AMD just teamed up with TensorStack to release Amuse 2.0 Beta, a local AI suite that runs Stable Diffusion and ControlNet workflows without any command-line setup. The biggest perk is XDNA Super Resolution, which speeds up creation and doubles output resolution on supported Ryzen AI processors and Radeon RX 7000 GPUs. It gives PC users a seamless on-device sandbox that bypasses the cloud entirely.
NVIDIA just broke down how its AI Foundry service helps companies build custom generative AI models using proprietary data. Powered by DGX Cloud, the NeMo platform, and support for Llama 3.1, it lets businesses forge and deploy tailored models as NIM microservices without building infrastructure from scratch.
The Verge's Decoder podcast just dug into the uncomfortable reality of AI's massive power appetite. While tech giants like Microsoft and Google blow past their emissions targets to pack data centers with GPUs, high energy use is already standard across video streaming and gaming. The real debate is how we decide if AI is actually worth the juice.
Elon Musk just fired up xAI's Memphis Supercluster, packing 100,000 liquid-cooled Nvidia H100 GPUs tied together on a single RDMA fabric. Musk boasts it is the world's most powerful AI training cluster and aims to deliver the strongest AI on every metric by December, though local utility contracts for its massive 100-megawatt power appetite are still not finalized.
Hugging Face just brought Google Cloud TPU v5e chips to its Inference Endpoints and Spaces, starting at $1.38 an hour. Powered by the open-source Optimum TPU library, developers can deploy and serve open models like Gemma, Llama, and Mistral in a few clicks. It gives builders a viable, budget-friendly alternative to scarce Nvidia GPUs.
Sequoia Capital partner David Cahn just calculated that the AI industry needs to generate $600 billion per year to pay off its massive hardware and data center investments. Even with aggressive revenue assumptions for tech giants, there remains a staggering $500 billion gap. Why it matters: Big tech is pouring fortunes into GPUs, but actual software monetization is falling far behind.
AI startup Etched just raised $120M to build Sohu, a specialized ASIC hardwired solely for Transformer models. By stripping away general-purpose GPU flexibility, the company claims it outperforms Nvidia's H100 by 20x. Why it matters: It could drastically lower LLM serving costs, but the chip hasn't taped out yet and is useless for any other AI architecture.
Elon Musk just revealed that Dell and Super Micro will supply the server racks for xAI's massive new supercomputer. Dell is assembling half the racks alongside Nvidia to build an "AI factory" aimed at powering Grok 3, which Musk says will need a staggering 100,000 Nvidia H100 chips.
Nvidia just dethroned Microsoft to become the world's most valuable public company, reaching a $3.34 trillion market cap after shares rose 3.6% on Tuesday. The chipmaker commands around 80% of data center AI processors, pushing its stock up over 170% this year alone. It confirms that the AI boom's biggest winner remains the hardware engine powering it all.
Microsoft CEO Satya Nadella just addressed the G7 Summit, revealing the company poured $5 billion into cloud and AI infrastructure across Indonesia, Malaysia, Thailand, and Kenya over the past three months. Teaming up with UAE-based G42, Microsoft is building trusted data zones and driving sustainable data centers to bend the productivity curve in emerging markets.
OpenAI is teaming up with Microsoft and Oracle to run Azure AI workloads on Oracle's cloud infrastructure. The deal shows OpenAI needs more capacity than Microsoft alone can provide to keep ChatGPT running smoothly. Frontier model pre-training stays on Microsoft supercomputers, but the partnership also puts Oracle in the awkward position of supplying both OpenAI and rival xAI.
Intel, Google, Microsoft, Meta, and AMD just formed the UALink Promoter Group to establish an open standard for interconnecting AI accelerator chips in data centers. UALink 1.0 will connect up to 1,024 GPUs in a computing pod. Notably missing are Nvidia and AWS, as tech giants push to break Nvidia's grip on AI hardware.
xAI just closed a massive $6 billion Series B round backed by Sequoia Capital, Andreessen Horowitz, and others. The company plans to use the capital to launch its first commercial products, build advanced infrastructure, and accelerate future R&D. Following its Grok-1.5 and Grok-1.5V releases, xAI now has the war chest to stay in the frontier compute race.
Apple just unveiled its next-generation M4 chip, debuting inside the redesigned iPad Pro. Built on second-generation 3nm technology, the processor features Apple's fastest Neural Engine yet, claiming to outpace the NPU of any AI PC today. Along with hardware-accelerated ray tracing and dynamic caching, Apple is positioning the tablet as an on-device AI powerhouse.
Google just announced a $3 billion investment to build and expand data centers across Virginia and Indiana to power its AI workload. Alongside the infrastructure boost, Google.org is launching a $75 million fund and the Google AI Essentials course to train one million Americans in basic AI skills for free.
Microsoft just hired former Meta VP of infrastructure Jason Taylor as corporate VP and deputy CTO to build its next wave of AI supercomputing systems. Taylor previously spent over a decade managing Meta's AI data infrastructure and hardware budgets. Why it matters: As frontier models demand massive compute, Microsoft is securing veteran talent to scale its foundational data center architecture.
NVIDIA just rolled out full-stack optimizations for Meta's brand-new Llama 3 across its data center GPUs, edge devices, and PCs. In early benchmarks, a single H200 GPU churned out roughly 3,000 tokens per second on the 70B model, enough to handle 300 users simultaneously. The 8B model is also landing on RTX PCs and Jetson Orin modules.
Intel just unveiled Hala Point, the world’s largest neuromorphic computer packing 1.15 billion artificial neurons across 1,152 Loihi 2 chips into a microwave-sized chassis. By storing and processing data simultaneously like a real brain, it solves optimization problems with 100 times less energy than conventional machines. While not built to train LLMs from scratch, it opens a promising path for ultra-efficient continual learning once software catches up.
Google DeepMind CEO Demis Hassabis just revealed that Google will eventually invest more than $100 billion into AI over time, topping the rumored cost of Microsoft and OpenAI's Stargate supercomputer. Hassabis insisted Google's processing power already outstrips rivals, with capital expected to flow into Axion chips and massive data center buildouts.
Intel just unveiled its Gaudi 3 AI accelerator at Vision 2024, claiming it outperforms Nvidia's dominant H100 GPU. Built on TSMC's 5nm process with 128GB of HBM2e memory, Intel says Gaudi 3 runs models like LLaMA up to 1.7x faster and is 2.3x more power-efficient. Air-cooled models roll out in Q3, followed by liquid-cooled units in Q4.
