repos into the canvas and build interactive elements directly. Why it matters: Figma is squeezed between AI-native creation tools and coding agents, and its answer is to pull more of the surrounding work into the canvas. ODSC Open Data ScienceOpen Data Science

Qwen rethinks how agents learn. Alibaba's Qwen team introduced Qwen-AgentWorld, a model that simulates seven different agent environments in one system and learns to model the environment itself, not just operate inside it. Why it matters: training agents in simulated environments instead of the real world could make them more autonomous and far cheaper to train. Medium — Akpovi MediumMedium

The AI IP war just named its biggest target yet.

Anthropic told U.S. senators and White House officials that operators linked to Alibaba's Qwen lab used nearly 25,000 fraudulent accounts to extract Claude's capabilities between April and June, in what it called the largest distillation campaign yet against a U.S. AI company. The campaign — described in a June 10 letter to Senate Banking Committee leaders Tim Scott and Elizabeth Warren and first reported by Bloomberg on June 24 — allegedly ran from April 22 to June 5 and generated more than 28.8 million Claude interactions. Distillation feeds carefully built prompts into a frontier model, harvests the outputs, and uses them to train a cheaper rival. The Next WebTech Times

The accusation is consequential because it's the first time Anthropic has named a major Chinese technology conglomerate, rather than smaller startups like the DeepSeek, MiniMax, and Moonshot labs it flagged in February. Several outlets stress these figures are Anthropic's allegation and remain unverified — Alibaba denies the allegations. The backdrop is tense: the Pentagon added Alibaba to its Chinese military companies blacklist on June 8, and Alibaba has sued the Defense Department to be removed. The Next Web + 2

The real takeaway for builders is structural. Because a distillation query looks identical to a legitimate one at the API level, the only foolproof defense is refusing access altogether — which conflicts with the core business model of a commercial AI company. Expect tighter API terms of service and routine access-gating on the most sensitive models. The Next Web · TechTimes Tech Times

OpenAI builds the full stack.

Jalapeño is an ASIC — less flexible than Nvidia's GPUs but cheaper and tunable for specific tasks — and OpenAI says it designed the chip in nine months, an unusually fast turnaround. Notably, the companies attributed that speed partly to using OpenAI's own models to accelerate the chip design process. Engineering samples are already running workloads in the lab, with initial deployment targeted for the end of 2026 and expansion in the years ahead. CNBC + 2

The strategic logic is about survival economics. As OpenAI lays groundwork for a heavily anticipated 2026 public offering, Jalapeño signals to investors that it has a plan to dig out of its financial hole — if it can drive down inference costs, it can recoup some of what it spends on training. Broadcom's CEO told CNBC that compute demand from its customers is "insatiable," extending into 2027 and 2028. The bet: owning more of the stack — chips, kernels, networking, serving systems — lets OpenAI optimize every layer around the same goal. CNBC · Tom's Hardware VentureBeatCNBC