Anyone wondering what is filling up their GPU's VRAM can now try gputaskman, a free, open-source tool from Microsoft VP Scott Hanselman. It runs in a terminal window with an htop-style layout and lists every process using the GPU with its memory footprint, engine activity, and loaded runtime, such as CUDA or Vulkan. A panel up top tracks utilization, clocks, temperature, and power. Hanselman shared the project on X last week and asked for pull requests.

Task Manager has shown GPU usage since 2017, and it will happily tell you that 15 GB of video memory is in use. It just won't say what is using it. gputaskman goes after that gap with a heavy focus on local AI. Its model inspector can identify the LLM a process has loaded and report its quantization, weights, and estimated KV cache cost, while Ollama and LM Studio users also get context size and proper model names. Hanselman calls it early, experimental software, and the AI breakdowns are estimates for now.

Gigabyte today announced the launch of AI TOP AERO, a new AI desktop powered by the NVIDIA RTX Spark platform. Positioned as a Personal AI Studio, AI TOP AERO combines an NVIDIA Blackwell architecture GPU, up to 128 GB of unified system memory, and Gigabyte's cooling and compact chassis design. The new desktop expands Gigabyte's AI PC lineup, bringing local AI development, content creation and gaming to personal workspaces.

NVIDIA RTX Spark unites NVIDIA AI, RTX graphics, and unified memory in one slim, powerful Windows PC. Designed to run larger AI models locally--without cloud dependency--and move between creative work, development, and immersive gaming without missing a beat. Connect apps, tools, and workflows into one seamless experience. Private by default. RTX Spark brings NVIDIA's full-stack AI platform, ready for the AI workloads of today and whatever comes next. Delivered on the familiar, trusted Windows platform.

NVIDIA is celebrating 20 years of the GeForce 8800 GTX, which hit store shelves on November 8, 2006, with a free tech demo on Steam. Called GeForce 8800 GTX - 20 Years of a Legend, it was created by Unreal Engine artist Dylan Browne, the developer of Through The Ages, with support from NVIDIA. The semi-interactive demo puts players in a 2006-era PC gaming room, where they can unbox the 8800 GTX, inspect it up close, and install it in an era-accurate PC tower, with GeForce easter eggs scattered around. The scene is rendered with path tracing, DLSS 4.5 Ray Reconstruction, Dynamic Multi Frame Generation, and RTX Mega Geometry. It is available now on Steam and is coming soon to GeForce NOW for Ultimate members.

Naturally, the 8800 GTX itself can't run any of this, which is quite ironic, to say the least. Steam lists an RTX 4070 as the bare minimum and recommends an RTX 5080, alongside 32 GB of system RAM and 30 GB of storage. That RAM figure alone is over 40 times the 8800 GTX's 768 MB of video memory. The recommended RTX 5080 is far from cheap right now either, with TechSpot's October survey putting its lowest US price at around $2,200. So a demo built around the thrill of unboxing a new GPU in 2006 wants one of the priciest gaming cards of 2026 to look its best.

The NVIDIA RTX PRO 5500 "Blackwell" workstation card has shown up at a US retailer, and it surprisingly costs more than the flagship RTX PRO 6000. As spotted by Alex Ziskind on X and corroborated by Videocardz, SHI Government Solutions listed the PNY-built model (VCNRTXPRO5500B-PB) with an MSRP of $17,999 and a discounted price of $17,408.63. NVIDIA's own US marketplace, on the other hand, lists the RTX PRO 6000 for $16,000, even though that card has more CUDA cores (24,064 vs. 21,760) and more memory (96 GB vs. 84 GB). So even at SHI's discounted price, the cut-down card costs around $1,400 more. NVIDIA hasn't announced official pricing since it quietly added the RTX PRO 5500 to its lineup in September, so it isn't clear whether the $17,999 figure comes from NVIDIA or PNY.

Videocardz points out that Chinese demand is probably a big factor behind the RTX PRO 5500's price. According to Reuters, ByteDance, Alibaba, and Tencent have ordered over 3 million RTX PRO 5500 cards at roughly $11,400 each, with total orders from China reportedly topping 6 million. One source said NVIDIA can make no more than 3 million of them in 2027. With the B300 barred from China and the H200 reviewed case by case, the RTX PRO 5500 is one of the strongest NVIDIA GPUs Chinese firms can still buy. The card also packs 28 GDDR7 modules at a time, even as memory prices keep climbing. With orders far ahead of supply, it's reasonable to expect RTX PRO 5500 pricing to stay high for a while.

