AI Compute

Twitter

Facebook

Copy Link

Key Facts

Key PlayersNVIDIA, Microsoft, Google, AWS, CoreWeave

Top News

Groq's Next-Gen LPU Adopts 3D Stacked DRAM to Support Expanding AI Models

Groq is shifting its next-generation Language Processing Unit architecture from SRAM to 3D stacked DRAM technology. This strategic move prioritizes memory capacity over raw bandwidth speed, positioning the hardware to handle increasingly large AI models as the industry scales up model sizes.

Technology2 sources

Groq's Next-Gen LPU Adopts 3D Stacked DRAM to Support Expanding AI Models

Technology2 sources

Vultr and AMD Launch 20MW GPU Cluster in Chennai—India's Largest AMD Instinct Deployment

Vultr announced a 20MW AMD GPU cluster in Chennai featuring AMD Instinct MI355X GPUs with 288GB of HBM3E memory. This marks India's largest AMD Instinct deployment and positions Chennai alongside Springfield, Ohio as one of the world's largest AMD AI infrastructure sites, addressing growing demand for sovereign cloud infrastructure.

Technology3 sources

Vultr and AMD Launch 20MW GPU Cluster in Chennai—India's Largest AMD Instinct Deployment

Technology3 sources

Elon Musk Announces ASML Will Expedite Shipments to Terafab, Targets 2028 Volume Chip Production

Elon Musk revealed that ASML and other semiconductor equipment suppliers are expediting deliveries to the ambitious Terafab facility in Texas. The joint SpaceX and Tesla venture targets volume chip production by Q3-Q4 2028, aiming to produce 1 terawatt of AI computing power annually—though Musk cautioned the timeline carries "big error bars."

Technology4 sources

Elon Musk Announces ASML Will Expedite Shipments to Terafab, Targets 2028 Volume Chip Production

Technology4 sources

Meta Keeps 11% of Nvidia GPUs Idle in Prometheus Supercluster to Prevent AI Training Failures

Meta revealed that over 11% of Nvidia GPUs in its Prometheus supercluster sit idle or dead as hot spares. The New Albany facility cut interruptions by 99% through redundancy, data parallelism, and checkpointing. The infrastructure improvements enable Meta to train models like Muse Wonder across nearly 1 million GPUs.

Technology3 sources

Meta Keeps 11% of Nvidia GPUs Idle in Prometheus Supercluster to Prevent AI Training Failures

Technology3 sources

SpaceX Revenue Jumps 161% to $14.6B as AI Business Overtakes Starlink in Q3 Earnings

SpaceX reported third-quarter revenue of $14.6 billion, up 161% year-over-year, beating Wall Street estimates as its AI business overtook Starlink for the first time. The company's Super Intelligence segment generated $7.6 billion, primarily from renting data-center computing capacity to firms like Anthropic and Google, while capital expenditures reached $19.6 billion.

Business9 sources

SpaceX Revenue Jumps 161% to $14.6B as AI Business Overtakes Starlink in Q3 Earnings

Business9 sources

Perplexity Computer Now Runs Entirely on Your Mac with Full Local AI Processing

Perplexity has launched a fully local version of its Computer app for Mac, allowing AI workflows to run entirely on-device without cloud dependency. Available to Pro, Max and Enterprise subscribers, the update requires Apple Silicon Macs with at least 36GB of RAM and supports Qwen 3.6 and PPLX Pro models for private, offline AI computing.

Technology7 sources

Perplexity Computer Now Runs Entirely on Your Mac with Full Local AI Processing

Technology7 sources

Cephalon's Lucy AI Studio merges local AI processing with private cloud storage in compact device

Cephalon unveiled the Lucy AI Studio, a compact mini PC and NAS combo that brings local AI processing and private cloud storage into one device. Powered by AMD Ryzen AI processors with up to 192GB unified memory, it runs large AI models locally while addressing privacy concerns about cloud-based AI systems.

Technology2 sources

Cephalon's Lucy AI Studio merges local AI processing with private cloud storage in compact device

Technology2 sources

Inception Mercury 3 Diffusion Model Claims 8X Speed Advantage Over GPT-6 Luna and Claude Haiku 5.5

Redwood City AI lab Inception released Mercury 3, a diffusion-based large language model that generates more than 1000 tokens per second on Nvidia chips. The model outperforms GPT-6 Luna and Claude Haiku 5.5 across benchmarks while running four to eight times faster, with a one million token context window and launch pricing of $0.10 per million input tokens.

