Latest AI News

Broadcom's new Jericho4 ethernet fabric router enables distributed AI infrastructure to connect over one million accelerators across multiple data centers. It delivers 51.2 Tbps deep-buffered Ethernet with lossless, low-latency networking optimized for AI workloads spanning geographic distances. Key innovations include HyperPort technology that improves network utilization by 70%, deep buffering to prevent data loss, line-rate security, and advanced congestion control. This technology addresses the growing need to interlink power-intensive AI data centers across distances while maintaining performance and reliability.

Palantir Technologies has raised its annual revenue forecast amid surging AI demand, reporting second-quarter revenue of $934-938 million that exceeded market expectations. The company's stock has more than doubled this year, reaching record highs as it becomes the S&P 500's best performer in 2025. Growth is particularly strong in the U.S. commercial segment (up 68%), driven by widespread AI adoption across industries. Palantir has increased its full-year growth guidance to 36% and projects third-quarter revenue of $1.08-1.09 billion. Despite this momentum, some analysts express concern about the company's valuation at 277 times expected forward earnings, significantly higher than industry peers.

Tower Semiconductor reported strong Q2 2025 results with revenues of $372 million (up 6% year-over-year) and net profit of $47 million. The company forecasts Q3 revenues of $395 million (±5%), exceeding analyst expectations, with a projected additional $40 million increase for Q4. CEO Russell Ellwanger credited growth to strategic factory repurposing for RF infrastructure and increasing demand from data centers and AI applications. Recent investments include $500 million in its Agrate fab and $300 million in New Mexico, targeting silicon photonics and germanium markets. Analysts view Tower's disciplined operations and strong position in the analog IC market favorably amid industry tailwinds.

Google has agreed to limit AI data center power usage during peak demand times following regulatory concerns about grid strain. This comes as AI systems like ChatGPT drive dramatic increases in energy consumption—Goldman Sachs predicts a 50% rise in global data center energy needs by 2027. Despite Google's data center energy use growing 27% in one year, the company has implemented efficiency measures including advanced TPU hardware and increased renewable energy purchases. The agreement establishes a precedent for balancing AI innovation with infrastructure and environmental priorities.