Data Center Dynamics· 1d agoInfineon is supplying silicon carbide devices for Eaton's solid-state transformers to support 800VDC power delivery in high-density AI data centers.
Read articleNews
HPC News Outlets
The daily sweep across the trade press — HPCwire, The Next Platform, insideHPC, Data Center Dynamics, ServeTheHome — rewritten in plain language with a straight link to the original. Lab and community announcements live on Community News.
Since
Sep 29, 202612
- Infineon will supply Eaton with SiC chips for 800VDC data center transformers
- Global Switch will host a large Nvidia Blackwell cluster in Paris
Global Switch claims the liquid-cooled installation in Paris will be one of the largest Nvidia Blackwell deployments in Europe.
Read article - SC26 gave its Test of Time Award to Nathan Bell and Michael Garland for GPU SpMV
HPCwire· 1d ago
The award honors their 2009 paper that showed how to map irregular sparse matrix-vector multiplication onto GPU architectures.
Read article - SEMIFIVE landed a $52M contract to build an AI inference ASIC
SEMIFIVE will design and produce an AI inference chip using LPDDR6 and PCIe Gen5 for an unnamed US company, targeting tape-out in 2027.
Read article - Dongarra says scientific computing needs trusted software ecosystems
Jack Dongarra claims computing advances depend on sustained software ecosystems to deliver reliable results rather than hardware performance alone.
Read article - Xanadu and Bluefors partner on modular cooling for quantum data centers
They say the modular cryogenic design will avoid massive industrial cryoplants for photon detectors in future quantum systems.
Read article - Samsung invested $1 billion in Helix Digital Infrastructure for AI data centers
The funding expands Helix's $10 billion pool to build out data centers, power, and cooling infrastructure for hyperscalers.
Read article - LiquidStack launched the CDU 2.X coolant distribution unit
They claim the unit pushes up to 3,750 liters per minute at 3.5 bar and handles facility water temperatures up to 45°C.
Read article - Gimlet Labs added Cerebras chips to its cloud inference platform
They claim pairing Cerebras wafer-scale chips with GPUs delivers up to 3,000 tokens per second for real-time AI inference.
Read article - AMD is buying Fei-Fei Li's World Labs for $8.2 billion
The deal brings founder Fei-Fei Li to AMD as chief scientist to help optimize spatial foundation models on its GPU hardware.
Read article - TACC partnered with Occam Foundry on Austin quantum campus
TACC signed an agreement with Occam Foundry to pursue funding and workforce development for a planned 16.5-acre quantum commercialization site in Texas.
Read article - New Mexico opened a 35,000-square-foot quantum lab with IonQ and QuEra
The 35,000-square-foot Albuquerque facility acts as an anchor for the state's $450 million quantum tech initiative.
Read article
Sep 28, 202610
- KIST connected Korean quantum startups with D-Wave and Microsoft
KIST arranged talks for Korean quantum startups to access D-Wave's systems and build partnerships with Microsoft and Japan's AIST.
Read article - Lenovo benchmarked OpenMoonRay ray tracing on Intel Xeon 6
Lenovo claims Intel's 128-core Xeon 6980P reached up to 2.11 times the rendering throughput of a 5th Gen Xeon in OpenMoonRay tests.
Read article - Jülich researchers won a EuroHPC award for multi-GPU P3M solver
They built a portable electrostatic solver library and benchmarked P3M scaling across multiple GPUs on MareNostrum5 and JUWELS Booster.
Read article - AMD will acquire World Labs for $8.2 billion
AMD agreed to buy spatial-AI startup World Labs for $8.2 billion in stock, adding Fei-Fei Li as executive VP and chief scientist.
Read article - AI data centers hit power limits driven by data movement bottlenecks
The IEA projects data center power draw will double to 945 TWh by 2030 as data movement overtakes raw compute as the key performance bottleneck.
Read article - NSF launched the Open Knowledge Network linking 43 data graphs
The public network links 43 federated knowledge graphs across federal agencies to give AI models grounded facts for scientific and enterprise use.
Read article - Europe's Chips JU opened €80M in funding calls for AI chips and systems
The funding splits into €10M for taped-out chip demonstrators and €70M to build an evaluation platform and procure multi-rack AI clusters.
Read article - DOE advisory committee outlines scientific AI roadmap for national labs
The DOE's Genesis Mission plan links national lab supercomputers and AI to tackle fusion control, biology, and materials supply chains.
Read article - Pasqal left France's defense quantum program to focus on civil R&D
The company exited the French defense PROQCIMA program to submit a new neutral-atom fault-tolerant roadmap to civil authorities.
Read article - Lenovo made few changes six months after buying Infinidat
Blocks & Files· 2d ago
Six months after a reported $500M buyout, Lenovo has yet to add GPUDirect or major AI storage features to Infinidat's platform.
Read article
Sep 27, 20261
- Nscale raised $2.36B as its UK AI data center faces grid delays
Supercomputing News· 3d ago
Nscale closed $2.36 billion in convertible debt, but reports suggest its planned UK AI cluster might lack grid power until the 2030s.
Read article
Sep 26, 20262
- SemiAnalysis tore down Intel Panther Lake to evaluate 18A
SemiAnalysis· 3d ago
SemiAnalysis found Intel's 18A compute logic matches TSMC N3E density but falls behind TSMC N3P, N2, and Samsung SF2.
Read article - NTUA deployed ICARUS, a 48-GH200 campus supercomputer
The 24-node HPE cluster delivers up to 3.2 theoretical FP64 Tensor Core petaflops across 48 Nvidia GH200 superchips over 100GbE networking.
Read article
Sep 25, 202615
- TekSynap won an NRC high-performance computing contract
· 4d ago
TekSynap was awarded a task order to support the Nuclear Regulatory Commission's High Performance Computing System under the GLINDA 2.0 vehicle.
Read article - SemiAnalysis mapped over 24GW of datacenter capacity in China
They found China has over 24GW of datacenter capacity, with ByteDance occupying roughly a fifth of the total.
Read article - ALCF highlights student research on supercomputers and workflows
Intern projects included multi-facility workflow observability tools for the American Science Cloud and new scheduler policies on Polaris.
Read article - Huawei revealed new interconnects and storage for massive AI clusters
They claim their new Unified Bus and M900 storage can scale up to one million NPUs to train 10-trillion-parameter AI models.
Read article - Microsoft opened a Maryland quantum center for DARPA benchmarking
The 15,000-square-foot facility will host Microsoft's topological quantum hardware for testing under DARPA's benchmarking program.
Read article - EuroHPC named its 2026 best users and papers in Dublin
Stephen Beller and Thomas Saillour won top user awards for scaling seismic inversion and ocean modeling codes across MareNostrum5 and MeluXina.
Read article - Naval Postgraduate School speeds up hurricane models 5x on GPUs
Researchers ported cyclone simulation code to GPUs on NCSA's Delta supercomputer, reporting a 5x speedup and lower energy use compared to CPU runs.
