AI Compute

Twitter

Facebook

Copy Link

Key Facts

Key PlayersNVIDIA, Microsoft, Google, AWS, CoreWeave

Top News

AI storage becomes the bottleneck as agentic AI pushes infrastructure beyond GPU limits

As AI evolves from simple chatbots to autonomous agentic systems, the industry faces a critical infrastructure challenge that has nothing to do with compute power. The real bottleneck is AI storage—specifically, managing massive context memory and KV cache that agentic AI generates. Storage technology is receiving what industry experts call a 'promotion' as enterprises discover that even the most powerful GPUs become idle when starved of data.

Technology5 sources

AI storage becomes the bottleneck as agentic AI pushes infrastructure beyond GPU limits

Technology5 sources

Penguin Solutions Posts Record Earnings, Raises Outlook as AI Infrastructure Demand Surges

Penguin Solutions delivered a record third quarter with revenue of $479 million, crushing analyst expectations by nearly $73 million. The AI infrastructure company posted earnings of 84 cents per share versus estimates of 54 cents, driven by memory solutions that more than doubled year-over-year. Management raised full-year revenue growth guidance to 22% from 12%, citing very strong agentic AI-driven customer demand.

Business4 sources

Penguin Solutions Posts Record Earnings, Raises Outlook as AI Infrastructure Demand Surges

Business4 sources

FuriosaAI brings low-power AI chip to Europe as sovereign compute demand grows

South Korean AI chip startup FuriosaAI has deployed its RNGD accelerators at Equinix's Lisbon datacenter, marking its first European expansion. The power-efficient chips offer 512 teraFLOPS while consuming just 180 watts, positioning the startup as a Nvidia alternative for enterprises seeking sovereign AI compute solutions. The company is also developing third-generation accelerators with Broadcom using HBM4 memory.

Technology2 sources

FuriosaAI brings low-power AI chip to Europe as sovereign compute demand grows

Technology2 sources

SpaceX files for Starlink Gen 3: 100,000 satellites to power AI infrastructure worldwide

SpaceX filed an application with the FCC for its Starlink Gen 3 constellation, proposing 100,000 satellites to deliver multi-gigabit broadband service and support billions of AI-powered devices. The system aims to provide ultra-low-latency connectivity for industrial automation, precision agriculture, telemedicine, and personal robotics, positioning itself as the communications backbone of the AI age.

Technology2 sources

SpaceX files for Starlink Gen 3: 100,000 satellites to power AI infrastructure worldwide

Technology2 sources

NVIDIA Rosa CPU with Rigel Core targets Agentic AI with higher per-core performance

NVIDIA has confirmed key details about its next-generation Rosa CPU featuring the new Rigel core. Built on an Arm v9.2-based design, Rosa delivers higher per-core performance than its Vera predecessor while maintaining the same silicon footprint. The chip includes a larger L2 cache, better instruction delivery, and more efficient memory handling, positioning NVIDIA to compete directly with AMD EPYC Venice and Intel Xeon Diamond Rapids in server CPU markets.

Technology2 sources

NVIDIA Rosa CPU with Rigel Core targets Agentic AI with higher per-core performance

Technology2 sources

Morgan Stanley backs Kingsoft Cloud as pure AI cloud play with 64% upside potential

Morgan Stanley initiated coverage of Kingsoft Cloud with an overweight rating and $15 price target, representing 64% upside. The investment bank views the Chinese company as a pure AI cloud play, forecasting AI will contribute over 60% of revenue by 2028 as it transitions from a mid-tier commodity cloud provider.

Business2 sources

Morgan Stanley backs Kingsoft Cloud as pure AI cloud play with 64% upside potential

Business2 sources

DeepSeek plans to make its own AI chips to sidestep US export controls and reduce Nvidia reliance

Chinese AI startup DeepSeek is developing its own AI chip for inference workloads, marking a strategic shift for the company behind globally popular models. The year-long effort aims to reduce dependence on Nvidia and Huawei amid US export controls that have reshaped China's semiconductor landscape. DeepSeek joins OpenAI and other AI leaders in seeking greater control over their hardware stack.

Technology11 sources

DeepSeek plans to make its own AI chips to sidestep US export controls and reduce Nvidia reliance

Technology11 sources

AMD Ryzen AI Halo brings 128GB unified memory for local AI development at $3,999

AMD launches the Ryzen AI Halo, a compact AI workstation designed for local AI development with 128GB of unified memory. Priced at $3,999, this powerful mini PC runs on the Ryzen AI Max+ 395 processor and directly challenges Nvidia's DGX Spark. The system supports models up to 200 billion parameters and ships with either Windows 11 or Linux with full ROCm software stack pre-installed.

