Artificial Intelligence - Page 3

AMD has just shown off its next-gen world-first 2nm EPYC "Venice" CPU with Zen 6 cores, and its Instinct MI455X AI accelerator, ready for its next-gen Helios AI racks.

The company unveiled its new Helios AI rack at its recent Financial Analysts Day 2025, promising some more performance numbers with class-leading performance and efficiency for AI workloads of the future. The new AMD Helios AI rack features a full liquid-cooled design with 4 x Instinct MI455X AI GPUs and a single Zen 6-based EPYC "Venice" CPU.

Helios AI racks use AMD's new Pensando "Salina" 400 DPU and Pensando "Vulcano" 800 AI NIC for networking and interconnection. AMD's next-gen EPYC "Venice" CPUs come with up to 256 cores based on the Zen 6c architecture, and each Instinct MI455X AI GPU packs a ton of GPU cores and next-gen HBM4 memory.

AMD confirmed at CES 2026 that its next-generation Instinct MI500 AI accelerator will be fabbed on TSMC's new 2nm process node, and be powered by the next-gen CDNA 6 architecture and next-gen HBM4E memory.

We will see AMD launching its next-gen Instinct MI500 series AI accelerators in 2027, as the company is moving into a faster annual cadence of releases in order to catch up with NVIDIA, and a similar way to how NVIDIA uses its standard and "Ultra" offerings. For example, NVIDIA has Blackwell and then Blackwell Ultra, Rubin and Rubin Ultra.

AMD provided some more concrete details about its MI500 at CES 2026 this week, confirming that it will be fabricated on an advanced 2nm process node at TSMC, the new CDNA 6 architecture (MI400 uses CDNA 5), and next-gen HBM4E memory (the next-gen standard after HBM4).

SK hynix showcased its next-gen memory solutions for AI at CES 2026, showing off its new 48GB HBM4, LPDDR6, SOCAMM2, and more for AI platforms of the future.

SK hynix showed off its next-gen 16-Hi HBM4 with 48GB, newer HBM4 that will succeed the upcoming 12-Hi HBM4 with 36GB that will arrive this year. The faster 16-Hi HBM4 48GB modules are bloody fast, with 2TB/sec of memory bandwidth per stack, destined for NVIDIA's next-gen Vera Rubin AI platform.

The company had its new 16-Hi HBM4 48GB running at the industry's fastest speed of 11.7Gbps, and is still under development at SK hynix, and will be released in the nearish future.

RTX Video Super Resolution is like DLSS for watching videos on YouTube or other streaming platforms: it takes a lower-resolution video, like 720p, and leverages AI to upscale it to 4K, delivering a sharper, more detailed image. RTX Video like DLSS leverages the Tensor Cores on GeForce RTX graphics cards for real-time upscaling.

At CES 2026, as part of a wide range of updates for RTX AI on GeForce RTX GPUs, NVIDIA announced that RTX Video will be coming to the popular, open-source AI platform ComfyUI in February. This means users with GeForce RTX GPUs will be able to take 720p AI-generated videos and upscale them "to 4K in seconds."

With the sheer computational power required to generate 4K AI video and images, most AI enthusiasts with a standard PC or laptop built for RTX AI create this content at lower resolutions, such as 720p.

NVIDIA founder and CEO Jensen Huang proudly took the stage at CES 2026, unveiling the company's next-generation Rubin AI platform.

NVIDIA's new Rubin AI platform is the successor to its dominant Blackwell AI chips, with Rubin being the first extreme-codesigned, six-chip AI platform, with Jensen adding that it's now in full production. NVIDIA is aiming to "push AI to the next frontier" with Rubin, not just offering far more computing power, but slicing the cost of generating tokens to around 1/10 of Blackwell, making large-scale AI "far more economical to deploy".

The use of extreme codesign means that designing all of the components together is essential because scaling AI to gigascale requires tighter integration innovation between chips, trays, racks, networking, storage, and software to remove bottlenecks. This massively reduces the costs of training and inference, added Huang.

