Every published Top AI Stories item tagged with AMD, newest first.
AMD released version 10.0 of ROCm, its open compute stack, jumping from 7.14 to clear up a scheme in which the stable release carried a lower number than the tech previews. The headline addition is a layer AMD calls ROCm.AI, shipping as three pieces: a unified command-line tool for installing, serving and diagnosing workloads; Hyperloom, an open-source agentic system that automates inference optimization; and AMD Skills, which packages AMD-validated stack knowledge in the Agent Skills format so that coding assistants such as Claude and Cursor answer questions about AMD hardware correctly. The software layer, not the silicon, has long been the argument against buying AMD for AI.
Stability AI raised $76 million in a Series B whose investor list is the story: Universal Music Group, Sony Music Group, Warner Music Group, Electronic Arts, AMD Ventures and Pacific Alliance Ventures. That brings the company to $232 million raised under chief executive Prem Akkaraju, who joined in 2024. The money goes to a creative production suite and a professional services arm, with the labels positioned as co-developers rather than licensees, which is a markedly different posture from the copyright litigation running elsewhere between rights holders and generative media companies.
Cerebras announced its CS-4 rack system, claiming more than 1,000 tokens per second on models above 10 trillion parameters and up to 30 times the speed of production graphics-processor systems. The headline part is what the chip is not: the **WSE-3 Turbo is the same 900,000-core, 5-nanometer wafer as the two-year-old WSE-3**, clocked from 1.4 gigahertz to 2.8 gigahertz rather than re-fabricated. OpenAI is using it for the Ultrafast tier, and AMD is pairing its graphics processors for prefill with Cerebras for token generation. Shipments start this quarter.
Taalas, founded in Toronto in 2023 by former Tenstorrent chief executive Ljubisa Bajic, burns model weights into a mask read-only memory fabric instead of streaming them from high bandwidth memory, which removes the memory bottleneck that dominates inference cost. The company said in February that its HC1 test chip served Meta's 8 billion parameter Llama 3.1 at 16,960 tokens per second, and its next chip targets 20 billion parameters. AMD did not disclose terms and expects to close by the fourth quarter, planning to sell the technology as systems alongside its Instinct accelerators. The catch is that a chip built around one model is hard to reuse when the model changes.
The inference company Wafer published benchmarks on July 31 putting Moonshot's Kimi K3, at 2.8 trillion parameters, at 952 tokens per second per node on AMD's MI355X against 1,568 on Nvidia's B300. The AMD part is slower in raw throughput but roughly 2.4-times cheaper per GPU, which works out to 48 tokens per second per dollar versus 33. Its 288 gigabytes of memory per GPU also lets the model fit in a single node where the Nvidia configuration needs two.
AMD used its Advancing AI event to launch Helios, its first full rack-scale system aimed squarely at Nvidia's grip on AI data centers. Each rack packs 72 of AMD's new Instinct MI455X GPUs with sixth-generation EPYC processors and Pensando networking, delivering up to 2.9 exaflops of inference performance. Microsoft will run Helios on Azure and Anthropic plans to install up to two gigawatts of the chips, with OpenAI and Meta also signed on. Systems ship in the second half of the year.