Technology7 sources

AMD Ryzen AI Halo brings 128GB unified memory for local AI development at $3,999

Technology7 sources

ASML stock surges 123% as Bernstein raises price target to $2,623 on AI-driven semiconductor demand

Bernstein raises its ASML stock price target to $2,623, implying 48% upside, as AI demand drives unprecedented expansion in advanced logic and DRAM capacity. The firm forecasts EUV revenue will grow at 30% annually, reaching €42.7 billion by 2030, more than 30% above street estimates, positioning the semiconductor stock for continued gains.

Business2 sources

ASML stock surges 123% as Bernstein raises price target to $2,623 on AI-driven semiconductor demand

Business2 sources

Nvidia Kyber AI rack delayed to 2028 as circuit board manufacturing hits roadblock

Nvidia's Kyber rack-scale architecture, designed to house 144 Rubin Ultra chips, has been pushed back more than 12 months to 2028 due to difficulties manufacturing a critical PCB midplane. The delay leaves Nvidia without a proven scaling solution for its most powerful systems, potentially opening the door for AMD and Google to gain ground in the high-end AI chip market.

Technology2 sources

Nvidia Kyber AI rack delayed to 2028 as circuit board manufacturing hits roadblock

Technology2 sources

AI Agents Consume 136.5 Times More Energy Than Chatbots, KAIST Study Reveals Grid Strain Ahead

New research from KAIST shows AI agents can consume up to 136.5 times more energy per query than standard generative AI chatbots. A single complex request burns through 348.41 watt-hours of electricity while GPUs sit idle for over half the time. If scaled to Google search traffic levels, these systems could demand nearly half of the entire U.S. electricity consumption.

Science4 sources

AI Agents Consume 136.5 Times More Energy Than Chatbots, KAIST Study Reveals Grid Strain Ahead

Science4 sources

AI-Driven RAM Crisis Threatens Budget Smartphones as Memory Costs Double and Shipments Plummet

Budget smartphone buyers face a challenging market as AI infrastructure demands drive memory costs to unprecedented levels. Omdia predicts a 22% decline in sub-$400 phone shipments through 2027, with memory now consuming nearly 60% of manufacturing costs. Chinese manufacturers are considering abandoning the budget segment entirely as razor-thin margins evaporate.

Business6 sources

AI-Driven RAM Crisis Threatens Budget Smartphones as Memory Costs Double and Shipments Plummet

Business6 sources

Meta repurposes DDR4 memory in DDR5 servers with custom chip to cut AI infrastructure costs

Meta revealed it's reusing DDR4 memory from retired servers in new DDR5-only systems using a custom CXL ASIC called Vistara. The approach cuts AI inference server count by up to 25% and reduces job-restart overhead by 33%, helping the company navigate memory supply shortages while lowering infrastructure costs.

Technology2 sources

Meta repurposes DDR4 memory in DDR5 servers with custom chip to cut AI infrastructure costs

Technology2 sources

Vertiv Opens Malaysia Manufacturing Facility to Meet Asia's Surging AI Infrastructure Demand

Vertiv has opened a new manufacturing facility in Johor, Malaysia, to address the growing demand for AI and high-density computing infrastructure across Asia. The facility will produce advanced thermal management and power solutions, including liquid cooling units and prefabricated power modules, with full operationalization expected by 2027 bringing hundreds of skilled jobs to the region.

Business2 sources

Vertiv Opens Malaysia Manufacturing Facility to Meet Asia's Surging AI Infrastructure Demand

Business2 sources

GMKtec EVO-X3 debuts at $3,600 with AMD Ryzen AI Max+ 395 and vertical tower design

GMKtec has launched the EVO-X3, a high-end AI mini PC workstation priced at $3,600, featuring AMD's Ryzen AI Max+ 395 Strix Halo processor and 128 GB of memory. The company abandoned traditional flat designs for a vertical tower chassis with triple-fan cooling. Despite using the same chip as its predecessor, the EVO-X3 represents a significant price increase—nearly 300% from GMKtec's original EVO-X1 model.

Technology2 sources

GMKtec EVO-X3 debuts at $3,600 with AMD Ryzen AI Max+ 395 and vertical tower design

Technology2 sources

Prev 1…678…78 Next