IBM and NASA released the NASA-IBM Lunar Foundation Model after training it on roughly 2 million lunar image tiles spanning two very different spatial scales. The open-source system gives researchers a shared starting point for mapping craters, studying volcanic features, and estimating where polar ice may remain stable. The release changes how scientists can approach decades of observations from separate instruments. Instead of building a specialized model from scratch for e

Micron has demonstrated a 512GB server memory module that reaches 9,200 MT/s, yet its planned production window remains almost a year away. The Micron 512GB DDR5 module gives the company another connection to AI infrastructure spending beyond high-bandwidth memory, or HBM. It also creates an immediate tension between impressive laboratory specifications and the slower work of qualifying a new component for commercial servers. The module can place 12TB of DDR5 memory inside a

Corning helped finalize the first SDM4 specification, giving four-core fiber a shared design after months of coordination among four optical infrastructure suppliers. The Corning SDM4 multicore fiber effort targets a growing physical problem inside AI data centers. Larger accelerator clusters need more optical links, but cable trays, connectors, and installation teams cannot expand without limits. AFL, Corning, Sumitomo Electric Industries, and TeraHop released Version 1.0 of

NASA has released a lunar AI model trained on nearly two million data tiles, opening a new route through decades of Moon observations. The NASA lunar AI model can be adapted to map craters, outline unusual volcanic terrain, and estimate where polar ice remains stable. Its release changes who can build those tools and how much task-specific data they need. NASA developed the model with IBM Research and academic partners, then published its weights, datasets, benchmarks, and fi

Meta AI deepfake rules failed in two Facebook cases involving women targeted through synthetic video, according to decisions published on September 17. The Oversight Board overturned Meta’s decisions to leave both posts online and called the company’s safeguards “consistently and fundamentally inadequate.” One video impersonated a Scottish Labour councillor and falsely placed inflammatory comments about refugees in her mouth. The other transformed an interview with a young Mu

SK Hynix raised the cash share of employee bonuses to 50% after workers rejected an earlier agreement by only 25 votes. The revised SK Hynix bonus deal ends an immediate labor dispute at a critical supplier of memory for AI systems. It also reveals a deeper conflict over who should carry the financial risk when extraordinary chip profits become employee compensation. Production workers approved the revised agreement on September 16 with 57.08% support. The deal splits profit-

Kaiko funding led by S&P Global has extended the crypto data provider’s Series B to $110 million, despite unresolved questions around round-the-clock tokenized markets. DRW Venture Capital, Nasdaq Ventures, Royal Bank of Canada, and Susquehanna Private Equity Investments joined the investment. Their participation makes this more than another crypto financing announcement. The investors represent several parts of the machinery behind capital markets. They include a financial i

LUMI is getting a digital twin as Finland prepares its Kajaani computing center for denser AI systems and far more demanding rack conditions. The project gives engineers a virtual environment for testing energy, cooling, and heat-recovery decisions before applying them to physical infrastructure. That matters because the facility is entering a difficult transition. Its current supercomputer already combines concentrated computing loads, warm-water cooling, and municipal heat

Nvidia RTX 40 Multi Frame Generation has crossed an unexpected line: an unofficial mod enables 4X output on Ada GPUs despite the feature's Blackwell-only positioning. Tests published on September 12 show the modification running in Cyberpunk 2077 across cards from the RTX 4090 to the RTX 4070. The results do not erase every advantage held by the RTX 50 series. They do show that Ada hardware can execute a feature Nvidia tied closely to its newer architecture. That distinction

NVIDIA officially launched DLSS 5 in one game, despite six months of fierce criticism over its generative approach to graphics. The NVIDIA DLSS 5 controversy is not simply another argument about sharper pixels. It concerns whether an AI model should reinterpret finished artwork while a game is running. NVIDIA calls DLSS 5 its most important graphics advance since real-time ray tracing arrived in 2018. Critics see something closer to a costly beauty filter that replaces delibe

