GPU

Elon Musk said xAI plans to expand its computing capacity to more than 1.2 million Nvidia GPUs by the end of the year, more than double its current count. The plan underscores the escalating race among AI companies to secure massive GPU fleets for training larger models, and highlights Nvidia's central role in supplying the global AI buildout.

China has reportedly relaxed restrictions on NVIDIA GPUs as part of a shift in its artificial intelligence strategy, according to a report in South Korean media. The move would allow Chinese firms greater access to NVIDIA's AI chips, which have been subject to US export controls and earlier Chinese procurement curbs. Details on the scope and timing of the relaxation remain limited.

China is reportedly considering allowing imports of NVIDIA GPUs as it seeks to accelerate its artificial intelligence development. The report, carried by South Korean outlet Chosun Ilbo, suggests a possible shift in Beijing's approach to securing advanced chips for its AI sector. Further details on the scope or timing of any decision have not been confirmed.

Elon Musk says SpaceX's AI effort, dubbed SpaceXAI, could reach GPT-6-level capability within three months and take the 'pole position' in artificial intelligence within six, backed by plans for a massive expansion of GPU computing power. The comments land amid fierce competition among AI labs and Musk's own xAI venture, drawing skepticism and attention in equal measure.

Nvidia has brought LG Electronics into its Partner Network as a preferred cooling provider, aiming to ease the power and heat bottleneck facing large-scale AI data centers. The move highlights that thermal management, not just GPU supply, is becoming a decisive factor in AI infrastructure, giving LG a direct route to supply cooling technology for Nvidia systems.

China is reportedly pursuing a 'hiding capabilities' strategy in artificial intelligence, built around NVIDIA GPUs. The approach suggests Chinese AI developers may be concealing the true extent of their computing power and model performance, likely to manage perceptions abroad and avoid triggering tighter export controls. The claim is circulating in international coverage of the US-China semiconductor rivalry.

Supersonic Labs has released Julia 1, an open decision model with 144.3 million parameters that is small enough to run on a standard CPU. Unlike large language models, decision models are built for making choices and taking actions rather than generating text. The low hardware requirement makes the model accessible to developers without expensive GPU infrastructure, which is drawing attention in the AI community.

CoreWeave, the GPU-focused cloud provider, is shifting from capitalising on shortages of AI chips to proving it can build a lasting, sustainable cloud business. Commentary centres on whether the company can convert its early advantage in scarce Nvidia GPUs into durable revenue as competition intensifies and chip supply improves. Analysts are watching its contracts, margins and ability to diversify beyond raw computing power.

Cerebras Systems will supply its CS-4 chips to support Gimlet's AI inference cloud infrastructure. The deal places the wafer-scale computing specialist's hardware at the core of a dedicated cloud service for running AI models, underscoring growing competition with GPU-based providers in the inference market.

A new report examines the thriving black market for advanced GPUs, arguing that US export controls aimed at keeping cutting-edge chips out of rival states have failed to stop smuggling and resale networks. Despite tightening sanctions, restricted hardware continues to change hands through intermediaries and gray-market dealers, and critics say enforcement is losing the race against demand for AI computing power.

Amazon Web Services has increased prices for reserved GPU capacity by 15%, citing surging demand for AI computing. The move affects customers locking in long-term access to graphics processors used for training and running machine learning models. Industry observers are debating what it signals about the cost of AI infrastructure and cloud providers' pricing power as demand for compute keeps outpacing supply.

A new technical writeup compares three approaches to rendering text on the GPU: signed distance fields (SDF), multi-channel signed distance fields (MSDF), and the Slug algorithm. The piece examines the trade-offs between them in quality, memory use and performance, drawing interest from graphics programmers weighing which method best suits real-time applications like games and UI engines.

Thieves in an unspecified location made off with a 20-ton shipment they believed contained valuable Nvidia graphics cards, only to discover the cargo was actually sand. The heist has become a viral talking point, with observers mocking the criminals' costly mix-up and questioning how the mistake happened. No word yet on arrests or the real value of what was targeted.

Developers are increasingly running AI models on Apple's Mac Mini instead of relying on Nvidia GPUs, according to a Fortune report. The shift is being attributed to the Mac Mini's lower cost and power efficiency, with Apple silicon offering competitive performance for local AI workloads. The trend highlights a challenge to Nvidia's dominance in AI hardware as smaller teams look for cheaper ways to build and test AI applications.

A detailed technical analysis of Qualcomm's Adreno X2 GPU, the graphics processor in its Snapdragon X2 PC chips, has been published by the hardware analysis site Chips and Cheese. The piece examines the architecture in depth, and the topic is drawing attention among chip enthusiasts and developers following Qualcomm's latest Windows-on-ARM push.

Chipmaker Cerebras Systems has agreed to deliver AI systems to Gimlet Labs, a cloud computing startup. Under the arrangement, Gimlet Labs will use Cerebras hardware to power its AI infrastructure offerings. The deal underscores growing demand for specialized AI computing among startups, as companies seek alternatives to GPU-based providers. Financial terms and delivery timelines have not been disclosed, and Cerebras has not publicly detailed the scope of the partnership.

The MLC team has published TIRx Harness, an open-source compiler harness designed for agentic GPU programming, letting AI agents write and optimize GPU kernels through a compiler-driven workflow. The announcement is drawing attention from developers interested in LLM-driven code generation and performance engineering, with early discussion focusing on how compiler infrastructure can make agent-written GPU code more reliable and verifiable.

ASUS's 16-inch TUF Gaming A16 laptop, equipped with a Ryzen 9 processor and an RTX 5060 GPU, is being offered at a 10% discount. The machine features a 165Hz display aimed at smooth gaming performance, and Japanese gaming media are highlighting the deal for players seeking a capable mid-range portable gaming PC.

Traverse Research has released Cossmology, a set of open-source GPU tools and benchmarks, drawing attention from developers including Jasper Bekkers and Max de Danschutter. The project is being shared in gaming and open-source software communities, where observers are highlighting its potential value for GPU performance testing and development workflows.

Free software news roundup of 27 September 2026 highlights an improved GPU reset mechanism coming to Raspberry Pi boards with Linux kernel 7.4, as reported by Phoronix. The change should make it easier to recover the graphics processor when it hangs, without a full reboot, improving stability for Pi users running the mainline kernel.

Valve engineers are working on foveated rendering support for the open-source Turnip graphics driver in connection with the company's Steam Frame headset. Foveated rendering cuts GPU load by drawing only the area the user is looking at in full detail. The work signals Valve's continued investment in open-source graphics software for its VR hardware.

AWS has introduced the P6-B300, an instance built around Nvidia's Blackwell B300 GPUs for large-scale AI work. A QuantumBytz article argues the instance changes the real buying question for AI infrastructure: rather than raw GPU count, organizations must weigh how clusters balance compute, networking and cost. The piece is being shared in Linux and open-source tech circles, where cloud GPU pricing and cluster design are active topics among engineers planning AI deployments.

A new open-source project called Nesbox, published by Nestrilabs on GitHub, offers a fast micro virtual machine with support for sharing GPUs across workloads. It is drawing attention among developers interested in lightweight virtualization and more efficient use of accelerated hardware, a growing concern as AI and compute costs rise.