Nvidia CEO Jensen Huang Pushes AI Tokens and Aims to Own Full AI Factory Stack

Nvidia used its GTC megaconference to unveil a suite of next‑generation GPUs and a refreshed software stack aimed at consolidating the company’s role as the “AI factory” for enterprises. The hardware announcements included the H100‑X2, a dual‑die version of the H100 that doubles tensor‑core density, and the new Blackwell‑B GPU, which Nvidia says can deliver up to 2.5 × the inference performance of the previous generation while consuming 30 % less power (Reuters). Alongside the silicon, Nvidia introduced a set of updates to its CUDA‑based development tools, the Nvidia AI Enterprise suite, and a tighter integration with its DGX Cloud platform, positioning the entire workflow—from model training to edge deployment—under a single Nvidia‑controlled umbrella (SiliconANGLE). By bundling these components, Huang signaled that Nvidia intends to capture not only the compute layer but also the orchestration, monitoring, and lifecycle‑management services that traditionally belong to cloud providers and system integrators.

The “AI factory” narrative was reinforced with a series of partnership announcements. Nvidia disclosed new OEM agreements with major server manufacturers, including a joint effort with Dell to ship pre‑configured DGX systems that embed the Blackwell‑B GPU and the latest version of Nvidia’s AI‑optimized operating system. In parallel, the company announced a strategic alliance with Microsoft Azure to make the new GPUs available as a native offering in Azure’s AI infrastructure, while also extending Nvidia’s software stack into Azure’s managed services (SiliconANGLE). These moves aim to lock customers into a vertically integrated pipeline where Nvidia supplies the silicon, the software, and the cloud‑hosted execution environment, effectively reducing the need for third‑party middleware.

In a departure from the industry‑wide trend of trimming headcount in response to AI‑driven automation, Huang introduced “AI tokens” as a supplemental compensation mechanism for Nvidia engineers. According to CNBC, the tokens will be awarded based on the contribution of individual employees to internal AI agents that automate routine coding, testing, and deployment tasks. The tokens are convertible into equity‑like units that vest over a three‑year horizon, aligning employee incentives with the broader corporate goal of accelerating AI‑powered productivity gains rather than resorting to layoffs (CNBC). Huang framed the initiative as a way to “reward the people building the AI factory,” arguing that the tokens will motivate staff to embed AI into more aspects of the development workflow and help Nvidia maintain its technology lead.

The financial implications of this strategy are underscored by Nvidia’s expectation that AI inference will become a $1 trillion revenue opportunity over the next decade, a figure cited by Reuters in its coverage of the GTC announcements. By expanding its product portfolio into the inference‑centric segment—where the new Blackwell‑B GPU’s low‑latency performance is most valuable—Nvidia hopes to capture a larger share of the downstream AI market that is currently dominated by cloud providers. The company’s push to own the full stack also positions it to extract higher margins from software licensing and services, complementing the traditionally hardware‑heavy revenue model.

Analysts have noted that Nvidia’s aggressive vertical integration could raise antitrust scrutiny, but the firm appears confident that the combined hardware‑software‑cloud offering will be a decisive differentiator for enterprise customers seeking a turnkey AI solution. As Huang concluded at GTC, the goal is to make the “AI factory” a single, self‑contained entity that eliminates the friction of stitching together disparate components from multiple vendors. If successful, Nvidia would not only solidify its dominance in AI compute but also reshape the economics of AI deployment across the industry.