Last week, modders found a way to double the tensor throughput of GeForce RTX 20, RTX 30, and RTX 40 cards, though Zagitta's original tool only ran on Linux. Now, Chinese modder physicsdolphin has released nv-uefi-unlocker, a UEFI tool that lifts the same limit on Windows. NVIDIA runs FP16, BF16, and TF32 tensor math with FP32 accumulation at half rate on GeForce cards, while the same silicon in professional cards runs at full speed. The tool removes that cap before the driver even loads, without any VBIOS flashing. An earlier Windows port from impacifist targeted a single RTX 3090 system, while this one covers all three generations.

The gains look substantial. One user's RTX 4070 Ti Super went from 57.00 to 150.93 TFLOPS in FP16 (2.65x), though board power also climbed from 227 W to nearly 292 W. An RTX 2070 reportedly jumped from 36 to around 60-63 TFLOPS. Not every card has been tested; RTX 20 owners have to enable GSP firmware mode first, and RTX 50 isn't supported at all. The unlock resets on every power-off, so it runs from a USB stick on each cold boot, with Secure Boot disabled or a self-signed certificate enrolled. That could be a problem for games whose anti-cheat requires Secure Boot.

NVIDIA's GeForce RTX 30-series and RTX 40-series graphics cards are based on the same dies as their professional-grade GPUs. This means the compute potential of consumer-oriented RTX GPUs is restricted by NVIDIA's firmware stack, but modders have managed to unlock the instruction issuing rate equivalent to their professional GPU counterparts. Starting with the GeForce RTX 30-series "Ampere," a researcher developed an unlock that effectively doubles the instruction issuing speed of FP/BF16 FMA into FP32 operations on GA102, GA103, GA104, and GA106 cards. As a result, cards like the GeForce RTX 3090 Ti experience up to 80% faster matrix multiplication in BF16, which is crucial for running nearly all neural networks. The unlock supports GA102, GA103, GA104, and GA106-based GeForce RTX graphics cards.

Next, we have the GeForce RTX 40-series "Ada Lovelace" with another project that has demonstrated a massive 2× speedup in FP16 and BF16 tensor throughput with FP32 accumulation. When the GeForce RTX 4090 was tested with FP16 tensor throughput and matrix accumulation, the GPU achieved 165.6 TeraFLOPS, while the unlock pushed that number to 331.6 TeraFLOPS. This is intriguing as these computational performance numbers are typically reserved for the higher-end professional segment like the NVIDIA RTX 6000 "Ada Lovelace." It seems NVIDIA artificially limited the AD102 silicon's performance in compute-sensitive matrix multiplication and accumulation, affecting the performance of running large language models. For this unlock, AD102, AD103, and AD106 are officially unlocked.

NVIDIA today announced the availability of NVIDIA RTX Spark laptops, bringing together RTX graphics and NVIDIA's full-stack AI platform in slim, premium Windows PCs for accelerated creation, immersive gaming and powerful local AI—rolling out from ASUS, Dell, HP, Lenovo, Microsoft and MSI.

Combining decades of NVIDIA innovation into a single superchip, RTX Spark delivers up to 128 GB of unified memory and 1 petaflop of FP4 AI performance to run large language models and AI agents locally. On a single device, creators, developers and gamers can move seamlessly between their everyday workflows and emerging AI agents on the familiar Windows platform.

ASUS today announced that the ProArt P16 and P14 laptops, powered by NVIDIA RTX Spark, are now available for purchase in Canada, with select configurations shipping starting today through the ASUS Store, Amazon Canada, Best Buy Canada, Memory Express and Newegg Canada. Purpose-built for creators, AI developers and workflow builders, these Windows PCs combine high-capacity unified memory, an NVIDIA Blackwell RTX GPU and an NVIDIA Grace CPU to support demanding local AI and creative workloads. The NVIDIA CUDA ecosystem and an expanding ASUS-exclusive agent-and-application solution make local AI easier to deploy, personalize and use, with the flexibility to connect cloud services when needed.

Together, Hermes Agent and ComfyUI deliver a compelling local-first, cloud-flexible AI experience. By bringing industry-validated workflows and agent-connected creative tools to users' own PCs, they enable generative AI creation and AI-assisted development, with the flexibility to integrate cloud services when needed.