Technology2 sources

Inception Mercury 3 Diffusion Model Claims 8X Speed Advantage Over GPT-6 Luna and Claude Haiku 5.5

Technology2 sources

HPE Spaceborne Computer-4 to Run First AI on the Moon Aboard Astrolab FLIP Rover in Late 2026

HPE's Spaceborne Computer-4 will launch to the lunar south pole aboard Astrolab's FLIP rover in late 2026, marking the first deployment of AI and high-performance computing on the Moon. The fanless edge computing module will test large language models, run HPC benchmarks, and process LiDAR data for autonomous navigation in extreme lunar conditions.

Technology6 sources

HPE Spaceborne Computer-4 to Run First AI on the Moon Aboard Astrolab FLIP Rover in Late 2026

Technology6 sources

Vertiv Launches Guide to Help Enterprises Scale AI Infrastructure from Pilot to Production

Vertiv and Deloitte unveiled a comprehensive framework to address a critical enterprise challenge: moving AI workloads from experimentation to sustained production. The Designing Infrastructure for Enterprise AI Guide provides a workload-based approach to determine deployment locations and assess whether existing physical infrastructure can support AI at scale.

Technology2 sources

Vertiv Launches Guide to Help Enterprises Scale AI Infrastructure from Pilot to Production

Technology2 sources

MiniMax M3.1 Flash Slashes AI Costs by 70% but Independent Tests Reveal Accuracy Trade-Offs

Chinese AI startup MiniMax released M3.1 Flash, delivering 70% cost reduction and 3X faster inference speeds for enterprise workflows and software development. Independent benchmarks show competitive performance against DeepSeek at lower costs, but Artificial Analysis found the model generates incorrect answers rather than acknowledging uncertainty.

Technology4 sources

MiniMax M3.1 Flash Slashes AI Costs by 70% but Independent Tests Reveal Accuracy Trade-Offs

Technology4 sources

doubleAI's Maxwell code execution model slashes compute costs by 67% as Nvidia and Dell invest $220M

Tel Aviv startup doubleAI, led by Mobileye co-founder Amnon Shashua, has launched Maxwell—the first code execution model that rewrites machine code during runtime to optimize software efficiency. The AI model that cuts compute costs helped one early customer reduce spending by 67% while doubling inference speed for LTX's multimodal models.

Technology5 sources

doubleAI's Maxwell code execution model slashes compute costs by 67% as Nvidia and Dell invest $220M

Technology5 sources

Tenstorrent Raises $1.5 Billion at $6.2 Billion Valuation to Scale Open-Weight AI Chips

AI chip startup Tenstorrent raised $1.5 billion at a $6.2 billion valuation, more than doubling its worth from 2024. Led by Jim Keller, the company chose independence over takeover offers from Intel and Qualcomm to scale its Galaxy servers for open-weight AI models and video generation workloads.

Startups7 sources

Tenstorrent Raises $1.5 Billion at $6.2 Billion Valuation to Scale Open-Weight AI Chips

Startups7 sources

Amazon and Google Launch Tools to Challenge Nvidia's AI Chip Dominance with PyTorch Support

Amazon and Google released TorchNeuron and TorchTPU, open-source tools that let AI developers run PyTorch code on their custom AI chips with minimal changes. The move directly challenges Nvidia's market dominance by making it easier to switch from Nvidia's GPUs to alternative hardware without rewriting applications.

Technology13 sources

Amazon and Google Launch Tools to Challenge Nvidia's AI Chip Dominance with PyTorch Support

Technology13 sources

Modal Deploys Nvidia Vera CPU After Tests Show 2x Throughput Gains Over Intel and AMD Chips

AI cloud provider Modal will deploy Nvidia Vera CPU alongside its existing Intel and AMD processors after internal tests revealed doubled throughput for AI agent workloads. The move marks Modal's first adoption of Arm architecture as the $15.75 billion company builds a heterogeneous server fleet optimized for agentic AI workflows.

Technology8 sources

Modal Deploys Nvidia Vera CPU After Tests Show 2x Throughput Gains Over Intel and AMD Chips

Technology8 sources

Prev 123 4…100 Next