AI accelerators
Coverage of AI accelerators in the Nexus archive.
- Chinese companies are ditching Nvidia’s advanced accelerators for domestic AI suppliers
Chinese companies are shifting from Nvidia's AI accelerators to domestic alternatives amid US-China tensions, with 46% of future budgets allocated to local products. Tencent, Alibaba, and Huawei are leading this transition, supported by a 2 trillion yuan government initiative to build AI-driven data centers using domestic technologies.
- Nvidia Is World’s Most Valuable Company and Cheaper Than Hershey
Nvidia has become the world’s most valuable company and is currently cheaper than Hershey. Jensen Huang confirmed that three major memory chipmakers are certified to supply high-bandwidth products for Nvidia’s AI accelerators.
- Broadcom and Apple extend custom silicon pact to 2031
Broadcom and Apple have extended their custom silicon partnership until 2031, with Broadcom supplying components like Wi-Fi, Bluetooth, and charging chips. The collaboration, ongoing since the late 2000s, includes developing custom ASICs for multiple Apple product generations.
- DeepSeek's new models are so efficient they'll run on a toaster ... by which we mean Huawei's NPUs
DeepSeek has released DeepSeek V4, an open weights large language model that significantly reduces inference costs and supports Huawei's Ascend AI accelerators. The model is now available in preview and aims to rival top American proprietary LLMs.
- Forget one chip to rule them all: With TPU 8, Google has an AI arms race to win
Google unveiled two new in-house AI accelerators at its Cloud Next conference, designed to enhance training efficiency and reduce model serving costs. The company is shifting from x86 architecture to Arm-based Axion cores for its TPU 8 chips, signaling a strategic move in the AI hardware space.