Jensen Huang, Nvidia chief executive officer (CEO), again rebutted the "software doomsday theory" that has been spreading this year. At Nvidia's annual developer conference, GTC 2026, he called concerns that AI will cannibalize the software industry—the "SaaSpocalypse" (software as a service + apocalypse)—"completely wrong." He added that "AI will not replace software corporations but will boost corporations' development productivity."
Anthropic, an AI corporation dubbed a rival to OpenAI, fanned the "software doomsday theory" when it released the AI agent "Claude Co-Work" in January, stirring worries that it would render existing subscription software useless, and major software corporations' stock prices fell in unison, remaining weak compared with the start of the year. At the time, Huang called this "an illogical claim" and stressed that "as the number of AI agents individuals use for work increases, usage of software tools will also grow exponentially."
Huang's view that the software industry will remain robust is also evident in Nvidia's strategy to expand its AI ecosystem. From the 16th to the 19th (local time) in San Jose, the United States, at GTC 2026, Nvidia announced partnerships with major software and infrastructure corporations. More than 20 corporations—including Siemens, Adobe, Dassault Systèmes, Cadence, Cisco, Salesforce, Inc., Palantir, Red Hat, CrowdStrike, and Atlassian—announced alliances with Nvidia, joining its ecosystem.
"A new industrial revolution has begun in which physical AI and autonomous AI agents fundamentally reinvent how we design, engineer, and manufacture," Huang said. "Nvidia will unite a global ecosystem of major software corporations, cloud providers, and original equipment manufacturers (OEMs) to deliver full-stack accelerated computing that supports unprecedented scale and speed."
At the center of this strategy is Nvidia's "AI factory" initiative. Nvidia is no longer just a company that sells graphics processing units (GPUs) but aims to reinvent itself as an AI platform corporation that builds and sells "AI factories," AI-specialized data centers—an outline it presented again at GTC 2026 after first unveiling it last year.
An AI factory, as Nvidia defines it, is a plant that mass-produces intelligence. It is a specialized computing infrastructure that integrates and manages the AI lifecycle from data collection to training, curation, and inference. Just as traditional factories input raw materials to produce finished goods, AI factories mass-produce intelligence based on data. Here, intelligence is not mere information but deliverables corporations can use directly for decision-making, such as "predicting customer churn," "where and how much to place ads," "loan approval decisions," "predicting equipment failure timing," and "discovering drug candidates."
Huang's goal is to provide corporations with AI factory systems that continuously produce intelligence. To that end, Nvidia has expanded its product lineup from AI chips to networks, storage, and AI models to build a full-stack platform. Its software platform CUDA and the recently unveiled open AI model specialized for running AI agents, "Nemotron-3 Super," are representative examples.
Software is cited as a core element that makes up an AI factory. Huang said an AI factory consists of five layers (five layer cake): energy (power), AI chips (GPUs and CPUs), infrastructure (data centers, networks, and storage), AI models, and applications. Nvidia is expanding collaborations with major software corporations to enhance the performance of each layer.
In practice, the French software corporation Dassault Systèmes has integrated its virtual twin technology (a twin model created in a virtual space that is identical to reality) with Nvidia's GPU-based accelerated computing, and this will be used for simulation, operation, and design of AI factories to be built in the future. Red Hat co-developed with Nvidia an AI platform optimized for corporate environments pursuing large-scale AI adoption, "Red Hat AI Factory with Nvidia," and CrowdStrike agreed to embed its Falcon security solution into Nvidia's AI agent architecture.
Huang said that if competition among AI models centered on large language models (LLMs) drove the initial AI boom, going forward, the outcome will hinge on how well corporations implement AI-based industrial systems that combine various AI models, AI agents, and software. In the end, even in the AI era, software will not disappear but will see increased use as it integrates with AI infrastructure.