Meta is reusing DDR4 memory from retired servers by attaching it to newer machines through a custom Compute Express Link ASIC called Vistara. The Register, citing a Meta paper scheduled for ISCA 2026, reported that the system is in production across millions of servers and has cut the number of machines needed for some inference workloads by 25 percent.
The ISCA 2026 program lists Meta’s Vistara paper as a session on taking CXL from ASIC design and operating-system support into hyperscale deployment. The design gives Meta a path to put older memory back into service as newer platforms move to DDR5-era systems.
Featured source: The Register , ISCA 2026 .
Other Stories
Windows Server 2022 Hotpatch Support Extended Through October 2027
Microsoft said hotpatch update support for Windows Server 2022 Datacenter: Azure Edition is now extended through October 2027, one year beyond the previous October 2026 end date. The company said the monthly hotpatch cadence is unchanged and that administrators already receiving hotpatch updates do not need to take action.
Filed from: Microsoft Learn , The Register .
Palantir Launches Engine for NVIDIA Nemotron Models in Sovereign Environments
Palantir announced an engine for deploying NVIDIA Nemotron open models in sovereign environments, focused on U.S. government agencies and critical infrastructure. NVIDIA said the system combines Palantir AIP, Ontology, Foundry, and Apollo with NVIDIA AI Enterprise and Nemotron models, with controls for data authorization, secure perimeter enforcement, customer-specific isolation, data portability, erasure rights, and auditability.
Filed from: NVIDIA .
HP Scales OpenAI Frontier Partnership
HP Inc. and OpenAI announced that HP will scale its OpenAI Frontier strategic partnership after pilots across customer-facing work, software development, enterprise operations, and security. OpenAI said one HP engineer used its models across 122 pull requests in 43 projects, and that an HP security team used the models to remediate several software bugs in a day.
Filed from: OpenAI .
Ford Rehires Veteran Engineers After AI Quality Systems Fall Short
Ford executives said the company hired 350 veteran engineers, including former employees and supplier staff, after automated quality systems did not produce the product-quality level the company wanted. TechCrunch cited Bloomberg reporting that the specialists look for failure points before parts reach plants, and Ford hardware engineering executive Charles Poon said the company had expected artificial intelligence and ingested design requirements to produce higher-quality output.
Filed from: TechCrunch , Bloomberg .
Australia Investigates Five Platforms Over Under-16 Social Media Ban
Australia said eSafety is investigating possible non-compliance by Facebook, Instagram, Snapchat, TikTok, and YouTube under the country’s social media minimum-age law. The Prime Minister’s office said the government will legislate stronger information-gathering powers for eSafety and double the maximum penalty for systemic breaches to AU$99 million.
Filed from: The Register , Prime Minister of Australia .
BT and Verizon Form International Enterprise Networking Joint Venture
BT and Verizon said they will combine BT International with Verizon’s international enterprise wireline business in a 50:50 joint venture with roughly $4 billion in annual revenue. The Register reported that the proposed company would serve more than 3,000 multinational customers across 180 countries, with Verizon paying BT $625 million to balance the contributed assets and a 2027 close expected subject to regulatory approval.
Filed from: The Register .
Suno Spark Terms Include Remix, Exclusivity, and Non-Disparagement Conditions
Suno launched Spark, an incubator program for unsigned independent artists that offers grants, mentorship, marketing support, and access to Suno programs. Suno’s terms say participants must make content available for collaborations, remixes, or derivative works by third parties, give Suno permission to use content and name-and-likeness materials for marketing, avoid paid promotional work with other AI music companies for 60 days after the final content post, and follow a non-disparagement clause during and after the term.
Filed from: The Verge , Suno , Suno terms .
Proception Settles Tesla Trade-Secret Suit and Raises $11 Million
Proception, the robot-hand startup founded by former Tesla Optimus technical lead Jay Li, reached a settlement with Tesla and said it raised an $11 million seed round led by First Round Capital. The company said it is shipping its first batch of high-dexterity robotic hands to researchers and robotics companies, and court docket summaries show Tesla’s claims were dismissed with prejudice on June 2, 2026.
Filed from: TechCrunch , PacerMonitor .
Z.ai GLM-5.2 Draws Cybersecurity Benchmark Claims
The Verge reported that China’s Zhipu AI, also known as Z.ai, released the open-weight GLM-5.2 model and that researchers cited by The Wall Street Journal said it matched Anthropic’s Mythos in some cybersecurity bug-finding scenarios. The WSJ report said the model still trails Anthropic and OpenAI products in other tasks, with public benchmark detail limited in the accessible reporting.
Filed from: The Verge , The Wall Street Journal .
Prosecutors Used ChatGPT Logs in Palisades Fire Trial That Ended in Mistrial
The Verge reported that prosecutors used Jonathan Rinderknecht’s ChatGPT logs as part of their arson case over the 2025 Palisades fire, alongside phone location data, security footage, and witness testimony. AP reported that jurors were deadlocked on all three charges after two days of deliberations, while the Justice Department case page describes the indictment allegations against Rinderknecht.
Filed from: The Verge , AP , Justice Department .
Omen AI Raises $31 Million for Data Center Coolant Monitoring
Omen AI said it raised a $31 million Series A led by Nava Ventures to monitor liquid-cooling fluid in data centers with real-time spectrometer hardware. TechCrunch reported that Omen is working with a dozen data center customers, including TensorWave, and that the company has raised $40 million since its 2024 founding.
Filed from: TechCrunch , Omen AI .
From the Community
LibrePods Unlocks AirPods Features on Non-Apple Devices
LibrePods is an open-source Android app that reverse-engineers Apple’s proprietary AirPods protocol to enable features such as noise-control switching, ear detection, battery status, head gestures, and conversational awareness on non-Apple devices. The project implements AACP and ATT protocols from scratch, uses a Rust-based backend, carries a GPLv3 license, and supports up to two simultaneous device connections with switching.
Filed from: GitHub .
NanoEuler Builds a GPT-2-Scale LLM in C and CUDA
NanoEuler is a GPT-2-class language model of about 116 million parameters built from scratch in C and CUDA, without PyTorch, autograd, or machine-learning libraries. The project includes a hand-written byte-level BPE tokenizer, a FlashAttention kernel, a full pretraining and supervised-fine-tuning pipeline, and analytic-gradient checks against central finite differences.
Filed from: GitHub .