Intel's next-generation Jaguar Shores data center AI accelerator platform is rumored to be using newer HBM4E memory, which could launch sometime in the second half of 2027.

Back at the Intel AI Summit Seoul in South Korea back in July 2025, the company seemed to be ready for HBM4 from SK hynix for Jaguar Shores, with a release in 2026. However, Intel hasn't had a stable or successful time with its Gaudi AI accelerators and its release schedule, and it knows it has almost insurmountable competition from AMD and more so NVIDIA, so timelines can change, and specifications -- like using faster HBM4E -- can change, too.

The new information regarding Intel's use of HBM4E on its next-gen Jaguar Shores AI platform is from leaker @Bionic_Squash on X, who replied to @harukaze5719 regarding Jaguar Shores using HBM4, with a simply reply of "Jaguar is HBM4E".

Samsung, SK hynix, and Micron are all fighting each other in developing new 16-Hi HBM, because NVIDIA requested supply of the new memory chips for the second half of 2026.

16-Hi HBM hasn't been commercialized previously with many technological hurdles to overcome, including DRAM stacking, as things get far more complicated with more HBM stacks. In a new report from the Electronic Times, NVIDIA reportedly requested that domestic and foreign memory manufacturers deliver 16-Hi HBM memory chips by Q4 2026.

SK hynix and Samsung Electronics in South Korea, as well as US-based Micron, have all begun full-scale development work for the mass production supply of 16-Hi HBM memory chips to NVIDIA. The outlet reports that concrete contracts haven't been signed yet; it reports that discussions are happening internally regarding the initial production volumes of 16-Hi HBM chips.

NVIDIA has teamed with SK hynix on a next-gen ultra-powerful SSD solution for AI inferencing, that could offer 10x the performance for AI inferencing in the middle of the worst DRAM crisis ever.

SK hynix has formalized development of the next-gen SSD with NVIDIA, after the South Korean memory giant enjoyed great results from supplying HBM to NVIDIA for its AI GPUs, and its customer- and service-tailored product development is expanding into the NAND flash sector.

In a new report from Korean outlet Chosun, SK hynix Vice President Kim Cheon-seong said at the recent "2025 Artificial Intelligence Semiconductor Future Technology Conference" (AISFC) that SK hynix was developing a new SSD with 10x more performance with NVIDIA. This new SSD is dubbed "Storage Next" for NVIDIA and "AN-N P" (AI NAND performance) for SK hynix, a new proof of concept that is in the works with the goal of releasing a prototype before the end of 2026.

KIOXIA's open-source AiSAQ (All-in-Storage ANNS with Product Quantization) has been a game-changer for running complex AI models by offloading vectorized data from expensive DRAM to SSD storage. With memory limitations and costs playing a significant role in which AI workloads can or cannot run, AiSAQ delivers a low-latency, scalable solution for Retrieval Augmented Generation (RAG) pipelines.

This week, KIOXIA announced that AiSAQ has been integrated into Milvus, one of the world's most widely adopted open-source vector databases. Starting with version 2.6.4, AI developers and enterprises can tap into the power of AiSAQ to scale AI applications with SSD storage. With the growth in RAG demands and the size of vector databases for inference, scaling DRAM is often not an option due to the exponential increase in cost.

KIOXIA's open-source AiSAQ is groundbreaking because it dramatically reduces DRAM requirements for running complex AI workloads, opening the door to large-scale system deployment that's more affordable and easier to scale, thanks to large capacity and fast SSD storage.

US deep-tech AI startup Tiiny AI has just unveiled the world's smallest personal AI supercomputer, with the new Tiiny AI Pocket Lab, which has been officially verified by the Guinness World Record under "The Smallest MiniPC (100B LLM Locally)".

This is the first global unveiling of the new Tiiny AI Pocket Lab, which will fit in your hands -- or your pocket, duh -- and is capable of running up to a full 120-billion-parameter LLM (Large Language Model) entirely on-device, without the need of cloud connectivity, servers, or high-end GPUs.