Read article - NetApp to acquire PEAK:AIO for parallel storage in AI clusters
NetApp plans to integrate PEAK:AIO's parallel file system and metadata architecture into ONTAP to scale storage for GPU clusters.
Read article - Germany picked planqc for its €640M quantum competition
The LOGIQC consortium will develop an error-corrected neutral-atom quantum computer under Germany's €640 million initiative.
Read article - DOE picked winners in its quantum grid planning challenge
Teams benchmarked quantum and hybrid algorithms on real quantum hardware against classical baselines to optimize microgrid and storage placement.
Read article - Nvidia launched a qualification program for AI data center power and cooling
Nvidia's DSX Ready program certifies third-party battery systems and coolant distribution units against its AI data center reference designs.
Read article - IonQ built a real-time quantum error decoder that runs on a single CPU
The Next Platform· 5d ago
They claim their new decoder runs real-time error correction for hundreds of logical qubits in the background using just a single standard CPU.
Read article - NetApp buys Peak:AIO to add parallel NFS to ONTAP
NetApp says the deal adds parallel NFS and independent metadata scaling to ONTAP to prevent GPU stalls in AI clusters.
Read article - Microsoft delivered a Majorana 2 quantum system to DARPA
Microsoft delivered an on-site Majorana 2 quantum system to DARPA in Maryland, but they have not disclosed the qubit count or test protocols.
Read article - Intel and Lenovo benchmarked Xeon 6 chips on scientific workloads
A whitepaper claims Granite Rapids SP CPUs average 1.22x speedups over 5th Gen Xeon, with AP models reaching up to 2.05x on HPC tasks.
Read article
Sep 24, 202616
- Nvidia claims AI data centers waste a third of their grid power
Techzine· 6d ago
Nvidia claims AI data centers lose 30 to 40 percent of grid power to cooling, bad rack configurations, and system failures.
Read article - LRZ hosted a TPC Europe hackathon on AI for supercomputing
Thirty researchers gathered at LRZ to build agentic AI workflows and tools for coordinating jobs on scientific supercomputers.
Read article - BSC co-develops Spain's National Cancer Data Node
The platform will aggregate clinical, genomic, and imaging data across Spanish research centers using FAIR standards for oncology research.
Read article - DoE's Darío Gil rejected AI doomerism and promoted scientific AI
DoE's Darío Gil dismissed catastrophic AI fears, pointing to a $5 billion program that combines AI, quantum, and HPC for science.
Read article - PicoJool raised $27.5M to build optical interconnects for AI clusters
They claim their 200G VCSEL modules offer low-power optical links that can scale data centers to 1.6T and beyond.
Read article - EdgeCortix unveiled its RAIDEN modular AI chiplet platform
They claim the four-die X4 flagship delivers up to 3.36 PFLOPS of FP4 compute with 256 GB memory for edge AI systems.
Read article - Vertiv acquired King Environmental Services for liquid cooling work in Europe
Vertiv bought European firm KES to add fluid management and load-testing services for liquid-cooled AI and HPC data centers.
Read article - EuroHPC launched a supercomputing network with Latin American institutions
EuroHPC JU kicked off a two-year initiative linking European HPC centers with seven Latin American countries to share supercomputing and AI resources.
Read article - Bifrost Electronics named Doug Campbell as CEO
Bifrost named Doug Campbell CEO as it prepares to launch an unshielded quantum readout amplifier built without Josephson junctions.
Read article - Survey finds 90% of quantum companies rely on foreign suppliers
A survey of 160 quantum firms across 15 countries found 90% use foreign suppliers and 74% rely on international customers.
Read article - Framework outlines which HPC workloads belong in the cloud
It uses coupling topology and cluster utilization data to determine whether tightly or loosely coupled jobs fit cloud infrastructure.
Read article - memQ open-sources a compiler for distributed quantum computers
The open-source framework splits and schedules standard quantum circuits across networked QPUs with different qubit architectures.
Read article - Florida International University bought an IonQ 256-qubit quantum system
The 256 physical trapped-ion qubit system is scheduled for on-campus installation in late 2027 following completion of FIU's data center.
Read article - QC Design claims its AI cuts quantum logical error rates tenfold
QC Design claims its Meridian AI reduced logical error rates by over 10x across 100 quantum error-correction design tasks compared to published methods.
Read article - Cornell demonstrated standing-wave cooling for trapped-ion chips
Researchers used Nullspace ES electrostatic simulation to demonstrate standing-wave cooling on a chip-scale ion trap for quantum computing.
Read article - Infleqtion ran 30 entangled logical qubits on its Sqale quantum system
They claim to have entangled 30 logical qubits across a single circuit using 80 physical neutral-atom qubits on their Sqale platform.
Read article
Sep 23, 202614
- TACC will allocate 90 percent of its Horizon supercomputer to NSF users
TACC says 90% of Horizon's 4,000 Blackwell GPUs will go to NSF users, yielding around 27 million GPU-hours per year.
Read article - Q.ANT launched an open-source SDK for photonic processors
The SDK lets developers program optical chips via Python and C/C++, targeting non-linear algorithms with up to 30x power savings, they claim.
Read article - WhiteFiber launched a distributed GPU cluster architecture
They claim the setup links data centers 83 km apart with 136 Tbps of bandwidth at 0.9 ms round-trip latency.
Read article - Thinking Machines signs $65M deal to run inference on Crusoe's B200 clusters
The lab will run open model inference on dedicated Nvidia HGX B200 systems linked by Quantum-2 InfiniBand under a $65 million annual deal.
Read article - Supermicro started shipping liquid-cooled Nvidia Vera Rubin NVL72 racks
Each rack packs 72 Rubin GPUs and 36 Vera CPUs with 20.7 TB of HBM4, connected via NVLink switch trays delivering 216 TB/s of bandwidth.
Read article - Classiq and INGL tested pipeline optimization on IonQ quantum hardware
They ran hybrid quantum-classical pipeline simulations on IonQ Forte-1 and found valid operating points close to classical reference solutions.
Read article - Qunnect detailed quantum network security projects with US agencies
They say their entanglement hardware can verify device identity, confirm signal location, and detect physical fiber taps.
Read article - Alibaba laid out its roadmap for AI chips, models, and data centers
They plan to surpass 20GW of data center capacity by 2032 and say upcoming Qwen models will scale up to 10 trillion parameters.
Read article - Photonic and Microsoft collaborate on quantum resource estimation
Photonic and Microsoft are working together to model qubit counts and runtime overhead for distributed quantum systems in Microsoft's resource estimator.
Read article - IonQ will link its Superion quantum system to Nvidia GPUs
IonQ will connect its Superion 256 quantum system directly to an Nvidia GB200 NVL72 cluster via NVQLink in 2027.
Read article - AtlasBase built a DNA storage and compute chip
AtlasBase says its Thalia platform features 5.6 billion synthesis sites and can write up to 250 GB of synthetic DNA data per chipset every 24 hours.