ASUS today announced the global launch of the new ProArt P16 and P14 laptops, along with the compact ProArt GR1X mini PC, all powered by NVIDIA RTX Spark. Purpose-built for creators, AI developers and workflow builders, these systems boast high-capacity unified memory, a powerful NVIDIA Blackwell RTX GPU and an ultra-efficient NVIDIA Grace CPU. The NVIDIA CUDA ecosystem and an expanding ASUS-exclusive agent-and-application solution make sophisticated local AI easier to deploy, personalize and use. NVIDIA RTX Spark unites NVIDIA AI, RTX graphics, and unified memory in one slim, powerful Windows PC. Designed to run larger AI models locally—without cloud dependency—and move between creative work, development, and immersive gaming without missing a beat. Connect apps, tools, and workflows into one seamless experience. Private by default, RTX Spark brings NVIDIA's full-stack AI platform, ready for the AI workloads of today and whatever comes next. Delivered on the familiar, trusted Windows platform.

Together, Hermes Agent and ComfyUI deliver a compelling local-first, cloud-flexible AI experience. By bringing industry-validated workflows and agent-connected creative tools to users' own PCs, they enable generative AI creation and AI-assisted development, with the flexibility to integrate cloud services when needed.

Following what appears to have been a series of unfortunate events that led to a roadside laptop find resulting in confirmation that Microsoft will be launching a 20-core NVIDIA RTX Spark Surface Laptop, we now appear to have confirmation that Microsoft will also be launching a Surface Laptop Ultra on the other end of the Spark spectrum. According to Windows Central, Microsoft will reveal the full RTX Spark Surface line-up on October 7 at a special event in San Francisco.

Alongside the 20-core variant, the Windows Central report claims that Microsoft will launch a Surface Laptop with NVIDIA's 18-core RTX Spark. It is also claimed that the 18-core SKU will start with 24 GB of unified memory, while the 24-core version will start with 32 GB of memory. There will reportedly be a performance demonstration that shows off how the new Arm SoC delivers the same level of performance on battery life as it does when plugged into the wall.

Local AI is becoming more useful by the token. As AI agents move from experiments into everyday development, increasingly capable open models are shrinking to fit on more devices, giving builders more to run locally.

Coming this month, NVIDIA DGX Spark will be available with 64 GB of unified memory from top manufacturer partners—Acer, ASUS, Dell, Gigabyte, HP and MSI—giving developers, researchers and AI enthusiasts a new configuration with DGX OS and the NVIDIA AI software stack ready to use from day one.

Gigabyte today announced the launch of a new 64 GB unified memory version of Gigabyte AI TOP ATOM, expanding its existing 128 GB offering. Available starting October 23, the new configuration retains the hardware design of AI TOP ATOM, based on the NVIDIA DGX Spark platform, giving developers, researchers, and enterprise teams greater flexibility to select the configuration that best fits their AI workloads and memory requirements, while enabling dedicated on-premises AI development environments.

As generative AI and agentic AI applications continue to evolve, model testing, data processing, and application validation are becoming increasingly integral to everyday development workflows. Gigabyte AI TOP ATOM integrates AI computing capabilities into a compact desktop form factor, enabling users to perform model inference, prototype development, and data analysis in offices, laboratories, and educational environments. By running models and processing data locally, users can maintain greater control over development resources and project data while reducing reliance on cloud computing resources.

Gigabyte, a global leader in computer hardware, today announced the Gigabyte AI TOP 100 B850, a high-performance desktop engineered to bring workstation-class AI compute directly to the deskside. Designed to reduce cloud dependence and lower the barriers to local deployment, the system enables businesses, developers, and researchers to run advanced AI workloads locally, giving them greater control over sensitive data and AI infrastructure while reducing reliance on cloud and token-based services.

Built for Next-Generation AI Development
At the heart of the Gigabyte AI TOP 100 B850 is the 16-core, 32-thread AMD Ryzen 9 9950X processor, delivering the performance required for modern AI workloads ranging from local LLM fine-tuning and generative AI to deep learning research and agentic AI. The system comes equipped with 128 GB of DDR5 memory and a 2 TB PCIe Gen 4 SSD, providing ample capacity and high-speed access for large datasets, local model storage, and complex development environments.

