SoftBank-backed SambaNova Systems, a US chip startup, reports surging demand in Asia for its affordable AI chips as an alternative to Nvidia's costly offerings, fueled by the rise of open-source AI models like China's DeepSeek.
Chip giants like Nvidia, AMD, and ASML face billions in losses after Trump's export curbs on selling advanced chips to China, with Nvidia bracing for a $5.5 billion charge, AMD up to $800 million hit, and ASML citing demand uncertainty due to tariff restrictions.
DeepSeek, a Chinese AI platform accused of data theft, spreading propaganda, and violating U.S. export controls by covertly funneling American user data to the CCP and training on material unlawfully obtained from U.S. AI models, has drawn scrutiny from Congress over Nvidia chip sales.
Nvidia grapples with new US export restrictions requiring licenses to sell its H20 AI chips to China, citing national security risks, potentially leading to a $5.5 billion hit in its Q1 2026 fiscal year.
China's tech giants are fueling an AI boom by releasing open-source models, but startups face pressure from investors prioritizing practical applications over foundational research, and US chip restrictions pose challenges that drive innovation to stay competitive.
In a strategic move, Intel is spinning off its Altera FPGA unit into an independent company by selling a 51% stake for $8.75 billion to Silver Lake, enabling Intel to streamline operations while allowing Altera to invest in emerging AI markets like edge computing and robotics.
Safe Superintelligence, founded by former OpenAI chief scientist Ilya Sutskever, secures $2 billion funding at a $32 billion valuation despite having no public product, with major investors including Andreessen Horowitz, Alphabet, and Nvidia backing the 20-employee startup developing AI models on Google's TPUs.
Ex-Google AI Chief Ilya Sutskever's startup Safe Superintelligence (SSI), backed by $1 billion, is tapping Google Cloud's powerful TPU chips in its quest to develop safe, superintelligent AI systems, marking another AI computing deal between Google and a former employee's startup.
Google's Ironwood AI processor, designed for its Gemini AI models and simulated reasoning, unleashes an astounding 42.5 Exaflops of computing power through clusters of up to 9,216 liquid-cooled chips with enhanced interconnect, surpassing previous Google TPUs in speed and efficiency.
IBM has launched the z17 mainframe, engineered specifically for AI and generative AI workloads, featuring high-capacity Telum II processors, compatibility with IBM's Spyre accelerator chips, and new performance management and operating system tools.