AI Spending Ripples Across Tech Stack; Nvidia Acquires Hugging Face | Bloomberg Tech 9/03/2026
THE SO WHAT
Nvidia moving to acquire Hugging Face for $13B would pull one of the largest open model and tooling communities directly into the GPU vendor’s orbit—blurring the line between neutral ecosystem and vertically integrated stack. If you’ve standardized on Hugging Face, revisit your infra assumptions and concentration risk this quarter.
READ THE SOURCE
MORE FROM THE WIRE
Applied AICrusoe Signs Roughly $13 Billion AI Cloud Deal With Jane Street
A ~$13B AI cloud contract with Jane Street shows quantitative finance is locking in dedicated AI compute the way hyperscalers lock in power — long-term, at scale. If you’re in capital markets, assume model-driven trading and research will be gated by your ability to secure high-performance clusters, not just talent.
I'm a VC with 5 kids. I encourage them to use AI and not to pick the safest career.
A VC paying for her Gen Z kids’ AI subscriptions is a small but telling data point — early-career workers are being nudged to treat AI fluency as baseline, not bonus. Employers hiring into junior roles should assume candidates will arrive with tool expectations and design workflows that harness, not suppress, that behavior.
Applied AIAbliteration.ai is making a business out of removing AI guardrails
Commercializing unguardrailed access reframes “jailbreaks” as a product, not a side effect — and it will force enterprises and regulators to distinguish between model capability and deployment policy. If you run security or compliance, assume your adversaries — and some internal teams — can access fully uncensored models and plan controls accordingly.
Applied AIOpenAI says Astra was built on its largest-ever training run, using more than 100,000 GPUs at its Stargate site in Texas
Six-figure GPU counts for a single training run push frontier models into the realm of industrial infrastructure planning. If you’re not in the top tier of capital allocators, assume you’ll be building on — not competing with — these stacks and optimize for orchestration, data, and domain depth instead of raw model scale.