OpenAI Unveils Jalapeño, Its Custom Inference Chip
OpenAI has announced Jalapeño, a custom inference chip built with Broadcom, joining Google, Apple, and SpaceX in developing proprietary silicon.
104 articles
The chips and systems powering modern AI — GPUs, accelerators, data-center infrastructure, and the hardware roadmaps from Nvidia, AMD, and custom silicon makers driving training and inference.
OpenAI has announced Jalapeño, a custom inference chip built with Broadcom, joining Google, Apple, and SpaceX in developing proprietary silicon.
Amazon SageMaker AI now supports NVIDIA Blackwell GPUs, enabling larger batch sizes and longer sequences for training large models, with P6-B200 instances available for booking.
IBM announced its first sub-1 nanometer chip technology, capable of integrating nearly 100 billion transistors on a fingernail-sized chip, doubling the transistor density of its previous generation.
Cerebras Systems saw its stock drop nearly 20% after reporting first-quarter earnings, despite exceeding expectations.
European officials are resisting U.S. efforts to restrict Chinese access to advanced chipmaking equipment, which could significantly impact ASML's sales.
OpenAI and Broadcom announced Jalapeño, an AI accelerator designed to improve performance per watt for large language models, with early testing showing significant gains over current state-of-the-art solutions.
Princeton researchers use AI to design radio-frequency integrated circuits, achieving record performance and reducing design time by orders of magnitude.
Qualcomm has acquired Modular, a chip startup, for nearly $4 billion, including $300 million for employee compensation.
Eco Wave Power, using NVIDIA AI tools, is developing wave energy technology that could supply 60% of U.S. annual energy needs, according to the Energy Information Administration.
U.S. officials claim ASML may have sent EUV lithography tools to China, though the company denies it. The export ban, enforced since the Trump era, remains in place.
Amazon is exploring selling its Trainium AI chips to third parties, potentially creating a $50 billion business, according to CEO Andy Jassy's shareholder letter.
AMD's analysis reveals that Grok 4.0's 200,000-GPU cluster generates network loads orders of magnitude higher than smaller-scale efforts.