NVIDIA today quietly added another card to its pro-viz product lineup with the RTX PRO 5500 "Blackwell" workstation GPU edition. Featuring 84 GB of GDDR7 memory, it is equipped with a GB202 die that includes 21,760 CUDA cores. This is the same CUDA core count as the consumer GeForce RTX 5090 SKU, but it is now certified for professional workloads and has a much higher memory capacity. The 84 GB of GDDR7 is connected to the GB202 with a 416-bit memory bus that delivers about 1,398 GB/s of memory bandwidth. This would put each GDDR7 memory chip at about 26 Gbps per pin, which is a standard GDDR7 memory speed, but not the fastest available on the market. Additionally, to achieve the massive 84 GB capacity, there need to be 28 memory modules. NVIDIA placed 14 GDDR7 modules on the front side with the GPU die and 14 GDDR7 modules on the back of the PCB, effectively sandwiching the PCB.

Compared to the rest of the RTX PRO lineup, the new RTX PRO 5500 "Blackwell" sits just a tick below the flagship RTX PRO 6000 SKU, which has a slightly higher CUDA core count of 24,064 and a 96 GB memory capacity. However, the RTX PRO 5500 is also much more powerful than the RTX PRO 5000 in the 5000 tier. The RTX PRO 5000 "Blackwell" has a much lower CUDA core count of 14,080 cores, with up to 72 GB of memory. Technically, the new SKU is much closer to the RTX PRO 6000 than the RTX PRO 5000 SKU, just with a slightly lower core count and memory capacity. The official specification sheet follows, courtesy of PNY.

HP Inc. today announced a collaboration with Red Hat, the world's leading provider of open-source solutions, to give organizations more choice in where AI workloads run, whether locally, in the cloud or across both environments. In collaboration with Red Hat, HP is developing an open, enterprise-grade AI platform to deliver purpose-built AI infrastructure powered by Red Hat AI Factory with NVIDIA.

HP's open enterprise-grade AI platform aims to help companies maximize local AI inference throughput with up to 20 PFLOPS FP4 AI performance, reduce environment setup time and deployment risk, and improve GPU utilization through optimized CUDA libraries, scheduling, and multi-GPU workload orchestration. Red Hat AI Factory with NVIDIA is an integrated AI platform, built on the industry-leading infrastructure of Red Hat Enterprise Linux and Red Hat OpenShift, for deploying and managing AI models, agents and applications across the hybrid cloud.

Manli is marking its 30th anniversary with the launch of the UF (Unlimited Firepower) series, a new graphics card lineup that joins its existing Nebula, Stellar, Polar Fox, and Gallardo families. Currently, it seems that two models are available at launch, both based on NVIDIA Blackwell architecture: the Manli UF GeForce RTX 5060 OC and the UF RTX 5060 Ti OC. Both are 2-slot cards measuring 324 x 143 x 42 mm with triple 92 mm fans, an aluminium baseplate, large radiators, and three 6 mm composite copper heatpipes. Idle fan stop is supported. Both graphics cards have 8 GB GDDR7 memory with a 128-bit interface, 28 Gbps speed, and 448 GB/s bandwidth. For connectivity, three DisplayPort outputs and one HDMI are available. On the GPU side, the RTX 5060 OC has 3,840 CUDA cores at 2280/2550 MHz base/boost and draws 145 W. The RTX 5060 Ti OC steps up to 4,608 CUDA cores at 2407/2632 MHz base/boost and a 180 W TDP.

A notable feature on both cards is G-Sense, a built-in tilt sensor that monitors GPU alignment in real time and sends alerts through a new Manli App if abnormal sag is detected. The app also handles ARGB customization and hardware monitoring. No pricing or store availability has been announced yet.

At IFA today, NVIDIA confirmed that the upcoming N1X SoC for the RTX Spark PC platform will come in two configurations. The top-end model will feature a 20-core "Grace" CPU and a GPU with 6,144 CUDA cores based on the "Blackwell" microarchitecture. NVIDIA envisions this version for both laptops and mini PCs, offering between 24 GB and 128 GB of LPDDR5X unified memory. The second configuration is slightly less powerful, with an 18-core "Grace" CPU, a "Blackwell" GPU with 5,120 CUDA cores, and 24-32 GB of LPDDR5X unified memory. This version is specifically aimed at laptops, which will be classified as Copilot+ AI PCs running Windows 11 for Arm.

