All published stories.
AMD's EPYC Venice is the first x86 server CPU on TSMC's 2-nanometer process, and its Helios rack pairs 72 MI455X accelerators with 31 terabytes of HBM4. Microsoft will deploy the platform across Azure in the second half of 2026.
OpenAI lifted its projected compute spending through 2030 to roughly $750 billion, up 25 percent from the figure it set five months ago, and committed $20 billion to Project Camellia in Georgia, the first data center campus it will develop itself.
Z.AI, formerly Zhipu, has switched on a roughly 1-gigawatt data center running entirely on Chinese-made chips to train its GLM models, a direct answer to its January 2025 US blacklisting.
White House OSTP director Michael Kratsios has publicly accused Moonshot AI of distilling Anthropic's Fable to build Kimi K3, and of using Nvidia GB300 servers in Thailand to train it. The charge names a specific company and model for the first time, days before K3's open weights are due for release.
Moonshot AI's 2.8-trillion-parameter Kimi K3 matched or beat Claude Fable 5 and GPT-5.6 Sol on several company-run coding and agentic tests. Moonshot says the model still trails both systems overall.
Governor Kathy Hochul has paused incomplete state permit applications for data centers capable of consuming at least 50 megawatts while New York develops new environmental and grid rules.
A Commerce official told Congress this week that a 'trivial' number of H200 chips have shipped to China under new licenses. It is the clearest sign Washington has chosen to sell into China rather than wall it off, the opposite of what the GAIN AI Act tried to require.
Digi Power X and AIB Data Centers each filed updates last week casting themselves as AI infrastructure companies. Both were crypto ventures until recently. Their filings describe businesses built on electrical capacity and leased land, with customers supplying the chips.
A technique called MemGhost uses a single crafted email to make a persistent AI agent store attacker-controlled memories, hide the change from the user, and act on it in later sessions. It succeeded most of the time against agents built on leading models, and slipped past existing defenses.
The National Science Foundation, a primary funder of fundamental AI research at US universities, gave notice of a policy barring the researchers it funds from collaborating with institutions on federal restricted-party lists. The lists are dominated by Chinese universities and labs, and AI is among the fields where the split will bite hardest.
Huawei intends to enter South Korea's AI accelerator market in the fourth quarter with its Ascend chips and Atlas SuperPod systems, pitching them as cheaper alternatives to Nvidia. The move pushes a Chinese vendor into one of the United States' closest allies and the home of Nvidia's biggest memory suppliers.
Security researchers at Sysdig documented JADEPUFFER, what they call the first ransomware operation driven end to end by a large language model, with no human at the keyboard. The agent broke in, moved through the network, encrypted a database, and left a ransom note, adapting to errors in seconds.