Gavin Boulanger / @gavinsbaker: As inference splits into pre-filling and decoding, Nvidia’s Groq deal could enable a “Rubin SRAM” variant…
Trending
- AI is forcing consulting firms to move from hourly billing to fixed or results-based pricing, a transition that is proving slow and difficult for the industry (Mark Maurer/Wall Street Journal)
- DeepSeek details DSpark, a speculative decoding framework for its V4 models, saying it speeds up AI inference by up to 85% and has been tested on Gemma and Qwen (Ben Jiang/South China Morning Post)
- Jefferies: Memory prices are expected to rise 40-50% in the third quarter of 2026 and 30-40% in the fourth quarter as Chinese memory makers are unlikely to provide significant relief in the near term (Hassan Mujtaba/Wccftech)
- Indonesia says tech companies have deleted 4.7 million children’s accounts since government ban – The Diplomat
- Even Fox News Can’t Hide Trump’s Great Failing for Fairness in America
- Austria is pushing the EU to consider hosting Anthropic within its borders, highlighting EU efforts to strengthen the bloc’s independence from US and Chinese technology (Marton Eder/Bloomberg)
- Donald Trump now spends his days counting trees in the park
- Kunlunxin Technology, Baidu’s chip unit, plans an IPO in Hong Kong with a target valuation of $50 billion, asking investors to buy chips worth 3 to 7 times their IPO investment (Qianer Liu/The Information)