Read article - VAST previewed DataEnclave to run closed AI models on local GPUs
They claim it lets proprietary AI models run in secure enclaves on local hardware without exposing model weights or customer data.
Read article - DOE opened a $215M contest for 100-logical-qubit systems
DOE says it will award up to $215 million to private vendors that demonstrate at least 100 lab-verified logical qubits by 2028.
Read article - Researchers mapped surface code error correction to IBM heavy-hex QPUs
Researchers say they ran 2D surface error-correction codes on IBM’s heavy-hex layout without matching the physical grid.
Read article
Sep 22, 202615
- Nvidia launched a qualification program for data center power and cooling
ServeTheHome· Sep 22, 2026
The DSX Ready program qualifies power and cooling gear for high-density AI and HPC data centers.
Read article - SiPearl delivered the first Rhea1 CPU samples to Bull for JUPITER
The 80-core Arm Neoverse V1 processor features four HBM stacks and four DDR5 interfaces, bound for JUPITER's CPU partition.
Read article - QuEra and HPE teamed up to put quantum systems into Cray supercomputers
They plan to integrate QuEra's neutral-atom quantum hardware directly into on-premises HPE Cray systems for hybrid computing.
Read article - LANL and PEAK:AIO will present pNFS Lattice metadata architecture at SDC
The project decouples metadata and control services in parallel NFS so they can scale independently for large AI and HPC clusters.
Read article - Huawei upgraded its Stellar AI Fabric and launched new AI cluster switches
They launched new 100T and 51.2T switches and claim the fabric cuts buildout costs by 30% while lowering interconnect latency and power draw.
Read article - Standard databases absorb vector search while memory remains the main cost
Ten million 1,536-dimension float32 vectors require about 60 GB of RAM before indexing overhead, making memory the dominant operational cost.
Read article - HPE partnered with Nvidia on sovereign AI infrastructure
They claim their combined hardware and services keep sensitive models and data under strict local control for regulated sectors.
Read article - Internet2 and Cisco added Nvidia H200 PODs to the National Research Platform
The two new university sites add eight Nvidia H200 GPUs and 90 TB of NVMe storage to NRP's shared Nautilus cluster.
Read article - Coherent launched PhotonLink optics platform for AI hardware
The platform supports co-packaged and near-packaged optics for AI chips, with company revenue expected to ramp in late 2026.
Read article - DOE launched a center to bring national lab computing to heavy industry
Funded by the DOE, the new center connects heavy manufacturing with simulation, modeling, and AI capabilities at LLNL and Argonne.
Read article - George Mason will establish Virginia Quantum Hub with $3.7M grant
The hub will deploy an open-architecture quantum computing system built with TreQ to support regional research and workforce training.
Read article - Zapata opened early access to its Quantum Pilot software platform
The cloud platform uses AI tools to help enterprises map specific use cases to quantum hardware and estimate required computational resources.
Read article - SiPearl preps Rhea CPUs for Europe's upcoming exascale systems
Europe's JUPITER supercomputer will use Rhea1 chips by late 2026, while the upcoming Alice Recoque system is slated to deploy Rhea2 CPUs.
Read article - VAST Data launched DataEnclave to isolate AI models and weights
They say it uses CPU enclaves and Nvidia Confidential Computing to keep model weights and data encrypted inside GPU memory.
Read article - Nutanix bought Ryax to add GPU scheduling for Kubernetes and Slurm
They plan to integrate Ryax's bin-packing and scheduling tools, which Ryax claims can cut compute costs by up to 45% across Slurm and Kubernetes.
Read article
Sep 21, 202615
- EPB deployed an IonQ Forte quantum computer in Chattanooga
The utility paired IonQ's hardware with its quantum network to run power-grid optimization and university research workloads.
Read article - iPronics and BSC partner on optical networking for GPU clusters
The two-year project will test iPronics' programmable optical circuit switches inside BSC's GPU infrastructure to dynamically reconfigure AI networking.
Read article - ORNL built a GPU-ready solver to optimize data center power dispatch
ORNL says its relax-and-round method cut generator scheduling runtimes from 10 hours to under a second on a 276-generator system.
Read article - Crusoe raised $3.9B in Series F funding for AI data centers
The funding values the AI infrastructure provider at $30.9 billion as it expands to over 6 GW of contracted data center power capacity.
Read article - Gimlet Labs raised $300M to build multi-chip AI data centers
They route inference prefill to GPUs and decode to SRAM chips, claiming custom hardware pairing drastically improves data center efficiency.
Read article - IonQ will supply a Superion 256 quantum computer to SDT in South Korea
IonQ is selling its Superion 256 system and silicon-vacancy quantum memory module to SDT for a hybrid data center deployment in South Korea.
Read article - DKRZ joined the CLAIMA project for AI climate modeling on HPC
The four-year project will build data pipelines and deploy AI climate emulators on European HPC and AI Factory systems.
Read article - FCT unveiled flexible leak sensors for liquid-cooled AI and HPC servers
The custom flex assemblies wrap around cold plates, manifolds, and disconnects to detect coolant leaks directly inside server chassis.
Read article - Huawei introduced UnifiedBus interconnect architecture for large NPU clusters
They claim the architecture boosts interconnect bandwidth to TB/s levels and cuts round-trip latency to 2 microseconds.
Read article - EuroHPC launched its Access Resource Committee Bureau
The 22-member bureau will help 91 committee members review proposals and allocate time across European supercomputers, AI systems, and quantum hardware.
Read article - Dell and Diraq teamed up on hybrid quantum-HPC orchestration
Dell deployed an HPC cluster in Diraq's Sydney lab to test low-latency orchestration and real-time control for silicon spin qubits.
Read article - Georgia Tech and PNNL partnered on AI and research projects
The institutions signed an agreement to collaborate on AI-driven biomanufacturing and autonomous nuclear reactor operations.
Read article - Accelerated systems drove nearly 70 percent of server revenue in Q2
IDC data shows buyers spent $87.4 billion on GPU servers and $27.5 billion on other AI accelerator systems in the second quarter.
Read article - DapuStor detailed an SSD combining pSLC and QLC flash
DapuStor claims its dual-mode drive delivers over 7x the random-write IOPS of QLC-only SSDs with write latency below 8 microseconds.
Read article - InferenceX benchmarked Kimi K3 across Nvidia and AMD GPUs
SemiAnalysis· Sep 21, 2026
The benchmarks compare token throughput and TCO for the 2.8T model across H200, B200, GB300, and MI355X hardware in FP4.
Read article
Sep 20, 20261
- GenAI buildouts boosted Ethernet switch sales to $18.9B in Q2
IDC data shows global Ethernet switch revenues grew 43.4 percent year-over-year in Q2 2026, driven by datacenter AI network buildouts.
Read article
Sep 19, 20262
- SemiAnalysis benchmarked AI chips under time-to-first-token limits
Data from SemiAnalysis shows Nvidia B200 and Vera Rubin NVL72 achieve 55% to 88% lower token costs than AMD MI355X under strict latency caps.
