OpenAI announced its custom Jalapeño AI inference chip outperforms Nvidia's GB300 in throughput per watt and response latency in internal tests. Designed specifically for inference tasks, the chip operates at about 700 watts and excels on larger workloads, including OpenAI's unreleased models. Built on TSMC's advanced N3P process with high-bandwidth memory, the Jalapeño chip offers top-tier performance and scalability, with plans to deploy it later this year and ramp production through 2027. OpenAI is also developing second- and third-generation chips to further improve efficiency and power.
Broadcom Inc. shares have been roughly flat year-to-date but have underperformed recently, dropping 15% since a previous sell recommendation. The company faces increasing competition in the custom AI ASIC market, where it once held a dominant 70% sha...
Micron, Broadcom, and CoreWeave are leading the AI semiconductor and cloud infrastructure surge, each with bullish price targets for 2027. Micron aims for $1,200 driven by strong memory demand and $100 billion in contracted revenue. Broadcom targets ...
Broadcom is rated a hold ahead of its September 2 earnings report due to high market expectations, strong AI-driven revenue growth, but declining gross margins. The company’s AI semiconductor revenue rose 143% year-over-year in Q2, with over 70% mark...
As AI infrastructure expands, the demand for data movement between chips and data centers grows, spotlighting the connectivity layer as a key profit area. Arista Networks leads in Ethernet switching with strong revenue growth and a bullish outlook, B...
OpenAI announced its first in-house AI accelerator chip, Jalapeño, which reportedly outperforms Nvidia's Blackwell-generation systems in AI inference tasks, delivering up to 1.9 times more throughput and significantly lower latency. The chip, develop...