Both chips are officially called "N1X," and there is no "N1" as previously thought. They are simply different configurations of the same chip, likely due to binning efforts during production. NVIDIA will launch these chips next month, and we expect to see systems from Acer, ASUS, Dell, MSI, HP, Microsoft, Lenovo, and many other OEMs. Pricing details are currently unknown, but with more memory, we can expect higher prices from any OEM, as memory is currently the most scarce resource.

At IFA 2026, ASUS is showcasing its new generation of ProArt AI creator PCs - the P16 and P14 laptops and GR1X mini PC - powered by NVIDIA RTX Spark, with hands-on demonstrations at the ASUS IFA Media Showroom. RTX Spark pairs a Blackwell RTX GPU with 6,144 CUDA cores and a 20-core NVIDIA Grace CPU, giving creators and developers on-device performance with fully private local data and full workflow customization - without relying exclusively on the cloud.

The ProArt P16 & P14 come with the thinnest ProArt laptop yet with RTX performance due to its integrated chip design - plus up to 99Wh battery capacity. Both feature ASUS Lumina Pro OLED displays with Delta E < 1 color accuracy, up to 1,600 nits peak brightness and 120 Hz visuals. The compact GR1X mini PC brings performance to a 150 x 150 x 51 mm desktop built for 24/7 AI workloads, available in Nano Black or a new Neo White.

NVIDIA's GeForce RTX 5060 mid-range graphics card is officially based on the GB206 GPU die SKU, equipped with 3,840 CUDA cores, 120 TMUs, 48 ROPs, and a total of 8 GB of GDDR7 memory running on a 128-bit bus. This is the standard configuration, which is supposed to be the same across every RTX 5060 GPU you buy from any of NVIDIA's add-in board partners. However, a review of the ASUS DUAL GeForce RTX 5060 OC D7 8 GB by Intek & Company, Quasar Zone revealed that the GPU die inside this graphics card is actually a GB205-200-KB-A1 GPU die SKU. This die is usually found in NVIDIA's GeForce RTX 5070, RTX 5070 Ti Mobile, and RTX PRO 3000 "Blackwell" mobile graphics cards.

When NVIDIA does these GPU SKU crossovers, it usually comes down to binning. Sometimes, GPUs of a lower bin do not meet all the parametric yields, meaning not all CUDA cores are functional. In such cases, the company can repurpose these GPUs as a lower-tier design, as is the case with the RTX 5060. Instead of this SKU going into an RTX 5070 and lacking CUDA cores, ROPs, or TMUs, it is cut down to RTX 5060 specs and used as an RTX 5060 base. No performance difference should be present.

NVIDIA has officially brought its RTX Neural Texture Compression (NTC) technology to Windows-on-Arm, gearing up for a general release on its RTX Spark PC platform. First demonstrated at GTC 2026 in May, NVIDIA showcased how NTC can significantly reduce GPU VRAM usage by up to seven times. In a technology demo at one of the GTC 2026 sessions, NVIDIA revealed that NTC can lower VRAM usage from 6.5 GB to just 970 MB in certain scenes. NTC technology is an AI-driven texture output that enables games to feature high-quality, complex materials without sacrificing performance. Games can take advantage of the substantial VRAM reduction that NTC offers while maintaining the same texture quality.

For the upcoming RTX Spark, NVIDIA is enhancing the software ecosystem so gamers and future buyers receive a mature software that works seamlessly out of the box, eliminating the need for workarounds. Even Japan's legendary SEGA has released AAA titles optimized for RTX Spark and is dedicating more resources to support the ecosystem. NVIDIA has already started shipping its first preview build of the CUDA toolkit for the RTX Spark platform, specifically as a native Windows-on-Arm software toolkit. Now, emerging technologies like NTC will run on RTX Spark, which includes configurations such as one with 6,144 CUDA cores and another with 5,120 CUDA cores. Both configurations are based on the "Blackwell" GPU microarchitecture and are paired with a 20-core Arm CPU and up to 128 GB of LPDDR5X memory.

A long-time TechPowerUp forum user who goes by Fouquin just posted a surprisingly in-depth review of the upcoming Microsoft Surface Laptop Ultra, featuring the NVIDIA RTX Spark Arm-based SoC. In a bizarre stroke of luck, the user says that he found the Microsoft Surface Laptop Ultra on the side of the road and decided to pick it up "assuming it was trash."

