The long-rumored device from OpenAI will enter the smart speaker market at a premium price, according to a new report.

T-Mobile expanded its partnership while UScellular and Appalachian Wireless will route emergency and rural SMS traffic through Starlink satellites launching this fall.

Executives discussed selling reserved GPU capacity to enterprise customers after internal forecasts showed utilization gaps in several U.S. regions.

The rocket company raised $75 billion at a $2.1 trillion private valuation just as Nvidia, Apple, and peers posted their steepest monthly decline since March 2025.

The custom ASIC, developed in nine months with OpenAI roadmap input, targets better performance per watt than general-purpose accelerators in data-center serving workloads.

The 1.4-million-accelerator system trains foundation models for commerce and logistics, reducing reliance on Nvidia GPUs for internal workloads by a projected 38 percent.

Nothing scrapped the CMF Phone 3 Pro after RAM costs made a budget handset impossible. The bottleneck is high-bandwidth memory for AI servers — and analysts expect shortages through 2027.

The aerospace manufacturer plans to integrate Cursor's code compilation and validation models directly into its spacecraft guidance systems and ground control networks.

A research team at Tokyo Tech builds a prototype transceiver that reduces optical-electrical signal conversion energy by 35%.

The Phoenix campus is shipping Apple and AMD silicon at rates within five points of Taiwan benchmarks, easing doubts about U.S. onshore advanced packaging.

Amazon Web Services has announced a sweeping expansion of its AI portfolio, including the Nova Premier foundation model, a major upgrade to Bedrock Agents, and two new Trainium chips that the company claims deliver 40% better performance-per-watt than H100 GPUs.

The milestone, disclosed at Microsoft Build, positions Copilot 365 as the fastest-growing product in Microsoft's history, surpassing LinkedIn Premium in time-to-reach 50M paid users by three quarters.

Internal benchmarks reviewed by Credence Wire show Apple's new Neural Processing Unit achieves 38 TOPS while consuming 40% less power than the previous generation, enabling real-time translation, video editing, and Siri interactions without cloud connectivity.

The GB300 delivers 2.5 exaflops of AI compute per rack unit, a 3x improvement over the H200, and ships with a new speculative decoding architecture that reduces token generation latency by 60% for long-context applications.

The new platform processes over 100 billion AI inference requests daily and introduces a zero-trust security layer that inspects model outputs for data leakage, prompt injection, and unauthorized PII exfiltration in under 3 milliseconds.