AI Hardware News
Product launches, GPU shortages, and market shifts tracked since October 2025.
April 2026
Listings for the 9950X3D2 have started to pop up at many different vendors, with some even listing preliminary prices. We found the CPU going for roughly $1,000 at two Canadian retailers, and just a s
Wildcat Lake is Intel's upcoming family of low-budget and low-power CPUs intended for OEMs. We've already seen many leaks surrounding this family, but now a new product from Advantech has listed three
Nvidia delivered 'five architectural breakthroughs' with the launch of the Tesla P100 accelerator 10 years ago today.
Computer scientist showcases their hardware multiplication ISA card accelerator for Intel 8086 and Intel 8088 PC systems.
A 42-core SKU from the upcoming Nova Lake-S CPU family has reportedly been upgraded to 44 cores by swapping the 6P+12E tile with an 8P+12E tile, allowing the chip to achieve symmetry across its dual-t
The company behind the tiny box AI accelerator says that its macOS driver for Nvidia eGPUs has just been signed by Apple, making it a legitimate software for Macs and no longer needs workarounds to wo
The Arctic Freezer 4U-SP5 is something really interesting. As the name suggests, it is a 4U cooler that is designed to cool AMD SP5 platform processors. For those who are buying AMD EPYC 9004/9005 ser
Nvidia has just demoed its Neural Texture Compression technique again at a GTC talk, where it showed VRAM usage dropping from 6.5 GB to just 970 MB in a scene. NTC uses a neural network to decompress
Windows 11 24H2 systems are being upgraded to 25H2 automatically, as Microsoft aims to streamline updates before support ends.
Two new Rowhammer attacks for GPUs have been discovered that can cause bit flips in VRAM to gain arbitrary read/write access over it. These attacks target page files and the page directory that are ot
For the holiday weekend, we take a look at the Sparkle Intel Arc A310 ECO, a compact and low-power GPU for when you have modest needs The post SPARKLE Intel Arc A310 Eco 4GB Low Profile PCIe GPU Quick
Intel's chip matches similarly-priced AMD parts in gaming and mercilessly wallops them in multi-threaded work.
As hyperscalers seek efficiency and control from custom CPUs they build in house, they adopt Arm and 90% of servers running custom silicon will use the Arm ISA in 2029.
Score Intel's latest Core Ultra 7 270K Plus with a Z890 motherboard and 32GB of DDR5-6000 memory for $240 off with this bundle.
We interview experts, including Chris McGuire, senior fellow for China and emerging technologies at the Council on Foreign Relations, and former senior director in the U.S. National Security Council o
SemiAnalysis estimates on Thursday that memory will account for roughly 30% of total hyperscaler capex in calendar year 2026.
At NVIDIA GTC 2026, we saw two new MiTAC servers with next-generation CPUs that we have not seen before, along with their GPUs and storage The post MiTAC Shows Servers with Next-Gen CPUs and Solidigm
March 2026
In our ASRock Rack TURIND8-2L2T review, we see how this motherboard manages to fit big processors while also exposing 128 PCIe Gen5 lanes The post ASRock Rack TURIND8-2L2T Review A Flexible AMD EPYC S
Blackwell-based industrial platform delivering 5,581 FP4 TFLOPS for robotics, manufacturing, and medical edge AI. Developer kits available from distribution partners.
336B transistor dual-die Rubin GPU on TSMC 3nm with HBM4. NVL72 racks train models with 1/4 the GPUs vs Blackwell. AWS, Azure, Google Cloud deploying H2 2026.
Administration revokes the per-country GPU quota rule that would have mandated export limits on AI hardware by nation.
A new LLVM patch has added V_FMA_F32, a 3-operand fused multiply-add (FMA instruction and introduced the VOPD3 instruction format for RDNA 5. Both of these changes should make it easier for compilers
Meanwhile, the older 7G105 GPU name no longer appears on the site; all four cards now sit under the LX product branding.
A motherboard that can accept both DDR4 and DDR5 memory can be the difference between you being able to build a new PC or putting it off till the shortage is over. ASRock's new H610M Combo II is other
The Ryzen 7 9800X3D is down to an all-time low price, making it a more enticing option than the slightly more performant but also much more expensive Ryzen 7 9850X3D.
A Redditor only paid $86.98 for a Ryzen 5 7600X, a Gigabyte B850 Eagle motherboard, and a 32GB DDR5-6000 Corsair Vengeance RGB kit from Newegg. The CPU even included a free 240mm Cooler Master AIO wor
U.S. Department of Commerce withdraws draft export rule for AI accelerators that would give the U.S. government ultimate power over export controls and mandate investments of foreign firms in the U.S.