Ilit Raz pleaded guilty to securities fraud after prosecutors said fabricated business records helped Joonko obtain $27 million from investors. The Ilit Raz securities fraud plea converts a disputed startup collapse into a criminal admission. It also exposes a due diligence problem that reaches beyond one failed AI hiring company. Joonko presented itself as an artificial intelligence platform that connected employers with candidates from underrepresented backgrounds. Accordin

Nebius faces a new data center lawsuit after one property changed hands three times in a day, rising from $17.2 million to $83.5 million. The September 5 complaint does not simply challenge zoning, construction noise, or the environmental effects of a large computing campus. A Regions Financial shareholder alleges that bank insiders and outside parties deprived Regions Bank of $66.3 million through a coordinated land transaction. Those accusations remain unproven, and the def

Apple researchers have introduced a protein model with one training stage, despite a field increasingly built around specialized pipelines. The Apple SimpleDesign protein design system generates amino-acid sequences and three-dimensional structures together. Its central claim is architectural simplicity, not proven biological function. That distinction matters. SimpleDesign produced competitive results in computer benchmarks after training on more than 2 million sequence-stru

Analog Devices agreed to buy Alif Semiconductor for $1.35 billion, turning the Analog Devices Alif acquisition into a direct bet on local AI processing. The cash deal also includes up to $200 million in contingent consideration. It gives Analog Devices, commonly called ADI, processors designed to run machine learning beside sensors rather than in distant data centers. The transaction is more than another chip portfolio expansion. ADI already converts physical signals, such as

Google has begun the Google Pics launch after months of previews, putting Nano Banana image generation and editing directly inside everyday Workspace workflows. The conflict is no longer only about which AI model creates the best picture. Google now wants to control where teams generate, revise, translate, approve, and reuse that picture. Pics is available as a standalone product at pics.new and is rolling out across Google Workspace. It can generate several options from one

Panmnesia and Meta have proposed a CXL design that coordinates up to 960 accelerators inside one coherence domain, despite today’s rack boundaries. The Panmnesia Meta CXL architecture treats an AI datacenter as one managed computing system, rather than a collection of networked servers. That framing challenges the dominant approach to scaling AI infrastructure. The proposal appears in a CXL datacenter review published by Nature Reviews Electrical Engineering on August 10, 202

Qdrant released a 10-billion-document search dataset, confronting a basic conflict in AI infrastructure: production systems are huge, while many accepted tests remain comparatively small. The Qdrant FineWeb 10B benchmark gives researchers a public corpus with dense and sparse vectors, filtered queries, and exact nearest-neighbor results. Its scale makes familiar shortcuts harder to hide. The release does not declare Qdrant the fastest vector database. It creates shared infras

Amazon has launched Prime Video AI lip sync on one German series, while keeping human-recorded English dialogue and digitally changing the actors’ mouth movements. The feature is available globally on the English dubs of Maxton Hall Seasons 1 and 2. Amazon will also apply it to the third season, scheduled for December 9. More titles are planned, although the company has not identified them. That limited launch creates a larger conflict. Traditional dubbing changes the soundtr

Naver defended its AI training practices in court despite three broadcasters claiming their articles were used without permission. The Naver AI copyright lawsuit now centers on whether older news-sharing contracts authorized uses that publishers say they never knowingly approved. KBS, MBC, and SBS sued Naver in January 2025 over material allegedly used to train HyperCLOVA and HyperCLOVA X. They seek damages and an order stopping further training with their news content. Naver

Xiaomi founder Lei Jun smiled after an audience member shouted a price prediction during the September 7 Pengcheng launch. The brief exchange became technology news because it captured the tension surrounding Xiaomi’s first range-extender SUVs. Buyers wanted another aggressive deal, while Xiaomi needed to protect margins in a shrinking domestic car market. The viral clip does not provide a reliable record of the exact words from the audience. Xiaomi’s broadcast and available