Nvidia moves from chipmaker toward AI lab with $26 billion bet on “open‑weight” models

The announcement and strategy

Wired has reported that Nvidia will invest $26 billion over the next five years to develop so‑called open‑weight AI models and the infrastructure around them. It has been reported that the funding will cover model development, compute and datacenter expansion, research hiring and ecosystem building, with initial outlays rolled out over the next 18–24 months and first models expected by late 2026 or early 2027. The company is reportedly already pretraining very large models — including one with roughly 550 billion parameters — and frames the work as both a test of compute and a roadmap for next‑generation hardware.

Why open‑weight matters

“Open‑weight” here means publishing model parameters rather than keeping models strictly closed‑service; it is not necessarily the same as releasing models under a permissive open‑source licence. Nvidia argues that publishing weights tuned for its accelerators would help developers run and fine‑tune models on prem and further entrench the company’s hardware ecosystem. The move sits between OpenAI’s closed approach and Meta’s historically more permissive releases. It also addresses a market demand: enterprises want transparent, customizable models that they can run on their own stacks, and Nvidia’s control of specialized chips gives it an advantage in tightly co‑optimizing software and silicon.

Market and geopolitical context

The plan could reshape who sets standards for large‑scale AI — but geopolitics will complicate the picture. US export controls and trade frictions have already restricted access to top‑end accelerators in China and other jurisdictions; will open‑weight models trained on Nvidia hardware be usable where that hardware can’t be fully deployed? Chinese actors such as DeepSeek and Alibaba (阿里巴巴) have used open‑source strategies to build broad developer followings, and analysts reportedly say a 10% share of the foundational‑model market could add tens of billions of dollars in annual revenue for Nvidia. Ultimately this is a bet on vertical integration: more than chips, Nvidia aims to own a slice of the AI stack — but competition, regulation and international supply limits will determine how far that ambition goes.