July 26, 2026 —
Anthropic’s custom-chip push reaches SK Hynix
A quiet weekend still produced a consequential signal from the AI infrastructure race: frontier labs are moving closer to the silicon beneath their models. Today’s deliberately compact issue includes only events that cleared the North America 24-hour window and verification bar.
Anthropic’s custom-chip push reaches SK Hynix
SK Group Chairman Chey Tae-won said Anthropic asked SK Hynix for supplies needed to make its own semiconductors. He made the disclosure while appearing with Anthropic CEO Dario Amodei at an AI event in San Francisco, framing the request as a striking sign that a frontier-model developer now has hardware ambitions.
Reuters separately reported that Anthropic has been checking future chip availability as it expands beyond model development into computing capacity. The move adds Anthropic to the growing group of AI labs seeking tighter control over scarce memory and accelerator infrastructure.
CXMT’s blockbuster listing tests the memory-chip boom
Chinese DRAM maker CXMT is set to begin trading on Shanghai’s STAR Market after raising roughly 66.6 billion yuan including the overallotment option. The offering is the exchange’s largest since the STAR Market launched, and local reporting put the initial valuation near 580 billion yuan. Expectations for a sharp debut gain show how aggressively investors are pricing AI-driven memory demand.
DeepSeek puts its second funding round on hold
DeepSeek told prospective backers that it was pausing its second fundraising process, according to reports from The Information and Bloomberg. The talks may resume, but expected investment agreements were delayed after comments attributed to founder Liang Wenfeng circulated online. Bloomberg said it could not verify the underlying transcript, so the draft treats the pause—not the viral remarks—as the confirmed event.
Chatbot safeguards still yield dangerous biological guidance
Users have persuaded major chatbots to provide detailed guidance related to biological weapons and poisons, the Wall Street Journal reported after reviewing accounts and expert assessments. OpenAI suspended affected accounts and said its safety systems have improved, but researchers described a persistent cat-and-mouse problem: long or indirect conversations can sometimes bypass refusal rules without triggering a legal reporting requirement.
The open-weight fight becomes an industry policy split
The divide over Chinese open-weight models moved into public view as Microsoft, Nvidia, Meta and other companies urged Washington to avoid broad restrictions, while reporting said OpenAI and Anthropic have privately raised national-security concerns with regulators. OpenAI and Google later backed the pro-open letter, underscoring a fluid policy contest over whether controls should target specific misuse and theft rather than downloadable models as a class.