OpenAI Jalapeño AI chip delivers strong perf per watt for LLM inference after nine months of design with Broadcom, expanding full-stack infrastructure.

OpenAI and Broadcom launch Jalapeño, a custom AI chip built for LLM inference to raise performance, efficiency and scale.

According to Richard Seroter, Google Cloud Run now supports NVIDIA RTX 6000 PRO Blackwell GPUs for AI workloads, enabling teams to serve 70B-plus parameter models without managing underlying infrastructure (source: Richard Seroter on X and Google Cloud blog). He highlights that drivers are pre-installed, no capacity reservations are required, and serverless instances offer 20 to 44 vCPUs with 80 to 176 GiB memory to streamline large language model inference and other high‑throughput tasks (source: Richard Seroter on X and Google Cloud Run documentation). This update allows on-demand scaling of LLM inference on Cloud Run while removing GPU infrastructure administration overhead for developers (source: Richard Seroter on X and Google Cloud blog).

According to @paoloardoino, Tether has open-sourced an edge-first generalized LLM inference and LoRA fine-tuning framework built for heterogeneous GPUs, highlighting a concrete AI tooling release that crypto traders can verify directly. Source: @paoloardoino on X: https://twitter.com/paoloardoino/status/1995871771875283434 According to @paoloardoino, the announcement does not include any token, pricing, or on-chain integration details, which limits immediate valuation signals but marks a notable technical step by the USDT issuer into AI infrastructure. Source: @paoloardoino on X: https://twitter.com/paoloardoino/status/1995871771875283434 According to @paoloardoino, the post frames this as the start of an AI ubiquity era, indicating Tether’s strategic positioning around edge AI compute and generalized model inference capabilities that may shape narrative monitoring for traders. Source: @paoloardoino on X: https://twitter.com/paoloardoino/status/1995871771875283434