Tiiny has developed its super-small AI supercomputer for energy-efficient personal intelligence, and the Tiiny AI Pocket Lab runs within a 65W power envelope. The new Tiiny AI Pocket Lab enables massive AI model performance at a fraction of the energy and carbon footprint of traditional GPU-based systems.

NVIDIA has built a new location verification technology that would help the company know where its AI GPUs are being operated, in a bid to help prevent its AI chips from being smuggled into countries where US exports apply.

In a new report from Reuters, the outlet said that NVIDIA has demonstrated its new location verification feature privately over the last few months, but hasn't released it just yet. When it's released, it'll act as a software option that customers and data center operators can install, so they can keep a tighter eye on the AI chips in their fleets.

The newly-developed software from NVIDIA is an opt-in, customer-installed service that keeps an eye on GPU usage, configuration, and errors. It has a pretty decent feature set, providing data center operators with the following abilities:

US authorities have busted an AI chip trafficking network that was attempting to send $160 million worth of NVIDIA H100 and H200 AI GPUs to China, as the smugglers were changing the shipments' final destination.

In a press release issued by the U.S. Department of Justice, authorities reported a trafficking network in Houston, Texas, that has been convicted of smuggling NVIDIA AI chips to China using a "complex scheme". Court documents reveal two individuals -- Alan Hao Hsu, and those who worked for his company, Hao Global LLC -- attempted to export NVIDIA H100 and H200 AI GPUs worth $160 million by manipulating official paperwork and hiding the "ultimate destination of the GPUs".

The network itself was busted by the discovery of a wire transfer that began in the People's Republic of China (PRC), with the NVIDIA AI GPUs shipped to US warehouses and then rebranded as "SANDKYAN", allowing the group to misclassify the AI GPUs and then export them.

NVIDIA CEO Jensen Huang sat down with Joe Rogan last week, where he provided some thoughts on his opinion of LLMs of today, turning into Terminator, tomorrow. Check it out:

Joe asked Jensen what he thought about AI and LLMs (Large Language Models) of today, which include generative AI, edge AI, agentic workflows, and more AI workloads. LLMs are super advanced right now, where they will replace many humans in their roles in the coming years, but some worry AI-boosted capabilities will go "too far", where they replace human beings as the "apex species".

Joe Rogan asked: "Well, I don't assume that it would do harm to us, but the fear would be that we would no longer have control and that we would no longer be the apex species on the planet. This thing that we created would now be. Is that funny?"

With the release of NVIDIA CUDA 13.1, the company is introducing the "largest and most comprehensive update to the CUDA platform since it was invented two decades ago." Alongside new features and performance improvements, the arrival of NVIDIA CUDA Tile is set to be a game-changer for AI programming.

The initial release is limited to the current Blackwell generation of GPU hardware (future versions will support more architectures), with CUDA Tile programming allowing users to bring their code up a layer with specific chunks of data called tiles. From there, the compiler and runtime determine "the best way to launch that work onto individual threads," including using hardware such as tensor cores.

With the new CUDA Tile programming, removing the need to define each thread's "path of execution," it reduces the effort required to write code that performs well across various GPU architectures.

Samsung Electronics' GDDR7 memory dies have won presidential recognition this week, as Samsung's continued technological competitiveness gains industry respect after its huge turnaround earlier this year.

In a new report from the Korea Times, we're hearing that at the recent 2025 Korea Tech Festival in Seoul, which was hosted by the Ministry of Trade, Industry, and Resources, the South Korean government awarded a presidential honor to Samsung's world-first 12nm-class, 40Gbps, 24Gb GDDR7 memory.

This isn't the first time for Samsung either, as it's the 12th time that the memory giant has received the presidential commendation, which is the highest number awarded to a single company. It previously received presidential recognition with its 14nm-class DDR5 memory in 2022, and the 64-layer 3D V-NAND flash back in 2017.

NVIDIA is reportedly cutting its B40 AI GPU production from an estimated 1.5-2M units to 900,000 units as Chinese AI companies are reportedly using RTX 5090 gaming graphics cards, as well as Hopper and locally-sourced AI chips instead.

