Nvidia has introduced Nemotron 3.5 Lightning, a new open artificial intelligence model designed for businesses aiming to reduce token processing costs while maintaining access to advanced features. The announcement was made on Tuesday (the 11th) amid ongoing competition between closed and open AI models.
Kari Briski, Vice President of Generative AI Software for Enterprises at Nvidia, explained the company's strategy while speaking at The Wall Street Journal Leadership Institute. In addition to the new AI model, Nvidia released Nemo Switchyard—an open-source library that allows directing various AI agent tasks to different models.
The main goal of this development is to achieve a balance between speed, quality, and cost, as well as minimizing the risk of confidential information leakage. Nemotron 3.5 Lightning is designed with a more compact architecture, yet it retains the knowledge inherent in more powerful models, offering companies a less expensive alternative for specific tasks.
According to Briski, the economic viability lies in avoiding the use of larger models for routine operations. Switchyard complements this concept by deciding which model should perform each activity based on parameters such as price, processing time, and response quality.
Initially, the routing system was developed by Nvidia itself to meet internal company needs: efficiently managing the volume of consumed tokens in AI applications and maintaining control over its own intellectual property. Furthermore, she emphasized the protection of corporate information. For organizations operating in highly specialized fields, such as cybersecurity and materials science, there is a concern about sharing strategic knowledge with AI providers who might become competitors in these markets.
Briski noted that open models give companies greater flexibility, including the ability to train on their own data. For Nvidia, this combines with the ability to measure gains in both efficiency and accuracy.
This move came just one day after Meta Platforms introduced Muse Glimmer, another open model. The company described the tool as small enough to run on a Mac or PC equipped with only one consumer graphics card.
Nvidia is also actively involved in expanding AI infrastructure. The company has entered into agreements with Apollo, Blackstone, Goldman Sachs, and KKR to create computing power financing platforms, aiming to provide clients with over $500 billion in capital. The industry landscape also features Anthropic, which is seeking investors ahead of a potential $965 billion IPO, as well as a 20-year, $9.1 billion contract between the company and Riot Platforms to secure data center capacity in Texas.
In the semiconductor sector, Intel announced plans to raise $20 billion through a public offering of 210.5 million shares at $95 per share. Previously, the company had stated high customer demand related to the growth of investments in artificial intelligence computing power.



