// NODE-02 · NEURAL FEED · DAILY TRANSMISSION //
June 24, 2026 · WINDYVIEWS.COM
Noam Shazeer — a co-author of the foundational 2017 "Attention Is All You Need" paper that introduced the Transformer architecture — is leaving Google DeepMind to join OpenAI as Lead for Architecture Research. In the role he would own the underlying neural-network structures behind all of OpenAI's models, marking one of the highest-profile talent moves of the year in the frontier-lab arms race.
Anthropic launched Claude Fable 5 and Claude Mythos 5 on June 9, a new capability tier above Claude Opus 4.8. The release was significant enough that the U.S. government issued an export-control directive on June 12 citing national-security concerns, underscoring how frontier model weights are increasingly treated as strategic assets.
Claude Opus 4.8 took the #1 spot on the Artificial Analysis Intelligence Index with a score of 61.4 — the first model to clear 60 by a clean margin. The benchmark aggregates reasoning, coding, and knowledge tasks, and the result puts Anthropic narrowly ahead in a field where frontier scores have been clustering tightly.
On June 13, China's Zhipu AI released GLM-5.2, a large Mixture-of-Experts model with a usable 1-million-token context window, shipped under a permissive MIT license. The open-weight release continues a trend of Chinese labs pushing frontier-class capability into freely deployable models, intensifying competition with closed Western systems.
Qualcomm is in early talks to acquire Tenstorrent for between $8 billion and $10 billion, signaling a serious bid for a seat at the AI-hardware table currently dominated by Nvidia and AMD. A deal would give Qualcomm a credible training-and-inference accelerator line as demand for non-Nvidia silicon accelerates.
Nvidia and Microsoft introduced RTX Spark, a superchip integrating GPU and CPU with up to 1 petaflop of compute, designed to run agentic AI workloads locally on PCs. New Windows devices built around the chip are slated to ship in late 2026, pushing more inference to the edge and out of the data center.
China announced a $295 billion, five-year government AI infrastructure program — roughly $59 billion per year in state-directed spending on compute, data centers, and model development. The scale signals Beijing's intent to compete head-on with U.S. private-sector buildouts through coordinated national investment.
Google announced its biggest Search overhaul in over 25 years, with a new AI Mode running on Gemini 3.5 Flash and a redesigned, AI-powered results experience. The shift moves Search further from a list of links toward synthesized, conversational answers — a structural change with major implications for publishers and the open web.
The Trump administration issued an executive order, "Promoting Advanced Artificial Intelligence Innovation and Security," directing the Treasury to form an AI cybersecurity clearinghouse and advancing a deregulatory, innovation-first stance. It builds on the National Policy Framework's push to preempt "unduly burdensome" state AI laws while preserving carve-outs for child safety, compute infrastructure, and government procurement.
The frontier race is now fought on three fronts at once: models (Anthropic's Fable/Mythos 5 and chart-topping Opus 4.8, plus China's open-weight GLM-5.2), silicon (Qualcomm's Tenstorrent bid and the Nvidia–Microsoft RTX Spark pushing AI onto PCs), and policy (a $295B Chinese state plan against a deregulatory U.S. executive order). Talent remains the scarcest input — Noam Shazeer's jump to OpenAI shows architecture expertise is being contested as fiercely as compute. The throughline: capability is diffusing fast, but governments are increasingly treating model weights and chips as strategic national assets.