A KV cache, or key-value cache, speeds up autoregressive transformer inference by storing attention information calculated for tokens that have already been processed. Without the cache, the model would recompute those keys and values each time it generates another token. Reusing them reduces repeated work and improves response latency, especially for long prompts and outputs. The tradeoff is memory: cache size grows with sequence length, batch size, layer count, and model dimensions. Production systems manage this cost with paging, quantization, eviction, prompt caching, and request scheduling. KV caching changes inference efficiency, not the model’s learned knowledge, and cached data must be protected when prompts contain sensitive information.
Alphabet plans to raise $80 billion through stock offerings and a strategic investment from Berkshire Hathaway to expand AI infrastructure. The funding will support growing demand for Google’s AI services and computing capacity.
Major AI companies are increasingly focusing on humanoid robots as the next frontier of artificial intelligence. The shift is drawing investments from Nvidia, OpenAI, Tesla, Meta, and a growing group of robotics startups seeking to bring AI into the physical world.
OpenAI has launched a dedicated robotics division focused on building humanoid robots for infrastructure projects and future consumer use. The move expands the company’s ambitions beyond software and places it in direct competition with Tesla’s Optimus program.
SoftBank plans to invest €45 billion in three large-scale AI data centers in France as part of a broader €93 billion foreign investment push announced at the Choose France summit.
ElevenLabs has introduced Dubbing v2, a new AI dubbing model that preserves a speaker’s tone, pacing, and emotional delivery across more than 90 languages.
Intel has disclosed new details about Crescent Island, its next-generation AI accelerator designed for data centers and agentic AI workloads. The chip is expected to launch later this year as Intel seeks to compete more aggressively with Nvidia in the rapidly growing AI infrastructure market.
Nvidia has launched Vera, its first CPU designed specifically for AI agents and large-scale AI infrastructure. The processor promises up to 1.8 times faster performance than traditional x86 CPUs and will be adopted by major AI labs, cloud providers, and server manufacturers.
Nvidia has unveiled its first Arm-based PC processor, the RTX Spark superchip, which will power a new generation of Windows laptops and desktops.
OpenAI has launched the Rosalind Biodefense program and expanded access to its GPT-Rosalind model for trusted developers and government partners. The initiative aims to accelerate biodefense, pandemic preparedness, and public health research using advanced AI tools.
Nvidia has committed at least $6.5 billion to photonics companies in recent months as it seeks to overcome AI infrastructure bottlenecks. The investments target optical technologies that could reduce energy consumption and improve data transfer across future AI systems.