Jensen Huang's GTC Taipei Keynote: 'AI Is Now a Profit Generator' and Why Software Companies Shouldn't Fear the 'Age of Agents'

The core message for enterprise software leaders was unequivocal. 'A lot of people have said, "Jensen, agentic AI is coming, therefore all of the software companies are going to go out of business,"' he said, addressing the concern directly. His answer was that every software company must transform but won't disappear. 'Every single software company in the world needs an agentic system. Need an agent strategy,' he declared, framing the future of IT not as a wrecking ball for incumbents but as a new platform upon which they must build .

'AI is now a profit generator': The age of agents arrives

Huang's framing of the opportunity was not just technical but economic. 'AI is now a profit generator. AI is now a GDP generator,' he told the Taipei audience, signaling a shift from AI as a research curiosity or productivity toy to a core engine of macroeconomic value . The industrial metaphor he returned to throughout the keynote was the 'AI factory'—massive, specialized data centers purpose-built to produce intelligence as a continuous, token-based output. This reframes Nvidia's hardware not as a one-time capital expense but as the physical plant of a new information-age economy .

'Compute is now directly convertible into revenue,' analysts summarized, noting Huang's message that architectural choices made by IT leaders over the next few years would determine their competitive position for the next decade .

The enterprise agent toolkit: Nvidia's open-source software offensive

The practical centerpiece of this vision was the formal unveiling of the NVIDIA Agent Toolkit, a full-stack, open-source runtime for building, deploying, and securing autonomous agents across the enterprise . The toolkit represents Nvidia's most aggressive push into the software layer of the AI stack, bundling together large language models, an agent harness, and an enterprise-grade security runtime.

Key components of the tool suite include:

• NemoClaw Blueprints: Production-ready enterprise agent configurations that handle built-in security, privacy guardrails, and support for always-on autonomous agents. NemoClaw uses the Agent Toolkit to wrap the open-source OpenClaw agent framework with enterprise safety and management features, installable and optimizable with a single command .
• NVIDIA Nemotron 3 Ultra: A smaller, faster open model specifically designed for long-running agent workloads. Nvidia claims it delivers 5x faster inference and up to 30% lower cost for complex agentic tasks compared to larger general-purpose models .
• OpenShell: An open-source, isolated sandbox runtime that provides open models and a secure execution environment so that agents can 'run safely' on enterprise data and workloads .
• CUDA-X Libraries with Agent Skills: A collection of accelerated libraries that give agents specialized skills across Omniverse, Cosmos, and Metropolis platforms, particularly for physical AI and robotics workflows .
• Physical AI Agent Collection: An open-source library of agent-executable instructions that turn complex robotics, autonomous vehicle, vision AI, and industrial digital twin workflows into repeatable, optimized tasks .

The strategy is visibly open-source and partner-driven. Nvidia is counting on enterprise software leaders to integrate its tooling stack. IBM WatsonX was featured at the keynote as an early integration partner, demonstrating the ability to 'refresh global operations data in minutes' using Nvidia's agent platforms . ServiceNow also committed to its Autonomous Workforce of AI Specialists using the NVIDIA Agent Toolkit and the AI-Q blueprint for agentic search .

From data center to desktop: RTX Spark and the reinvention of the PC

Moving beyond the data center, Huang made a direct play for the client-side market with the unveiling of the RTX Spark, developed in partnership with Microsoft. Huang framed the machine not as a conventional PC but as a new class of personal computer purpose-built for on-device AI agents. 'Microsoft and NVIDIA are going to reinvent the PC,' a keynote presentation declared .

RTX Spark's headline specification is 1 petaflop of AI compute paired with 128 GB of unified memory, a hardware profile designed to run sophisticated, locally-hosted AI agents that Huang described as moving the PC 'from tool to teammate' . The collaboration with Microsoft goes deep; alongside the hardware, the two companies are delivering a native Windows experience for personal agents with new security primitives and the integration of NVIDIA OpenShell . Alongside the consumer-focused Spark, Nvidia also introduced the DGX Station for Windows, a higher-end deskside supercomputer aimed at data scientists and AI professionals .