Read article - SemiAnalysis published prefix cache benchmarks for LLM inference
The benchmarks track prompt token hit rates across HBM and host memory caches under increasing concurrency for hardware like Nvidia's B300.
Read article
Sep 18, 202611
- Engineers argue software efficiency must solve AI data center constraints
Cited research shows optimized prompt engineering can cut LLM inference energy and carbon emissions by 32% to 48%.
Read article - NSF launched the ELEGANT initiative for microelectronics in HPC and AI
The program funds projects to move low-dimensional semiconductor tech from labs into commercial manufacturing for AI and HPC hardware.
Read article - IEEE gave ORNL an award for its quantum computing collaborations
The lab received the IEEE Quantum Technical Community Synergy Award for its multi-sector research into hybrid quantum-classical HPC and quantum materials.
Read article - IQM will install a Spark quantum computer at Brazil's Eldorado Institute
The deal includes an on-site Spark quantum system by Q1 2027 alongside remote access to IQM's 54-qubit cloud system in Europe.
Read article - Mitsubishi Electric started R&D to scale quantum hardware
Japan's NEDO selected two projects to develop laser controls and cryogenic amplifier modules for neutral-atom, trapped-ion, and superconducting qubits.
Read article - DOE launched a $215M competition for fault-tolerant quantum systems
The program offers shared prize pools to vendors demonstrating verified quantum systems with at least 100 to 200 logical qubits by 2028.
Read article - DOE report sets roadmap for error-corrected quantum computing by 2028
A DOE advisory committee proposed a phased plan to reach error-corrected quantum computing by 2028 and integrate it with national lab HPC systems.
Read article - SemiAnalysis benchmarked DRAM and NVMe offloading for model inference
They found B200 DRAM offloading delivered 121 million tokens per dollar versus 52 million for SSDs at 125 tokens per second per user.
Read article - SK Hynix considers building a US NAND fab for Solidigm
Solidigm is scouting US East Coast sites for a NAND fab to bypass China export limits and build advanced flash for AI demand.
Read article - Huawei unveiled OceanStor M900 flash storage for AI clusters
They claim the all-flash system delivers 40 TB/s of bandwidth and cuts KV cache access latency to 60 microseconds for large AI clusters.
Read article - DeepMind published a 1PB dataset predicting 9 billion genetic variants
DeepMind says the one-petabyte dataset uses its AlphaGenome AI model to precalculate molecular effects for 9 billion possible single-letter DNA mutations.
Read article
Sep 17, 202613
- Lightmatter launched a bidirectional optical module for AI clusters
They say the 6.4 Tbps near-package optical engine halves fiber cabling by sending transmit and receive signals over a single fiber.
Read article - Bull operates as an independent supercomputer vendor after Atos split
Now fully owned by the French state, Bull is hiring 500 staff and pushing to source more supercomputing components from Europe.
Read article - NSF added three topics to its $1.5 billion X-Labs research program
The agency opened proposals for AI for physical systems and previewed upcoming funding for computing at the limits of physics.
Read article - SimOps Foundation outlined a framework for adopting HPC and AI simulation tech
The group argues engineering teams need continuous capability loops rather than one-off IT projects to keep up with fast-changing compute and AI tools.
Read article - DOE opened a $215M competition for 100-qubit fault-tolerant systems
DOE will fund up to $215M for private firms that can build fault-tolerant quantum systems running at least 100 logical qubits.
Read article - EUCLYD raised over 200M euros to build AI hardware
The Dutch startup says the funding will support developing custom AI silicon and low-power datacenter systems.
Read article - GlobalFoundries expanded SiGe chip supply for Marvell optical networking
GF will expand its Vermont fab to build more SiGe chips for Marvell's 200G-per-lane optical transceivers and co-packaged optics.
Read article - Alice & Bob partnered with CEA on hybrid quantum-HPC software
They plan to integrate Alice & Bob's quantum software into hybrid HPC workflows using Bull Qaptiva for many-body physics problems by 2030.
Read article - IonQ and Synopsys speed up LS-DYNA simulations using quantum algorithms
They claim using IonQ's trapped-ion hardware to reorder equations before solving cut Ansys LS-DYNA simulation times by up to 14.6%.
Read article - Nvidia posted first MLPerf inference numbers for Vera Rubin NVL72
Nvidia claims its Vera Rubin NVL72 rack delivered up to 3.7x higher inference throughput than GB300 NVL72 on Qwen3-VL.
Read article - Fujitsu will start selling its 144-core Monaka Arm chip in November
News Releases· Sep 17, 2026
The 2nm Armv9 processor targets datacenters and HPC, with Fujitsu claiming double the inference throughput of rival CPUs.
Read article - Bull beat HPE to build the Lumi-AI supercomputer in Finland
The upcoming upgrade will offer 10x the AI performance and around twice the FP64 throughput of the current Lumi-G partition.
Read article - Matsuoka concedes software 3D FFT in revised FP8 emulation papers
He says integer overhead slows 3D FFTs on GPUs, but models a hardware assist that could double emulated DGEMM speed on Nvidia Rubin.
Read article
Sep 16, 202610
- Infleqtion and Nvidia showed a 5.4-to-1 physical-to-logical qubit ratio
Using Nvidia's CUDA-Q Logical, they claim they created 18 logical qubits from 98 physical data qubits with a new error-correction code.
Read article - Engineer proposes Z-order memory layout to cut CPU cache misses
The proposal suggests baking Morton layout addressing into the MMU to keep multidimensional data clustered in CPU caches.
Read article - CoreWeave brought up multi-rack Nvidia Vera Rubin NVL72 clusters
The multi-rack clusters link hundreds of Rubin GPUs over Spectrum-X networking with 1.6 Tb/s of interconnect bandwidth per GPU.
Read article - Cohere and Aleph Alpha agreed to merge under the Cohere brand
The combined company will have over 1,000 staff across headquarters in Toronto and Berlin targeting sovereign AI for governments.
Read article - ORNL and IonQ used AI to speed up hybrid quantum optimization
Researchers say a transformer model kept quantum circuit synthesis to 28 seconds as problem sizes grew, replacing costly iterative tuning.
Read article - IBM's Anderon finalized a $1B CHIPS award for quantum wafer fab
The 300mm quantum wafer foundry in Albany finalized its $1 billion federal award and has begun running its first production wafers.
Read article - Google, Nvidia, and Emerald AI formed an alliance for flexible AI data centers
The group aims to standardize how AI data centers dynamically shift compute workloads and power draw to match grid constraints.
Read article - Arteris launched FlexGen Multi-Die NoC IP for AI and HPC chiplets
They claim its bidirectional UCIe support cuts PHY area, power, and I/O requirements by up to 50% across chiplet boundaries.
Read article - TotalEnergies and Mistral partnered to build AI models for oil and gas exploration
The companies launched a €100 million, three-year program to build AI models and a joint lab for subsurface geoscience and reservoir data.
