Can open source AI do without Jensen Huang ?

Can open source AI do without Jensen Huang ?
3 min read

The resignation of Jensen Huang, CEO of Nvidia, represents, according to the author of the Interconnects newsletter, a major setback for the development of an open model ecosystem in the United States. The company, he argues, retains financial interests in supporting these models, but their growth is more of a strategic puzzle than an engineering problem.

Nvidia’s operations are heavily dependent on Jensen Huang, which, the author continues, leaves several challenges ahead. He also mentions the possibility of further departures linked to investigations into chip smuggling, or to talent recruitment by other players. The company’s priority would be to stabilize the situation and deliver to its customers over the next 24 months. Open models were one of Jensen Huang’s most ambitious bets, and the funding he committed risks drying up, according to the author.

A still nascent economy

The author considers the open model economy to be in its infancy, and that the coming years will be decisive in establishing use cases, revenue streams, and motivations for continuing to train new generations of models. He sees one positive point: the open model economy has, in his view, correctly identified ‘ownership of intelligence’ and enterprise post-training (adapting a model for specific uses) as a strategic axis. The size of the resulting market remains to be determined.

The negative point lies in the scarcity of computing capacity. Young inference companies (running models to answer queries) struggle to obtain enough chips, as Anthropic and OpenAI have purchased a large portion of the available capacity. This results, according to the author, in little margin competition for open-weight models. Even if these models are also offered with high margins, as shown by DeepSeek’s release of DeepInfer, they are far from being a commodity. Closed labs, which benefit from economies of scale and integration from model to data center and chips, would then be best positioned.

Regulatory and security risks

The author is concerned about a wave of Democratic funding for candidates wanting to regulate the AI industry in the name of ‘safety’. In his view, many of these officials feel they must act and consider open models dangerous, without always weighing their benefits. His main short-term fear is poorly conceived state-level regulation, followed by national regulation.

He points out that a single high-performing Chinese open-weight model would be enough to trigger a crisis, a situation he describes as having been valid since this summer. No major hack involving a popular Chinese open-weight model has been observed to date, but it would be, in his opinion, a matter of timing rather than possibility.

For the author, the only lasting solution lies in the emergence of a leading player in open models in the United States. He cites Dylan Patel, CEO of SemiAnalysis, who emphasized in a podcast with ChinaTalk that the problem is not just financial: ‘It’s not that simple. It’s not a money problem, it’s a talent problem.’ Nvidia’s Nemotron series of models, according to the author, has ambition and competent people, without yet reaching the ‘mission accomplished’ stage. Thinking Machines, with its Inkling 2 Small model, is considered the best Western open-weight model for some time, but remains far behind Kimi K3.1 Pro, while expectations are high for the next update of Zhipu’s GLM.

The author concludes that the community must engage in collective reflection and build alliances to create a new platform, the question of the future of open models remaining, in his eyes, entirely open.

S’abonner
Notification pour
guest

0 Commentaires
Le plus ancien
Le plus récent Le plus populaire
0
Nous aimerions avoir votre avis, veuillez laisser un commentaire.x