This brief covers AI news from 2026-09-08 UTC.
Today’s biggest news: Nvidia confirms its $12.9B acquisition of Hugging Face, plus major model and tool releases.
Nvidia confirms $12.9B Hugging Face acquisition
What happened: Nvidia announced plans to acquire Hugging Face for $12.9 billion, subject to regulatory approval. The deal includes $11.9 billion in cash and up to $1 billion in equity retention. Nvidia CEO Jensen Huang said Hugging Face will remain an open platform for the entire AI ecosystem.
Why it matters: Developers who rely on Hugging Face for model distribution may see changes in governance and infrastructure, though Nvidia pledges openness.
IFM releases six open models, from 0.9B to 375B
What happened: The Institute of Foundation Models released K2 Horizon, a fleet of six Apache 2.0 models ranging from 0.9B to 375B parameters, along with training code, configs, and logs. All models are on Hugging Face with FP8 and GGUF builds, and support vLLM, SGLang, and Ollama.
Why it matters: Teams can prototype on smaller models and scale to 375B without changing serving stacks, all under a permissive license.
SpaceXAI launches Grok Build coding agent
What happened: SpaceXAI launched Grok Build, a terminal-based coding agent powered by Grok 4.6. It can read and write files, execute commands, and fix bugs. Available on Windows, macOS, and Linux, with access tiered by Grok subscription.
Why it matters: Developers get a new AI coding agent that integrates directly into their terminal workflow, potentially speeding up routine coding tasks.
iFlytek launches Spark X2.5 with edge models
What happened: iFlytek released Spark X2.5, a 293B-parameter MoE model trained on domestic Chinese infrastructure. It also released two smaller open-source versions (4B and 1.7B) that natively support 1 million-token context windows for edge devices.
Why it matters: Edge models with million-token context could enable long-document and coding tasks on local devices, reducing cloud dependency.
Anthropic walks away from $6B Decart talks
What happened: Anthropic has pulled out of negotiations to acquire Israeli AI startup Decart in a deal valued at roughly $6 billion. Decart builds chip-optimization software for efficient AI inference.
Why it matters: The collapse may signal a shift in Anthropic’s acquisition strategy, potentially affecting its IPO timeline and compute efficiency plans.
Claude formalizes Fermat’s Last Theorem in Lean
What happened: Anthropic announced that Claude completed a fully machine-verifiable proof of Fermat’s Last Theorem, working almost autonomously for 11 days and generating about 13 million lines of Lean 4 code. It is the first complete formalized proof of the theorem.
Why it matters: This demonstrates AI’s ability to handle long-horizon, rigorous reasoning tasks, which could extend to verifying complex software systems.
Arm unveils AGI CPU and Neoverse CSS N4
What happened: Arm announced the Neoverse Compute Subsystems N4 and provided updates on its AGI CPU, designed for agentic AI workloads. Arm says its rack-scale servers have overtaken x86 as the dominant accelerated computing platform.
Why it matters: New CPU options tailored for agentic AI could give developers more choice in building responsive, efficient AI infrastructure.