Read article - ParityQC added its Twine quantum compiler to IBM Qiskit
They claim the tool cuts two-qubit gate overhead and circuit depth so users can run denser optimization problems on IBM QPUs.
Read article
Sep 15, 202612
- ai& will deploy Rebellions AI inference hardware in its Tokyo data center
ai& plans to deploy up to 100 RebelRack units in Tokyo as part of its broader push to reach 40 MW of data center capacity by 2027.
Read article - Astera Labs expanded its Leo memory controller line for AI clusters
They claim the new fabric-attached controllers can cut time-to-first-token by 62% for AI workloads by offloading KV cache over PCIe.
Read article - SDSC won a $10M NSF award to build the Expanse2 supercomputer
The AMD-powered system will feature an all-NVMe file system and enter production in 2027 to replace the original Expanse machine.
Read article - Liquid Compute raised $15M to build an exchange for AI compute
The startup raised $15 million to build a trading marketplace and financial clearing layer for AI compute capacity.
Read article - Altera filed draft paperwork for a public offering
The FPGA vendor submitted a confidential draft Form S-1 with the SEC for a proposed public offering of its common stock.
Read article - VDURA added Sentinel proactive monitoring to its AI storage platform
They say the tool analyzes storage telemetry against fleet data to automatically open support tickets with fixes attached.
Read article - SiFive and AMD demoed ROCm on RISC-V datacenter servers
They demonstrated LLM inference using ROCm 10 on a SiFive P870-D RISC-V host paired with AMD Radeon AI GPUs.
Read article - Micron demoed a 512GB DDR5 memory module for servers
They claim the 512GB DDR5 RDIMM reaches up to 9,200 MT/s and cuts operating power by over 60% compared to four 128GB sticks.
Read article - Western Digital highlights HDD role in growing AI data lakes
WD cited an IDC survey showing 61% of organizations experienced data growth over 25% from AI, arguing HDDs remain necessary for capacity.
Read article - HPE and Nvidia pitched integrated AI Factory infrastructure
HPE and Nvidia say their bundled infrastructure pairs Nvidia accelerators with HPE management tools for enterprise and sovereign AI workloads.
Read article - VDURA added automated fault monitoring to its parallel storage software
They say the tool analyzes fleet telemetry against known failure patterns to automatically diagnose faults and dispatch parts.
Read article - SemiAnalysis says datacenter moratoriums delay only 2.3GW in the US
They claim moratoriums have delayed only 2.3 GW of capacity, leaving 38 GW of US datacenter additions on track for 2027.
Read article
Sep 14, 202617
- Photonic proposed a $500M CAD quantum and AI chip fab in Canada
The company pitched a 500 million CAD multi-tenant facility to build quantum and AI semiconductors domestically.
Read article - Phasecraft used Nvidia GPUs and cuQuantum to speed up quantum chemistry
Phasecraft ran over 3,000 VQE molecular emulations on Hopper GPUs using cuQuantum, claiming a 15x simulation speedup.
Read article - Infleqtion integrated its qLDPC library with Nvidia CUDA-Q Logical
Infleqtion claims the setup validated an error-correction code using about six physical data qubits per logical qubit.
Read article - BlueQubit opened a $150,000 grant program for quantum and GPU compute
The initiative provides researchers with cloud credits for IBM QPUs and Nvidia GPUs on AWS to develop quantum algorithms and classical simulations.
Read article - AirJoule bought BitSink to expand into AI and HPC cooling
AirJoule paid $27 million upfront, with up to $40 million in earnouts, to acquire data center cooling and power vendor BitSink.
Read article - IonQ presented nine quantum computing papers at IEEE Quantum Week
They claim quantum-accelerated graph partitioning sped up finite-element simulation workflows by up to 14.6% on meshes with 35 million elements.
Read article - PPPL and UKAEA link AI supercomputing platforms for fusion
PPPL is federating its STELLAR-AI system with UKAEA’s SUNRISE supercomputer to share data and train AI models across both fusion labs.
Read article - Alchip showed off 2nm ASIC designs and advanced CoWoS packaging
The company says 90% of its designs are sub-7nm and highlighted 2nm development alongside TSMC CoWoS packaging.
Read article - Fujitsu plans November 2026 launch for MONAKA CPU and server
Fujitsu claims the 3.8GHz 2nm chip reaches 8800MT/s memory speeds and doubles AI inference throughput compared to competing CPUs.
Read article - Seagate says only 38% of enterprises are ready for AI storage growth
Seagate found that 99% of IT leaders expect AI to increase storage demands, but only 38% say they are fully prepared.
Read article - UCLA and Nvidia accelerated quantum state control simulations with AI
They claim a Fourier Neural Operator model predicted quantum dynamics up to 10 million times faster than GPU numerical simulations.
Read article - Mat3ra runs DOE project using AI to speed thin film design
They claim the project will use closed-loop AI and simulations to speed up semiconductor thin film development by up to 30x.
Read article - SemiAnalysis benchmarked Nvidia's Rubin NVL72 on agentic inference
SemiAnalysis claims Nvidia's Rubin NVL72 delivers up to 7x higher token throughput per megawatt than Blackwell on agentic inference workloads.
Read article - Huawei and ThinkParQ put BeeGFS directly onto OceanDisk arrays
The system runs BeeGFS inside storage controllers with 192 cores, delivering up to 175 GBps and 2 PB of raw NVMe capacity, they claim.
Read article - Quandela and Nvidia linked a photonic QPU to GPUs over NVQLink
Their white paper outlines how Nvidia GPUs handle decoding and simulation over RoCE while Quandela's FPGA controller runs the physical photonic QPU.
Read article - Oracle reported surging cloud revenue and heavy datacenter capex
Oracle posted $19.18 billion in quarterly revenue, driven by $6.5 billion in cloud CPU and GPU rentals while capex reached $28 billion.
Read article - SemiAnalysis published MI300X specs and inference benchmark data
The tracker shows the 192 GB MI300X hitting 2,615 FP8 TFLOP/s with estimated cloud pricing between $0.95 and $1.30 per chip-hour.
Read article
Sep 13, 20261
- IO500 moved Sugon ParaStor to the research list, returning Aurora to No. 1
The committee cited a lack of architectural details and limited file-system availability, moving the Chinese submissions off the production rankings.
Read article
Sep 12, 20262
- d-Matrix joins Nvidia's NVLink Fusion platform
AI chip startup d-Matrix has joined Nvidia's NVLink Fusion ecosystem to connect its silicon with Nvidia hardware.
Read article - Japanese startup ai& deploys alternative AI chips in data centers
More Than Moore· Sep 12, 2026
Startup ai& is building a sovereign AI cloud in Japan using brownfield sites and deploying accelerators beyond Nvidia and AMD.
Read article
Sep 11, 20267
- Article outlines data sovereignty hurdles across distributed AI infrastructure
The author argues that distributed GPU clusters and cross-border data pipelines make traditional geographic sovereignty rules inadequate for modern AI workloads.
