AI accelerators news
RSSIllegal Nvidia Blackwell Ultra B300 GPUs found inside state-owned China Telecom's network
A security firm has been able to identify 32 Nvidia Blackwell Ultra B300 AI chips running inside a China Telecom network, but it's unclear where the export-barred components are physically located.
FCC issues first bans under new chip-level rule
FCC bans a dashcam and wireless video transmitter produced by HiSilicon, a subsidiary of Huawei for containing an AI chip that advertises face detection algorithms.
Deepx Building Open Software Ecosystem to Scale Physical AI Beyond the Data Center
Deepx is combining low-power AI accelerators and portable software with industry partnerships to bring physical AI into production. The post Deepx Building Open Software Ecosystem to Scale Physical AI Beyond the Data Center appeared first on EE Times.
HAWK: Heterogeneity-Aware KV Cache Transfer for Efficient GPU-NPU Phase-Disaggregated LLM Inference
Large language model (LLM) inference exhibits distinct computational characteristics across its two phases: prefill is compute-intensive, whereas decode is more memory-bound. As a result, AI accelerators achieve varying levels of efficiency across the two phases. GPUs are well suited to prefill, but are often…
(PR) Synaptics Announces Astra SR200 Multimodal Edge AI Processors with Google Coral NPU
Synaptics Incorporated (Nasdaq: SYNA) today introduced the xMCU, a new class of microcontrollers that runs multimodal Edge AI, with the launch of the Synaptics Astra SR200 Series. Unlike conventional MCUs constrained by limited on-chip memory and compute, the xMCU combines heterogeneous, AI-native compute and…
MX-Like Tensor Core for LLM Inference with NPU Integration via Hardware-Software Co-Design Exploration
Block-scaled quantization formats such as OCP Microscaling (MX) and NVFP4 have been widely adopted as efficient data formats for LLM inference. However, the design space of these formats spanned by shared scale, weight, and activation format combinations has not been systematically explored, and the joint…
SK hynix to place custom tiny CPU cores in its SSD controllers to ease Arm load as drive speeds rocket
SK hynix detailed H-TPU, tiny in-house cores that offload routine I/O from the Arm CPUs in its next enterprise SSD controllers.
A New GPU, From Facebook ?!
Meta builds its own AI accelerators, and at Hot Chips 2026 it presents the next two: the MTIA 300 for training and the MTIA 400 for generative AI and recommendation. Recommendation compute at Meta grows 3x a year, and GPUs sit idle waiting on memory for those workloads. I also show the first two MTIA generations on…
Former Groq engineers sue board over $20 billion Nvidia deal
Two former Groq engineers sued Groq’s board in Delaware over the Nvidia licensing deal behind the Groq 3 LPU.
Why Everyone's Building AI Chips in Korea // The AI Hardware Podcast S2E12
South Korea's AI chip scene is booming - DEEPX, HyperAccel, Mobilint, FuriosaAI, and Rebellions are all racing to carve out a niche between the edge and the data center. Ian Cutress and Sally Ward-Foxton give their honest takes on which companies have the traction, funding, and roadmap to actually make it - including…
Inside Korea's AI Silicon // The AI Hardware Show S2E12
South Korea has quietly become one of the most active fronts in the AI chip race, and Sally Ward-Foxton and Ian Cutress are on the ground to find out why. In this episode of The AI Hardware Show, they break down five of the country's boldest AI silicon players — DeepX, HyperAccel, Mobilint, Furiosa AI, and Rebellions…
[54] The Man Who Built OpenAI's First Chip
Today's sponsor is https://www.amtechmicro.com Richard Ho leads hardware at OpenAI. He joined Google's TPU program as one of its first engineers, built Anton supercomputers at D. E. Shaw before that, and now runs the team behind Jalapeño. He tells me how close he came to a blank cheque, why the team bet on wiring HBM…
The Microwatt Chip Race // The AI Hardware Podcast S2E11
Forget the data center - the real AI action might be happening in microwatts. Sally Ward-Foxton and Ian Cutress dig into the tiniest AI chips on the market, from analog compute upstarts to ARM's revived Ethos NPU line, and debate why an "AI ASIC" isn't really an ASIC at all. Plus: a mysterious defense-linked startup…
How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip
On 25 August, OpenAI fully unveiled Jalapeño, the company’s debut AI accelerator chip. Jalapeño delivers up to 13.4 petaflops of 4-bit compute and accesses 232 gigabytes of the most advanced memory available, linking to it at a blazing 15.4 terabytes per second. Benchmarks cited by OpenAI show that Jalapeño can reduce…
Broadcom Rides Rocketing Trend For Custom AI Accelerators
The ABF substrate data center roadmap — solving the supply crunch and material wall beneath every AI accelerator
ABF substrates underpin today’s most advanced AI chips, but soaring demand and expanding accelerator packages are creating new supply and technical bottlenecks that the industry is currently racing to resolve
Samsung and Mistral AI Announce Strategic Partnership for Intelligence-Driven Semiconductor Infrastructure
Samsung Electronics today announced it has entered into a strategic partnership with Mistral AI to enhance its semiconductor engineering and manufacturing capabilities that will reinforce its leadership in AI chip technologies. The announcement of the partnership was made during the state summit held in Paris between…
TPU Inference Externalization Full Steam Ahead - InferenceX
InferenceX, Up to 50% Better Performance per Dollar, Rapid Externalization of TPU stack, Growing Customer Base, Ironwood, TPUv8i, Reducing CUDA Moat
AI Chips Drive Around A Third Of TSMC Revenues
Stacking Chips Sideways Gives AI More Memory
GPUs and other big AI chips just can’t get enough memory. Today’s data center computers stack DRAM 12 dies high. But as memory makers try to build higher stacks to pack in more bits and bandwidth, experts worry this high bandwidth-memory (HBM) will trap enough heat to cook itself into oblivion. That will be especially…
Taiwan raids 12 locations in its first formal crackdown on Nvidia AI chip smuggling
Getting banned Hopper or Blackwell chips into mainland data centers just became exponentially more fraught as Taiwan begins to crack down on smugglers.
IBM's Spyre AI Accelerator Deep Dive
Enterprise AI Inference, Eight Years in the Making
Jensen Groks Groq
On Nvidia's fascinating 'acqui-something' of AI chip startup Groq
TPU Mania
Is GPU vs TPU the CISC vs RISC contest of the modern era?