LIVE · TUE, AUG 04, 2026 --:--:-- ET

Issue Nº 105 COST TOTAL $15060.18 ARTICLES TODAY 5 TOKENS TOTAL 9.81B

Running the wire

Policy NIST joins Genesis Mission with AI centers for manufacturing drones, critical infrastructure cybersecurity Policy NSF launches TechAccess program: $224M for 56 state AI coordination hubs, workforce readiness Policy FCC moves toward ban on Chinese optical transceivers for U.S. data centers; Innolight holds 27% global share Chips Kioxia and SanDisk hit record 3D NAND density: 332 layers, 37 Gb/mm², 4,800 MT/s I/O Research Alibaba Qwen 3.8-Max (2.4T) and 27B Open-Weights Models Released; Aims at Long-Horizon Coding and Agentic Work Breaking Databricks Joins Open Secure AI Alliance as Founding Member; Contributes Omnigent Meta-Harness, DASF 3.0 Security Framework Chips NVIDIA Open-Sources cuFile APIs for GPU-to-Storage Direct Access; Launches Storage-Next Initiative Chips NVIDIA Alpamayo 2 Super open reasoning model now available for commercial AV deployment Funding EU's €5B Scaleup Europe Fund now operational; targets AI, quantum, clean energy Market Texas halts 1,800 data center approvals; 474 GW power queue stalls pending audits Breaking Tata Electronics hack exposes Apple supply secrets; India pivot risks amid China competition Funding EU's €5bn Scaleup Fund opens investment window; Mistral, Exploration Company rumored targets Chips SanDisk & SK hynix unveil HBF spec: 512GB GPU memory at 3 TB/s bandwidth via UCIe Market Texas halts 1,800 data center projects pending power audit; 474 GW pending requests 5x peak Funding HappyRobot hits $1.2B unicorn status with $150M Series C from Prysm & Eurazeo Funding July broke venture records: 14 billion-dollar rounds, $65B capital, 53% to AI Policy NIST joins Genesis Mission with AI centers for manufacturing drones, critical infrastructure cybersecurity Policy NSF launches TechAccess program: $224M for 56 state AI coordination hubs, workforce readiness Policy FCC moves toward ban on Chinese optical transceivers for U.S. data centers; Innolight holds 27% global share Chips Kioxia and SanDisk hit record 3D NAND density: 332 layers, 37 Gb/mm², 4,800 MT/s I/O Research Alibaba Qwen 3.8-Max (2.4T) and 27B Open-Weights Models Released; Aims at Long-Horizon Coding and Agentic Work Breaking Databricks Joins Open Secure AI Alliance as Founding Member; Contributes Omnigent Meta-Harness, DASF 3.0 Security Framework Chips NVIDIA Open-Sources cuFile APIs for GPU-to-Storage Direct Access; Launches Storage-Next Initiative Chips NVIDIA Alpamayo 2 Super open reasoning model now available for commercial AV deployment Funding EU's €5B Scaleup Europe Fund now operational; targets AI, quantum, clean energy Market Texas halts 1,800 data center approvals; 474 GW power queue stalls pending audits Breaking Tata Electronics hack exposes Apple supply secrets; India pivot risks amid China competition Funding EU's €5bn Scaleup Fund opens investment window; Mistral, Exploration Company rumored targets Chips SanDisk & SK hynix unveil HBF spec: 512GB GPU memory at 3 TB/s bandwidth via UCIe Market Texas halts 1,800 data center projects pending power audit; 474 GW pending requests 5x peak Funding HappyRobot hits $1.2B unicorn status with $150M Series C from Prysm & Eurazeo Funding July broke venture records: 14 billion-dollar rounds, $65B capital, 53% to AI

Chips Tuesday, August 4, 2026 at 04:03 PM

NVIDIA Open-Sources cuFile APIs for GPU-to-Storage Direct Access; Launches Storage-Next Initiative

At the Future of Memory and Storage conference, NVIDIA announced it is open sourcing cuFile, the application programming interfaces powering GPU-to-storage direct memory access (DMA) transfers that bypass the CPU. cuFile is the core component of NVIDIA GPUDirect Storage, enabling GPUs to read from and write to storage directly using hundreds of thousands of GPU threads and high-bandwidth memory, with data access times in the microsecond range. The open-source release includes the vertical storage software stack, positioning cuFile for interoperability across storage vendors and GPU platforms.

The release addresses a critical bottleneck: as AI agents consume massive amounts of data, GPUs now initiate thousands of concurrent storage requests. Storage systems must encrypt, compress, verify and reconstruct data for each request, and these data services can become bottlenecks at scale. NVIDIA Vera BlueField-4 STX storage processors deliver up to 3.21x higher throughput than x86 CPUs in dual-stage compression-encryption pipelines, absorbing the AI data flood with significantly less compute infrastructure. cuFile enables this direct path without CPU bouncing, freeing CPU resources and reducing latency.

Alongside cuFile, NVIDIA launched Storage-Next, a vendor consortium uniting over 40 storage and flash suppliers (DDN, KIOXIA, Micron, and others) with GPU vendors and standards bodies to align on how GPU-driven storage should behave. The initiative includes SCADA (Scaled, Accelerated Data Access), a framework allowing massively parallel GPUs to pull only necessary data directly from storage into their own high-speed memory. NVIDIA also launched the Open Secure AI Alliance with Google, Intel, and Meta as inaugural maintainers, supporting open APIs for secure, distributed AI storage access.

For infrastructure teams building AI data pipelines, the open-source cuFile and Storage-Next standardization reduce vendor lock-in and accelerate adoption of GPU-direct storage across heterogeneous environments. The shift from passive storage to active data-path participation fundamentally changes memory-versus-storage economics in the microsecond era.

Sources