Read article - QuantrolOx opened its first US office at UC Berkeley
The qubit automation startup opened its first US office at UC Berkeley to collaborate on automated tuning for quantum processing units.
Read article - Crux AI launched to build 500 MW of Google TPU infrastructure
Backed by $5 billion from Blackstone, the startup plans to bring 500 MW of Google TPU capacity online in 2027.
Read article - QumulusAI deployed Nvidia Blackwell B300 GPUs in Philadelphia
The rollout activates an $18 million, two-year take-or-pay contract with a GPU cloud marketplace.
Read article - EuroHPC picked NVision for its Quantum Grand Challenge
EuroHPC awarded NVision €293,455 to develop a scaling roadmap for its photon-interfaced spin-qubit architecture.
Read article - Argonne trained 75 researchers in extreme-scale computing at ATPESC 2026
The two-week program covered MPI, PETSc, quantum computing, and parallel I/O for graduate students, postdocs, and computational scientists.
Read article - Nvidia committed $530B to backstop AI data centers and compute supply
SemiAnalysis says Nvidia holds $530B in off-balance sheet guarantees covering AI datacenters, power leases, and memory supply.
Read article
Sep 10, 202614
- HP and Red Hat launched an AI workstation with Nvidia Grace Blackwell
HardwareBrief· Sep 10, 2026
They claim the HP ZGX Fury workstation delivers up to 20 PFLOPS of FP4 AI compute using Nvidia Grace Blackwell chips.
Read article - d-Matrix will integrate its memory-centric XPUs into Nvidia MGX racks
d-Matrix is pairing its upcoming Raptor 3D-RAM inference chips with Nvidia GPUs, claiming 100 TB/s of memory bandwidth for decode workloads.
Read article - ACM and IEEE CS named 2026 George Michael HPC fellows
Jie Ren, Amit Samanta, and Xuan Wu received fellowships for research on GPU solvers, hybrid cloud scheduling, and scientific data compression.
Read article - Ayar Labs raised $150M to manufacture co-packaged optics
The new round brings their 2026 total to $650M to scale up high-volume production of co-packaged optical interconnects for AI clusters.
Read article - CIQ updated RLC Pro AI to the Linux 6.18 kernel
They claim the updated stack loads AI models 33.8% faster and improves time to first token under CPU contention versus standard Enterprise Linux.
Read article - BSSw opened applications for 2027 scientific software fellowships
The program offers up to $25,000 to selected fellows at US institutions to fund projects that improve scientific software sustainability.
Read article - Qoro and Hartree Centre partner on hybrid quantum-HPC software
Qoro will integrate its Divi and Maestro tools with Hartree's Slurm scheduler via QRMI to route hybrid quantum-classical workloads.
Read article - EU project RItrain-DIGIT launches to train research staff in HPC and AI
The €1.5 million Horizon Europe project brings together PRACE, HLRS, and EGI to upskill IT staff running European research computing systems.
Read article - d-Matrix plans NVLink Fusion rack system with NVIDIA for AI inference
The system will pair d-Matrix's Raptor inference XPUs with Nvidia MGX racks using NVLink Fusion interconnects.
Read article - Positron AI raised $875M to tape out memory-heavy inference chips
They claim their TSMC N3P Asimov chip pairs with up to 2.3 TB of commodity LPDDR5X to avoid HBM supply bottlenecks.
Read article - IQM plans a three-stage rollout for the LUMI-IQ quantum system
They plan to deploy a 150-qubit machine in 2027 and upgrade it to run up to 9 logical qubits by 2029.
Read article - Bull won a €31 million contract to upgrade Spain's weather supercomputers
Bull will build a dual-site supercomputing system for Spain's AEMET that they claim will increase performance by 10%.
Read article - WEKA integrated its AI storage platform with Backblaze B2
The companies say the pairing lets AI clusters keep hot data on flash while offloading training checkpoints and raw datasets to B2 object storage.
Read article - Broadcom forecasts $230 billion in AI chip and networking sales by 2028
The company projects custom AI chip and networking revenue will hit $230 billion by fiscal 2028 across 30 gigawatts of compute for major AI labs.
Read article
Sep 9, 202615
- Qualcomm and Amazon will co-develop AI inference chips for AWS
Amazon committed to $60 billion in hardware through 2036 to co-develop custom inference silicon and 1.6 Tbps optical interconnects.
Read article - TSMC plans High-NA EUV production by 2030
TSMC plans to bring High-NA EUV lithography into volume manufacturing by 2030, initially using standard 6x6-inch photomasks.
Read article - CXMT started manufacturing eight-layer HBM3 memory
CXMT is producing 8-layer HBM3 stacks with reported manufacturing yields between 25% and 30% for domestic supply chains.
Read article - Panmnesia and Meta proposed a multi-rack CXL architecture for AI clusters
They claim custom CXL switches and controllers can create a single coherent memory pool across multi-rack AI accelerator clusters.
Read article - Lightbits launched Inferra to tier GPU KV cache to NVMe storage
They claim the software tiers KV cache data to SSDs over RDMA to enable up to 16x more concurrent inference sessions on existing GPUs.
Read article - Alice & Bob report warns quantum software stack lags hardware
They say quantum hardware is advancing, but compilers and schedulers needed to integrate systems with supercomputers remain unready.
Read article - Parallel Works and CoreWeave deployed an HPC cloud for DARPA
The managed platform gives DARPA's NODES researchers dedicated Nvidia HGX H100 systems and Slurm-on-Kubernetes orchestration for biological modeling.
Read article - Hyperion and Alice & Bob surveyed supercomputing sites on quantum integration
Based on 15 supercomputing center interviews, the report warns that immature software stacks risk eroding trust in quantum computing.
Read article - Amazon partnered with Qualcomm to develop custom AI inference chips
They plan to co-develop multiple generations of AI accelerators and 1.6T optical networking for Amazon data centers.
Read article - IDC study shows AI workloads drive enterprise storage demand
An IDC survey found 61% of organizations saw data volumes grow by 25% or more over the past year due to AI adoption.
Read article - Google plans to spend €13B on Finnish data centers and AI infrastructure
The company plans to invest at least €13 billion over 2027–2028 to expand data center and AI infrastructure across four sites in Finland.
Read article - RIKEN, IBM, and Cleveland Clinic named Gordon Bell finalists for quantum HPC run
They used IBM quantum processors alongside Fugaku and Miyabi-G to simulate a 12,635-atom protein complex.
Read article - Meta and Panmnesia design CXL fabric to connect 960 AI accelerators
They claim the CXL design cuts cross-rack latency to hundreds of nanoseconds and scales cache coherence to 960 accelerators.
Read article - Xanadu and ASML partner on lithography for photonic quantum chips
They are working together to improve patterning control and cut optical losses caused by line edge roughness in photonic quantum processors.
