Jason Calacanis says Nvidia is “taking the gloves off” with Nemotron and aims to control more of the AI stack

Nvidia has long been viewed as the critical supplier for the modern AI era, powering training and inference with its accelerated chips. But venture capitalist Jason Calacanis argued the company is moving beyond being the indispensable hardware vendor and is preparing to take on a larger role in artificial intelligence, including the software models themselves.

Calacanis said Nvidia is pairing its “dominant AI hardware business” with increasingly capable open-weight large language models through its Nemotron effort. Open-weight models are AI language systems whose weights are available to others, allowing wider customization and deployment compared with fully closed models. In his view, that combination could allow Nvidia to “own the whole stack,” meaning not just chips but also the model layer that many companies use to build AI products.

On the All-In Podcast, Calacanis also predicted that Nvidia’s open-weight model offerings could become good enough for many everyday use cases. He suggested users would not be able to reliably distinguish between an open model from Nvidia and leading alternatives from companies such as Anthropic for a large share of typical search and assistant tasks.

Calacanis further framed the company’s potential endgame as a challenge to OpenAI and Anthropic, two firms that have used proprietary models as major differentiators. His comments were speculative, but they align with a broader industry question that has been growing louder: whether AI progress will be dominated by a small number of closed foundation-model providers or by a more open ecosystem where model distribution and infrastructure determine leverage.

What is known from the public comments is centered on strategy and positioning, not on disclosed product timelines or performance benchmarks. The available reporting and quotes do not provide technical measurements for Nemotron, details on model releases, or any formal commitments by Nvidia about future model capabilities or adoption targets.

Sector context may help explain why the comments resonated. Nvidia’s core business has been tightly linked to demand for compute, but the economics of AI can shift as model development becomes increasingly important to end-user experiences. If a chip supplier can meaningfully influence the model layer, it may capture more value across inference deployments and tooling, rather than relying solely on hardware refresh cycles.

Still, important specifics are missing from what has been reported so far. Neither the podcast remarks nor the business write-ups reviewed here include confirmation from Nvidia about the exact lineup of Nemotron models, how Nvidia intends to commercialize them, or whether Nvidia plans to compete directly with frontier labs through model releases, partnerships, or licensing. For now, observers will likely treat Calacanis’s “own the whole stack” framing as a hypothesis rather than a company statement.

Going forward, investors and AI developers will be watching for concrete indicates from Nvidia that bridge the gap between strategy talk and operational details, such as model releases, benchmarks, platform integration, or clearer go-to-market partnerships around open-weight systems. Any formal disclosure from Nvidia, whether in product announcements, developer documentation, or investor communications, would help determine whether Nemotron is primarily a technology test bed or the foundation for a broader competitive push.