In a new market report posted by insider @Jukan on X, we're hearing that NVIDIA's new B40 AI GPU, which complies with US export restrictions and is ready for China, is seeing its production cut, with an expected 1.5M to 2M units in 2H 2025, reduced to just 900,000 units.

The reason behind this is that the analysts' last "field trip" throughout Asia suggests that China's AI inference compute is still "largely dependent" on RTX 5090 gaming graphics cards (thanks to 32GB of speedy GDDR7 memory) and "previously improved" Hopper AI GPUs, and "some local chips".

NVIDIA will be the first customer for TSMC on its next-generation A16 process (1.6nm) with production taking off in 2027 at its Taiwan fab, while Apple will reportedly skip A16 and use the A4 (1.4nm) process.

We've been hearing rumors that NVIDIA would be the first customer for TSMC on its new A16 process node in 2026, but it looks like that will now take place in 2027, while other rumors said Apple was "not yet in talks" with TSMC to use its A16, with this new report suggesting the company is going directly to A14.

The new DigiTimes report also confirms that TSMC's plans to build 3 more 2nm semiconductor fabs in Taiwan in order to keep up with the ever-growing demand, with industry insiders saying TSMC's 2026 capex could reach $48-$50 billion, up from $40-42 billion this year. TSMC's advanced capacity at its Arizona fabs will also be going through some changes:

Microsoft has proclaimed on multiple occasions that Windows 11 and Windows in general are transforming into an 'Agentic OS,' and the latest 'Experimental Agentic Features' included in a recent Windows 11 preview build offer a first honest look at a Windows PC becoming an AI PC. The quick summary is that AI Agents will have their own accounts and privileges and run in the background while you're using your PC, leading to a situation where multiple users are logged in to your PC, with you being the only human.

Basically, you'll be able to interact with your PC using natural language. At the same time, these AI Agents will handle everything from launching office apps and creating charts to browsing, finding a deal, buying a new appliance, and searching through images to find something specific. These agents will run in the background, with Copilot as the primary interface.

Microsoft notes that you'll be able to monitor AI Agents like you can apps, while also confirming that these agents are prone to hallucinating and can even be tricked into installing malware or sending sensitive data and files to bad actors, which makes you wonder why anyone would enable these 'Experimental Agentic Features' when Microsoft is adamant that they pose a real security risk.

AMD is reportedly working on a couple of new Radeon AI PRO R9000 "RDNA 4" series consumer-grade AI cards, with a tease of the R9700S and R9600D leaking out.

AMD has officially listed its new consumer-grade Radeon AI PRO R9700S and Radeon AI PRO R9600D cards on its own support page, spotted by @KOMACHI_ENSAKA and @RubyRapids on X. We don't have any specifications on the new cards, and nothing official was posted on AMD's support page apart from the names of the two new Radeon AI PRO R9000 series "RDNA 4" cards.

Currently, AMD has its Radeon AI PRO R9700 workstation GPU on the market for $1299, featuring the full Navi 48 GPU core and up to 32GB of VRAM for the desktop. On the laptop side however, RDNA 4 hasn't graced any gaming laptops so far... so the new Radeon AI PRO R9700S could be a flagship RDNA 4-based offering for laptops, with the same Navi 48 GPU + 32GB of GDDR6 memory as the desktop R9700 variant.

The path towards combining both GPUs and HBM memory chips together is here, with NVIDIA and Meta "reviewing plans" to mount GPU cores on HBM.

The general concept is that placing GPU cores into the base die at the bottom of the HBM stack, with Meta and NVIDIA working with HBM leaders SK hynix and Samsung. In a new report from Korean outlet SEDaily and multiple industry insiders "familiar with the matter" that "next-generation 'custom HBM' architectures are being discussed, and among them, a structure that directly integrates GPU cores into the HBM base die is being pursued".

HBM (High Bandwidth Memory) is widely used on AI GPUs from both NVIDIA and AMD, with next-gen HBM4 right around the corner, and HBM4E not far behind it, and it's built perfectly for AI applications and mass amounts of data through its incredibly high memory bandwidth.