Read article - Sandia developed analog memory for more efficient computing
Sandia researchers built ETCRAM, an analog memory they claim offers 100x higher precision and 1,000x greater dynamic range than current tech.
Read article
Sep 8, 202613
- DeepSeek ordered 160,000 Huawei Ascend 950DT chips
DeepSeek is deploying 160,000 Ascend 950DT chips in Inner Mongolia to target 160 exaFLOPS of peak FP8 inference performance.
Read article - Arm launched Neoverse CSS N4 and AGI CPU for data center AI
Arm claims CSS N4 packs up to 128 cores per die, supports PCIe Gen 7, and offers 2x the performance and 1.75x the memory bandwidth of CSS N3.
Read article - Former Cray CEO Pete Ungaro says data movement matters more than FLOPS
Ungaro argued that memory bandwidth and networking matter far more than raw FLOPS as workloads hit data movement bottlenecks.
Read article - Imec built superconducting circuits with 30nm wiring for HPC and AI
Imec says its new NbTiN superconducting circuits reach 3.8 million junctions per square centimeter with 30nm wires on 300mm CMOS processes.
Read article - Qualcomm and AWS partner on AI inference chips and 1.6T optics
Qualcomm and AWS are collaborating on custom AI inference silicon and optical connectivity solutions up to 1.6T for data centers.
Read article - Intel Foundry and ASML detail High NA EUV chip manufacturing progress
Intel says it has processed over one million wafers with High NA EUV across R&D and production for select chip layers.
Read article - GlobalFoundries finalized a $375M CHIPS Act award for quantum R&D
The five-year award from the Commerce Department will support domestic manufacturing R&D for quantum chips and cryogenic CMOS.
Read article - EuroHPC and NAISS inaugurated the Arrhenius supercomputer in Sweden
The €68.5M HPE-built system delivers over 60 petaflops and placed 42nd on the June 2026 Top500 list.
Read article - PsiQuantum finalized a $100M CHIPS Act award for quantum hardware
The federal funding will support US manufacturing of optical switches, photon detectors, and packaging for fault-tolerant quantum systems.
Read article - Rigetti signed a $100M CHIPS Act deal with the US government
The US Commerce Department finalized a $100 million award for Rigetti to scale superconducting quantum hardware in exchange for a minority equity stake.
Read article - SkyWater launched a quantum semiconductor foundry unit
SkyWater is offering dedicated superconducting and silicon-nitride photonics manufacturing services for quantum computing and networking hardware.
Read article - HPE reported higher quarterly profit as datacenter equipment prices rose
HPE posted $12.21 billion in Q3 revenue, up 33.7 percent year on year, as customers absorbed higher prices for CPUs, GPUs, and networking gear.
Read article - Malaysia considers Huawei chips for sovereign AI project
Bloomberg reports Malaysia is evaluating Huawei's Ascend 910C accelerators for a RM2 billion sovereign AI infrastructure initiative.
Read article
Sep 7, 20264
- Dell predicts AI will make up 75 percent of datacenter demand by 2030
Dell claims AI will drive 75 percent of datacenter demand by 2030 as its AI server backlog reached $95 billion alongside surging legacy server upgrades.
Read article - SemiAnalysis tests Google TPUv7 inference against Nvidia Blackwell
SemiAnalysis reports Google TPUv7 delivers up to 50% better performance per dollar than Nvidia B200 and B300 chips in FP8 inference benchmarks.
Read article - Canonical detailed its enterprise RISC-V roadmap at Hot Chips
Canonical says 95% of Ubuntu packages build for RISC-V, but test pass rates sit at 55.1% compared to roughly 92% on x86.
Read article - University of Málaga used HPC to model tsunamis for Andalusia's emergency plan
Cheese2· Sep 7, 2026
Researchers used Spanish Supercomputing Network clusters to simulate 1,000 km of coastline at five-meter resolution to map tsunami risks.
Read article
Sep 4, 202612
- InferenceX published specs and benchmarks for Nvidia's B300 GPU
InferenceX says Nvidia's B300 packs 268 GB of HBM3e and delivers 13,500 TFLOP/s of dense FP4, 1.5x higher than the B200.
Read article - AMD unveiled the Threadripper Halo Station with MI350P GPUs
The liquid-cooled tower pairs a 96-core Threadripper Pro with up to four Instinct MI350P accelerators with 144GB of HBM3E memory each.
Read article - IBM launched Nighthawk r2 quantum processor
IBM claims its 120-qubit Nighthawk r2 runs over 100,000 circuits per second by adding dedicated reset elements to cut idle time.
Read article - Quantum Foundry Copenhagen plans European quantum chip fab
Backed by the Novo Nordisk Foundation, they plan to open a 5,300-square-meter quantum chip fab in Copenhagen by 2027.
Read article - Malaysia's APU opened an AI lab with Nvidia DGX Spark
APU opened a lab with Nvidia DGX Spark systems, claiming it lets students run and fine-tune models locally with up to 200 billion parameters.
Read article - Berkeley Lab scaled BerkeleyGW to simulate 10,000-atom materials on GPUs
Berkeley Lab and USC researchers used GPU supercomputers to run many-body excited-state simulations on moiré superlattices with roughly 10,000 atoms.
Read article - Oracle will deploy HPE Juniper networking across its AI data centers
Oracle will use HPE Juniper switches for RoCEv2 backend networks and work with HPE on fabric telemetry to catch cluster congestion faster.
Read article - Scientek partnered with Classiq to sell quantum software in Taiwan
Scientek will distribute Classiq's hardware-agnostic quantum design platform to semiconductor and research institutions across Taiwan.
Read article - Tape shipments rebounded in early 2026 after dropping in 2025
Tape archive capacity fell 9.2% to 160.3 EB in 2025, but Q1 2026 shipments jumped 57% year-over-year according to LTO consortium data.
Read article - InferenceX compared AMD MI325X and MI300X specs
The MI325X increases memory to 256 GB HBM3e and bandwidth to 6 TB/s while pulling 1,000 W TDP at unchanged FP8 compute.
Read article - InferenceX detailed Nvidia's GB300 NVL72 rack specs and benchmarks
The 72-chip rack packs 20 TB of HBM3e at 1,400 W per chip and reportedly delivers up to 2.83x higher throughput than the GB200 on DeepSeek-V4.
Read article - NEC reportedly halted quantum hardware development
Reports say NEC ended quantum computer development over long commercialization timelines while Fujitsu joined Japan's Moonshot project.
Read article
Sep 3, 202611
- Nvidia agreed to buy Hugging Face for $12.9 billion
The $12.9 billion deal is set to close in early 2027, bringing Hugging Face's hub of three million AI models under Nvidia.
Read article - TSMC raised its 2026 capex target to $64 billion
TSMC increased its 2026 capital budget to $60–$64 billion as projected equipment needs surged to nearly double original estimates.
Read article - George Mason University bought a $7.7M quantum system from TreQ
The university plans to install TreQ's open-architecture quantum system in early 2027 with commissioning expected later that summer.