In another major market expansion, Huang officially confirmed Nvidia's first PC processors, the N1X Arm-based chips for Windows laptops. The move marks Nvidia's most direct challenge yet to the longstanding x86 duopoly of Intel and AMD in the client PC market . The dual announcement—a new chip for the traditional laptop market and a completely new AI-native PC category—represents a two-front hardware strategy designed to make Nvidia's AI platform nearly inescapable across both cloud and edge computing .

Vera Rubin ramps full production, Taiwan bets $150 billion

In parallel with the PC announcements, Huang confirmed that the Nvidia Vera Rubin next-generation AI data-center platform is ramping into full production. Announced earlier at the San Jose GTC, the platform was positioned in Taipei as specifically engineered for the 'age of agents,' promising 10x higher agent throughput at scale compared to the previous-generation Grace Blackwell architecture . Taiwan's role could not be more central. The island's top server makers are manufacturing Vera Rubin-based systems at scale for global hyperscalers, AI labs, and cloud providers .

The keynote also provided a lens on Nvidia's deepening physical investment in Taiwan. Huang reportedly discussed an increased commitment to the region, with Nvidia deepening its Taiwan investment to $150 billion . The deepening entanglement with Taiwan's manufacturing ecosystem is a strategic necessity as Nvidia attempts to supply the global 'AI factory' build-out.

Market reaction: a stock dip, bullish retail, and an 'amazing' ARM bump

The immediate market reaction to the keynote was layered, reflecting a mix of enthusiasm for the product vision and anxiety over macro and regulatory headwinds. Nvidia's stock price (NVDA) saw a modest dip on the day of the keynote, a move that Tom's Guide reported was likely 'not due to a bum keynote' but rather to simultaneous news that the U.S. Commerce Department was tightening export controls. The regulatory action closed a loophole that allowed Nvidia's AI chip shipments to reach Chinese firms through Malaysian subsidiaries, threatening to block hundreds of thousands of chips .

Despite the share price decline, other sentiment signals were positive. Retail investor sentiment for both NVDA and Microsoft (MSFT) rose to a 'bullish' level following the announcements . Jim Cramer weighed in on social media with a specific beneficiary in mind. 'The Nvidia superchip is obviously additive,' Cramer posted on X, calling the reveal 'amazing for club holding ARM!' ARM stock responded by closing 5.37% higher on the preceding Friday and surging over 13% in premarket trading on the Monday of the keynote . The linkage is strategic: Nvidia's deepening use of Arm-based architectures for its Vera Rubin CPU, its PC processor push with the N1X, and its overall shift to Arm-designed cores directly benefits Arm Holdings as a licensor.

Broader market context for the event was buoyant. The S&P 500 and Nasdaq Composite had both closed at record highs on the Friday before the keynote, with the Nasdaq reaching 26,972.62. The Korean KOSPI had also recently topped 8,000 for the first time, driven in part by anticipation for the Computex announcements . Asia stocks, more broadly, were 'count on AI boom to offset Gulf risks' as regional markets priced in continued AI infrastructure spending .

The five key shifts for enterprise AI strategy

Synthesizing the keynote, analysts for enterprise-focused platforms distilled five strategic shifts, the most foundational of which was the maturation of the open-source agent operating system. Nvidia's OpenClaw framework, launched earlier at the San Jose GTC, was described by Huang as 'the operating system of agentic computers' . OpenClaw agents can navigate file systems, spawn sub-agents, run scheduled tasks overnight without supervision, and decompose complex problems step-by-step. The open-source nature of the tooling is designed to make this layer nearly default, ensuring that the orchestration of AI runs on Nvidia's stack.

NemoClaw takes that open framework and hardens it for production, adding the guardrails, privacy controls, and management tooling that a Fortune 500 audit committee will demand. The combined play is a classic open-core land-and-expand for the data-center operating system market: seed the open-source runtime and monetize the enterprise-grade security, scheduling, and orchestration layer .

For CIOs and enterprise architects, Huang's Taipei keynote made the strategic fork in the road impossible to ignore. The message was that the question is no longer if agentic AI will reshape the enterprise computing stack, but whether a company's agent strategy will be built on an accelerated, vertically integrated platform purpose-built for the 'age of agents' or cobbled together from legacy cloud infrastructure not designed for thousand-fold compute spikes. The software industry, far from dying, is about to get a lot busier.