Chinese tech giant circumvents US export controls by accessing NVIDIA Blackwell GPUs through third-party Malaysian datacenter.
Microsoft says "monitors can now report refresh rates higher than 1000 Hz" in the patch notes for its latest Windows 11 Insider builds. These updates are part of the Release Preview channel, which mea
AMD, Nvidia, Microsoft, Broadcom, and Meta have formed an Optical Compute Interconnect Multi-Source Agreement (OCI MSA) to develop a standardized optical interconnect for AI data centers, with the asp
At GDC 2026, Nvidia held a presentation somehow aimed at gamers and not data center clients. Still, it was sprinkled with AI pat-on-the-backs, with the company touting that its future gaming GPUs will
Nvidia says it is permissible for ByteDance to use AI clusters outside of China to develop its AI prowess as long as these clusters are built in compliance with the U.S. export controls.
Score on a great combo deal with AMD's Ryzen 7 9850X3D, Asus ROG Crosshair X870E Apex, 32GB Corsair DDR5-6400 RAM for $1,187.99 - it's $436 off retail, effectively making the RAM free.
DirectStorage 1.4 brings along key upgrades to the API, including support for Zstandard compression as well as CreatorID for improved GPU scheduling.
V-Color announces new 1+1 DDR5 memory kits with one real memory module and one filler module for AMD-based systems.
Meta shared more on its MTIA 300 AI accelerator, already in production, and future MTIA 400, 450, and 500 generations for its infrastructure The post Meta Outlines New MTIA Accelerator Roadmap for its
AMD, Broadcom, and Nvidia team up with Microsoft, Meta, and OpenAI to develop protocol-agnostic optical scale-up interconnects for AI clusters.
We take you on a tour looking a the Supermicro NVIDIA B300 generation versus the NVIDIA B200 generation and the other components of an AI factory The post Checking out the Supermicro NVIDIA B300 Solut
New Intel Core Ultra 250K and 270K Plus chips refresh the company's Arrow Lake lineup for enthusiasts who want a bit more The post Intel Intros Core Ultra 250K and 270K Plus Chips Refreshing Desktop A
Potential new restrictions under consideration beyond existing China-focused AI hardware rules.
Up from $1,999 MSRP at launch. Meaningful retail availability not expected until Q3 2026.
February 2026
Founders Edition jumps $700 from $3,999 due to DRAM supply constraints driven by AI datacenter demand.
Firmware update reduces cores on already-sold cards. Company claims 1-2% performance impact but offers no explanation.
Six weeks into the GPU shortage, NVIDIA pauses new RTX 50-series production due to AI-driven DRAM demand.
Second-largest penalty in Bureau of Industry and Security history for exporting ion implantation equipment to China.
financial
Liquid-cooled RISC-V workstation with four Blackhole ASICs runs 120B parameter models locally. Ships Q2 2026.
January 2026
Four weeks after launch, RTX 5090 aftermarket prices reach $5,000. Retail shelves empty across Amazon, Best Buy, and Micro Center.
Global DRAM memory crisis driven by AI datacenter demand creates consumer GPU shortages. RTX 5090 supply constrained at launch.
Tariffs imposed on chips meeting AI performance thresholds under Section 232 of the Trade Expansion Act.
HBM3e production constraints push H100 and H200 server pricing upward as AI companies buy memory years in advance.
First client processor on Intel 18A node. 14 SKUs with 50% performance uplift over Lunar Lake, shipping January 27.
MI400-series on TSMC 2nm with CDNA 5 architecture. Helios integrates 72 MI455X accelerators with 31TB HBM4 delivering 2.9 exaFLOPS.
Next-gen Rubin platform combines Vera CPU + two Rubin GPUs. 5x inference performance, 10x lower cost per token vs Blackwell. Available H2 2026.
December 2025
NVIDIA purchases 214.7M Intel shares at $23.28/share after FTC approval. Companies to jointly develop datacenter and PC products using NVLink.
Blackwell architecture comes to professional workstations with 72GB VRAM for AI development.
November 2025
Datacenter GPU prices up 15-30% year-over-year heading into 2026 as enterprise AI deployments accelerate.
October 2025
World's smallest AI supercomputer delivers petaflop performance with 128GB unified memory. Available from Acer, ASUS, Dell, Lenovo, MSI.