Jensen Huang's Computex 2025 Keynote: AI DCs, factories and AGI

Techday NZ

19-05-2025

Imagine the excitement of being invited to see Jensen Huang himself in person, only to be told 'well, you'll have to livestream from your hotel room because you aren't really invited to see Jensen Huang himself in person.' If this wasn't an exciting enough development at Computex 2025 taking place now in Taipei, then perhaps the keynote itself was: as for your humble correspondent, it is available for streaming from multiple platforms, and you don't have to travel to Taiwan to take in the spectacle.
No prizes for guessing the topic du jour. Helmed by Huang, Nvidia makes the chips (and other bits) that make the AI work, and he was talking AI data centres and 'AI factories' as the digital sweatshops for a new industrial revolution.
Huang highlighted NVIDIA's technological advancements and aligned with broader industry discussions about artificial intelligence's trajectory, including recent observations from former Google CEO Eric Schmidt on artificial general intelligence (AGI). Schmidt's warnings about AGI's rapid emergence and potential to outstrip human control provide a critical lens through which to view Huang's ambitious plans for AI infrastructure.
But, we keep in mind the comment from some anonymous Facebook commenter that perhaps AGI will get a Nietzschean dose of self-awareness and delete itself. Wouldn't that be a development.
The data centre as AI factory
The DC today has evolved beyond data storage and processing to generating 'tokens'—units of intelligence that power applications like robotics, chemical simulations, and large language models. "For the first time, we have a new type of data centre that is about AI generation," Huang declared, producing intelligence at scale.
This vision builds on NVIDIA's technological foundation, particularly the Blackwell platform and the forthcoming Blackwell Ultra AI series with GB300 chips, set to launch later in 2025. He said Nvidia's software ecosystem, including Omniverse, CUDA, and NVIDIA Inference Microservices (NIMs), further enables developers to leverage these AI factories for domain-specific applications across industries like healthcare, manufacturing, and finance.
Huang's ambition for AI factories includes a bold claim that there are "no laws of physics" preventing data centres from scaling to a million chips. He predicted a millionfold increase in computing power within a decade, driven by AI factories operating 24/7 to produce intelligence accessible to (almost) all.
Eric Schmidt's AGI warnings: A sobering counterpoint
While Huang's keynote was optimistic about AI's potential, Eric Schmidt's recent observations on AGI introduce a cautionary perspective. Speaking at the AI + Biotechnology Summit in April 2025, Schmidt anticipated AGI emerging within three to five years, potentially by 2028–2030. He warned that once AI achieves recursive self-improvement (the ability to improve its own performance), it "won't have to listen to us anymore," leading to artificial superintelligence (ASI) surpassing all human intelligence combined by 2031.
Schmidt's timeline aligns with Huang's vision of scaling AI factories, but he noted the risks of unchecked development. AI systems could soon rival the best human experts in fields like mathematics, physics, and coding, with generative models potentially supplanting most programmers within a year.
This capability, while enabling Huang's AI factories, raises concerns about control – and, getting back to existential issues, the role of people in all of that. Schmidt has advocated for "pulling the plug" if AI becomes unobservable or uncontrollable, a stance he reiterated in discussions about agentic AI, where systems autonomously make decisions.
The biggest question is both simple and profound: what is the purpose of it all? There's an abyss out there wondering who, or what, will stare into it.
Implications for AI Factories
Schmidt's warnings cast a shadow over Huang's optimistic vision. The AI factories Huang champions rely on massive computational power, which Schmidt argues is the foundation for AGI's rapid advance. The GB300 chips and million-chip data centers could accelerate the path to AGI, potentially outpacing society's ability to establish ethical and regulatory frameworks. Schmidt's call for transparency—"If we can't observe it, we unplug it"—clashes with the proprietary nature of NVIDIA's ecosystem, where closed-source models and hardware dominate.
Moreover, Schmidt's concerns about misuse resonate with Huang's acknowledgment of geopolitical risks. Huang hinted at diversifying Nvidia's manufacturing beyond TSMC in Taiwan (where air raid shelters are to be seen on the immaculate streets), citing U.S.-China tensions and supply chain vulnerabilities.
Schmidt, in a 2024 interview, noted that China lags two years behind the U.S. in AI but faces unique challenges due to its censorship regime, which could limit generative AI's spread. Both leaders recognize the global stakes, but Schmidt's focus on governance contrasts with Huang's emphasis on technological and economic gains.
The abyss beckons, but if not for thee, for whom, or what, does the bell toll?