While the review is not an official TechPowerUp review, so it cannot be directly compared to our existing benchmarks for other laptops and CPUs, the user has done surprisingly thorough testing of the Laptop Ultra using the Phoronix test suite. While the Surface Laptop Ultra in question was running very old graphics drivers initially, Fouquin was able to update the drivers to NVIDIA preview version 616.00, which added CUDA support for the built-in Blackwell GPU. The version of the Spark SoC present in the test unit is the top-end variant with 6,144 cores and a 20-core CPU.

NVIDIA has released a developer preview of GeForce drivers for its RTX Spark PC platform, confirming the variants of the N1X chip and the SKU separation that will occur. In GeForce driver version 616.00, this marks the company's first driver package compiled for the Windows-on-Arm (WoA) platform. Thanks to Windows Latest, which extracted the installation file and examined the information file, it appears there are two main RTX Spark N1X chip SKUs: one with 6,144 CUDA cores and another with 5,120 CUDA cores. Both are based on the "Blackwell" GPU microarchitecture and are paired with a 20-core Arm CPU and up to 128 GB of LPDDR5X memory.

NVIDIA is working with Microsoft to optimize Windows 11, and developers now have access to the optimization techniques needed to get their AAA titles, productivity software, and other applications ported and optimized for running on NVIDIA's latest platform. A few days ago, we noted that NVIDIA is rolling out the CUDA toolkit for its RTX Spark platform, specifically as a native Windows-on-Arm software toolkit, and this driver version 616.00 is part of that toolkit.

NVIDIA has officially released its first preview build of the CUDA toolkit for its RTX Spark platform, specifically as a native Windows-on-Arm software toolkit. This means that NVIDIA's partners, ecosystem developers, and game developers now have the means to perform native platform optimization for the latest NVIDIA chips. This is particularly important since the platform is one of the few Windows-on-Arm designs available on the market. For decades, modern PC gaming has focused on x86-based operating systems, giving Intel and AMD an advantage in gaming scenarios. Consequently, most indie and major AAA titles have been developed natively for x86 processors, with Arm making little progress in this area. However, NVIDIA expects this to change as the company begins rolling out optimization toolkits to create native, optimized Arm builds for its RTX Spark mini PCs and laptops.

Earlier this year, we learned that NVIDIA might be planning to introduce a 9 GB edition of the GeForce RTX 5050 graphics card. However, later rumors suggested that the development was paused, and the latest reports claim that the project has been entirely canceled. According to MEGAsizeGPU, the GeForce RTX 5050 9 GB edition has been canceled because NVIDIA is currently relaunching its GeForce RTX 3060 12 GB edition GPU, which would compete with its own revived product. To avoid this overlap, the company has reportedly canceled the RTX 5050 9 GB edition launch. Initially, NVIDIA aimed to introduce this 9 GB version to address memory shortages in the lower-end segment. Instead of using four GDDR6 modules with 2 GB capacity each, NVIDIA planned to switch to three modules of GDDR7 with 3 GB capacity each, reducing the number of memory modules needed for this GPU.

For example, the current RTX 5050 uses 8 GB of 20 Gbps GDDR6 memory on a 128-bit bus, providing 320 GB/s of bandwidth. The 9 GB model was expected to use three modules of GDDR7 memory, each with 3 GB of capacity, resulting in a total of 9 GB across a 96-bit memory bus. While the narrower bus decreases the interface width, the switch to the new 28 Gbps GDDR7 memory would increase total memory bandwidth to 336 GB/s, a roughly 5% improvement, along with a 12.5% boost in VRAM capacity. Earlier leaks also claimed that NVIDIA was using the GB206 as a GPU base, while the older RTX 5050 8 GB with GDDR6 used the GB207 die. However, the core count remained at 2,560 CUDA cores. Since the GeForce RTX 3060 12 GB GPU is already on a path toward revival, launching this GPU wouldn't make sense.

SK hynix Inc. and NVIDIA today announced a multi-year technology partnership to advance next-generation memory for the global AI factory buildout and accelerate semiconductor design and manufacturing. The agreement builds on years of deep co-engineering collaboration that has powered some of the world's most advanced AI computing platforms.

"SK hynix and NVIDIA have been building toward this for years, and this partnership reflects the depth of that collaboration," said Chey Tae-won, Chairman of SK Group. "Together, we are co-developing the next generation of memory for AI factories and applying AI to how we design and manufacture semiconductors - work that will shape the future of AI infrastructure."