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.
Z.ai released GLM-5.2, a flagship coding model with a usable 1-million-token context, shipping it to paying subscribers first while withholding benchmarks, an API and open weights.
Tech firms have cut close to 150,000 jobs in 2026 while citing AI, even as AI-driven IPOs mint vast new fortunes, sharpening questions about who benefits.
Google AI Overviews liability sets a precedent as a German court rejects the click-to-verify defense, a ruling that could reach ChatGPT, Claude and Perplexity.
Anthropic’s Fable 5 and Mythos 5 shutdown, the first US export control on an LLM, has Carney and EU lawmakers pushing to reduce reliance on American AI.
Anthropic has disabled Claude Fable 5 and Claude Mythos 5 globally after a U.S. government export-control directive restricted access to the models for foreign nationals, prompting the company to challenge the decision publicly.
Mistral’s reported €3 billion round would fund a 1-gigawatt European compute buildout, yet leaves the OpenAI rival far smaller than US and Chinese rivals.
SK Hynix, Nvidia’s largest memory partner and the HBM market leader, awaits SEC clearance for an ADR listing as its stock trades at a persistent Korea discount
Anthropic committed $150 million to Claude Corps, a fellowship paying 1,000 early-career workers to spend a year embedding AI in US nonprofits.
SpaceX is pitching investors on a vertically integrated AI strategy spanning chips, data centers and even compute satellites in orbit, bankrolled by Starlink’s profits.
SpaceX IPO sold 555.6 million shares at a fixed $135, nearly tripling the prior record as heavy demand bets on Musk despite steep losses and no path to profit.