OpenAI and Broadcom said they have co-designed a dedicated ASIC called Jalapeño, optimized for large language model inference and slated for deployment in data centers within the year. The chip is tailored to OpenAI’s technology roadmap, aiming to improve energy efficiency and reduce reliance on third-party GPU providers. The companies said the effort is intended to lessen dependence on external suppliers such as Nvidia as OpenAI pushes deeper vertical integration. No measured performance results have been published yet, and a technical report is expected in the coming months.

A Tesla Model 3 crashed into a home in Houston, killing 76-year-old Martha Avila, after the driver said Autopilot was enabled. Her family is suing Tesla for more than $1 million, alleging defects in Full Self-Driving including Sudden Unintended Acceleration and failed obstacle detection. The National Highway Traffic Safety Administration has opened a formal investigation. Tesla denies the technology was responsible, arguing the crash resulted from human actions at high speed.

The White House issued an executive order moving up deadlines for high-value asset systems to adopt post-quantum cryptography, setting Dec. 31, 2030 for key-establishment schemes and Dec. 31, 2031 for digital signatures. The order applies to federal agencies and critical infrastructure suppliers, requiring them to accelerate replacement of quantum-vulnerable algorithms such as RSA and ECC. Google and Microsoft also tightened their migration timelines to 2029 in March–April 2024.

Bernie Sanders has proposed a $7 trillion AI sovereign wealth fund, funded by a one-time 50% tax on the stock of the largest AI companies and paired with a requirement to split AI and non-AI businesses. The plan would apply to any AI firm with $200 million in annual AI sales, and to new firms once they reach that threshold. Tesla could be swept in because of its ties to xAI and its approach to integrating AI into its business. The proposal is not viewed as likely to pass, but it puts fresh regulatory focus on vertically integrated tech-and-auto companies and has fueled debate over valuation models and compliance costs.

NVIDIA GEAR实验室联合CMU、UC Berkeley推出ENPIRE智能体框架，支持AI编码代理自主设计、测试并优化机器人训练策略，在GPU安装、扎带切割等任务中达99%成功率。该进展凸显NVIDIA在物理AI基础设施层的领先能力，并与Unitree合作推出参考人形机器人、推进与现代汽车集团的量产合作。事件强化市场对其AI硬件生态主导地位及长期变现路径的信心。