Read article - Argonne and JPMorgan simplified quantum optimization simulations
Researchers mapped large QAOA problems to spin-boson systems to simulate algorithm performance on DOE supercomputers without full circuit runs.
Read article - NSF moved NAIRR to full operations with a $35M grant to SDSC and TACC
NSF gave SDSC and TACC $35 million over three years to run the operations center as the national AI research resource exits its pilot.
Read article - RIKEN adopted QunaSys SDK for its Fugaku hybrid platform
The SDK lets Fugaku and ROQUO users run hybrid quantum algorithms across IBM and Quantinuum hardware using MPI-based circuit simulation.
Read article - Jülich and eleQtron launched the JION trapped-ion quantum computer
The system connects to Jülich's supercomputers via JUNIQ so researchers can test hybrid classical-quantum workflows.
Read article - Quantinuum and Aramco partnered on industrial quantum computing
The companies signed a non-binding MOU to benchmark quantum systems and prepare for research into fault-tolerant computing for energy use cases.
Read article - Quantum Motion extended its Series C round for silicon quantum chips
They claim their silicon CMOS approach cuts quantum system costs 100-fold by using standard chip fabrication lines.
Read article - ZettaLane built a Lustre filesystem backed by cloud object storage
In MLPerf Storage tests, they claim MayaNAS reached 32.42 GB/s checkpoint write bandwidth writing straight to Google Cloud Storage buckets.
Read article - Greg Kurtzer created OpenWALDO to track AI training data provenance
The project uses a federated index to give AI models a verified bill of materials for their training data and licensing.
Read article
Sep 2, 202619
- UBS says China's lithography tools lag ASML by two decades
HW Compass· Sep 2, 2026
UBS reports Chinese toolmakers still struggle with 45nm mass production, putting domestic capabilities roughly on par with ASML in 2004.
Read article - UBS says Chinese lithography tools lag ASML by two decades
A UBS analyst reports China's domestic lithography tools mirror ASML's 2004 capabilities, with no local vendor yet reaching 45nm mass production.
Read article - Brookhaven Lab partnered with PsiQuantum to develop quantum algorithms
Researchers will use PsiQuantum's Construct platform to build fault-tolerant quantum algorithms for the DOE's Quantum Genesis initiative.
Read article - Enterprise SSD revenue doubled in Q2 2026 on high-density NAND demand
TrendForce projects top enterprise SSD vendors hit $37.59 billion in 2Q26, up 103.6% quarter-over-quarter as suppliers shift to 300-plus layer NAND and PCIe 6.0.
Read article - Hot Chips 2026 covered RISC-V standards and adoption
Presenters at Hot Chips 2026 detailed updates to RISC-V architectural standards and broader ecosystem adoption.
Read article - Vertiv bought UtilityInnovation Group for $1.45B
Vertiv agreed to pay $1.45B for UIG to bring microgrid controls and onsite power engineering in-house for AI data centers.
Read article - Equinix and CPP Investments completed their $4B buyout of atNorth
The $4 billion deal gives the buyers eight Nordic data centers targeting AI and HPC workloads, with plans to expand capacity across the region.
Read article - NMSU deployed VDURA storage for AI and HPC research
The system pairs NVMe flash and high-density HDDs into a single namespace on the university's InfiniBand fabric.
Read article - Everpure posted top MLPerf Storage results for model checkpointing
They say FlashBlade//EXA reached 877.52 GiB/s write bandwidth during a 1.25T parameter checkpoint benchmark with 30 nodes.
Read article - iPronics raised $125M to commercialize optical switches for AI clusters
They raised $125 million to ramp up production of their silicon photonics switches, which reconfigure AI cluster networks in real time.
Read article - IBM Ventures invested in BQP to back its hybrid quantum-classical physics solver
BQP raised an IBM Ventures-backed round to expand BQPhy, claiming its physics solver utilizes GPU capacity that classical sparse solvers leave idle.
Read article - Pasqal and True Nexus encode protein structures on neutral-atom quantum hardware
The companies say they successfully encoded protein structures on Pasqal's neutral-atom processor to model molecular mechanisms behind protein gelation.
Read article - Nvidia plans to acquire Hugging Face for $12.9 billion
Nvidia says Hugging Face will stay open and users will not be required to run workloads on Nvidia compute.
Read article - AMD unveiled a workstation with a 96-core CPU and four MI350P GPUs
The system pairs a 96-core Threadripper PRO with up to four liquid-cooled Instinct MI350P accelerators packing 576 GB of total HBM3E memory.
Read article - LTO consortium pushed its tape roadmap to LTO-14
The group reported 160.3 exabytes shipped in 2025 and says LTO-14 will hit 576TB native and 1,440TB compressed capacity per cartridge.
Read article - Nvidia plans to acquire Hugging Face for $12.9 billion
Nvidia announced a $12.9 billion acquisition of Hugging Face, stating the model repository will remain open without requiring Nvidia compute.
Read article - TSMC raised its 2026 capex budget to $64 billion
TSMC bumped its 2026 capex guidance to as much as $64 billion to build new fabs in Taiwan and the US as fab tool shortages persist.
Read article - CXMT and YMTC stockpiled ASML lithography tools ahead of tighter export rules
CXMT and YMTC secured enough DUV systems to cover their fab expansions for three years before expected new export curbs take effect.
Read article - LTO vendors shipped 160 exabytes of tape storage in 2025
LTO shipments reached 160.3 EB in 2025 as the roadmap expanded to LTO-14 with up to 576TB of native capacity per cartridge.
Read article
Sep 1, 20267
- MLCommons added KV cache, vector DB, and S3 tests to MLPerf Storage
MLPerf Storage 3.0 adds benchmarks for LLM KV cache offloading, Milvus vector database queries, and direct S3 object store access.
Read article - Samsung unveiled a roadmap to stack DRAM directly over processors
Samsung plans to vertically integrate DRAM directly over accelerator compute dies with zHBM, though they haven't set a release date yet.
Read article - ALCF automated supercomputing workflows for fusion experiments
Researchers use Globus and ALCF's Polaris supercomputer to run plasma simulations within the 15-minute window between fusion shots.
Read article - Nvidia detailed its use of RISC-V cores inside GPUs
At Hot Chips 2026, Nvidia showed how it integrates internal RISC-V cores across its GPU architectures.
Read article - Nvidia detailed NVHBM custom memory for AI accelerators
Nvidia claims NVHBM delivers 30% more bandwidth and 15% lower power than standard HBM4E by shifting memory controllers into the base die.
Read article - South Korea plans $919B buildout for 18.4GW of AI data centers
South Korea plans to deploy 18.4 gigawatts of AI data center capacity by 2035 through a $919 billion investment by domestic tech giants.
Read article - VMware launched on-premises AI cloud stack
VMware introduced an on-prem AI stack with OEM ReadyNodes and AMD Instinct MI350 GPU support for private model deployment.
Read article