NVIDIA’s Jensen Huang on Reasoning Models, Robotics, and Refuting the “AI Bubble” Narrative

Task vs. Purpose Framework in Labor (9:03 - 12:31)

• The core idea is that AI automates tasks, but the underlying purpose of a job often remains and can even expand, leading to increased productivity and demand for human roles.
• Huang uses his own typing as an example: AI automates typing tasks, allowing him to do more work and become busier, not less.
• Increased productivity in fields like healthcare leads to greater demand for services, allowing for more hiring, as seen with radiologists.

Solving Labor Shortages with Robotics (12:31 - 15:14)

• Robotic systems are presented as a solution to severe and worsening labor shortages in industries like manufacturing and trucking, driven by aging populations.
• Beyond filling gaps, robotics will create new jobs, such as maintenance and repair for a projected billion robots, becoming the largest repair industry globally.
• The argument extends to other shortage-prone fields like accounting and nursing, where AI can help fill critical gaps.

The Layer Cake of AI Technology (15:14 - 18:39)

• Huang outlines AI as a five-layer technology stack: energy, chips, infrastructure (hardware/software), AI models (diverse modalities beyond language), and applications.
• AI is fundamentally about the automation of intelligence, combinable with mechatronics for embodied tasks.
• AI models are diverse, encompassing human, biological, chemical, physical, financial, and healthcare information, not just chatbots.
• Applications like full self-driving and humanoid robots (Figure) are examples of AI embodied in mechanical forms.

The Importance of Open Source (18:39 - 21:52)

• Open source is deemed essential for the AI ecosystem, enabling startups, companies in various industries (manufacturing, transportation, healthcare), and academic research.
• Without open source, AI development would be stifled, as it provides fundamental pre-trained technology that can be adapted and fine-tuned.
• Policymakers are urged to protect open-source innovation, recognizing its pervasive impact on data centers, mobile devices, and industrial applications.

The Myth of “God AI” and Monolithic Models (21:52 - 23:54)

• Huang dismisses the concept of a singular, all-knowing 'God AI' as unhelpful and unrealistic, emphasizing that AI is a diverse and evolving field.
• The idea that one monolithic model will dominate is impractical, as different industries require specialized AI capabilities (e.g., understanding molecular language vs. human language).
• AI is framed as the next computer industry, and just as no single computer company dominates all aspects, no single AI will solve all problems, necessitating diverse development.

Addressing the “Doomer” Narrative and Regulation (23:54 - 29:25)

• The 'doomer' narrative, portraying extreme dystopian futures, is criticized as unhelpful, damaging to the industry, and potentially leading to misguided regulations that stifle innovation.
• Huang suggests that some extreme narratives might be driven by intentions of regulatory capture, aiming to limit competition for established players.
• He emphasizes that the first aspect of AI safety is its ability to perform as advertised, which requires aggressive investment in advancing technology, as seen in 2025 with improvements in grounding and reasoning.
• The marginal cost reduction of AI will lead to more AI monitoring other AIs, enhancing safety and security, akin to having more police to ensure societal safety.

The Plummeting Cost of Compute and Tokenomics (29:25 - 35:09)

• The cost of generating AI tokens has dramatically decreased, with GPT-4 equivalent models dropping over 100x in 2024, making AI more accessible.
• NVIDIA's architectural advancements (e.g., Volta, Ampere, Hopper, Blackwell, Reuben) deliver 5-10x performance improvements annually, far exceeding Moore's Law.
• This rapid progress could lead to a billion-fold reduction in token generation costs over a decade, making AI training and inference incredibly cost-effective.
• The argument that only large companies can afford to train frontier models is challenged, as the decreasing costs allow new entrants to stay competitive, especially with open-source contributions like DeepSeek.

The Return to Research (35:09 - 37:49)

• The AI field is experiencing a renewed focus on research, alongside continued scaling efforts, leading to diverse approaches and specializations.
• AI labs and startups will increasingly bifurcate into niches, focusing on specific domains (e.g., coding, accessibility) rather than trying to 'boil the entire ocean.'
• This specialization allows for significant breakthroughs in micro-niches, leveraging open-source foundations to achieve superhuman capabilities in specific areas.

Future of Coding and Software Engineering (37:49 - 43:20)

• Coding is identified as a prime example of an application-level business benefiting from AI, with tools like Cursor enhancing developer productivity.
• The purpose of a software engineer is problem-solving and discovering new problems, with coding being just one task; AI automates coding, freeing engineers for higher-level work.
• NVIDIA's commitment to programmable architecture ensures flexibility and broad compatibility for evolving AI algorithms, driving down costs and fostering innovation across its large installed base.
• MVLink72 is highlighted as the lowest-cost token generation machine, demonstrating NVIDIA's focus on efficient inference.

The Industries Due For Their “ChatGPT” Moments (43:20 - 46:00)

• Several industries are poised for their 'ChatGPT moment' due to advancements in multi-modality, long context, and synthetic data generation.
• Digital biology is a key area, with rapid progress in protein understanding (e.g., NVIDIA's LaPrina model) and an impending shift to protein generation and chemical understanding.
• The combination of generative AI and multimodal capabilities will create a data flywheel for foundational models in proteins and cells.

The Evolution of Self-Driving Cars and Robotics (46:00 - 54:06)

• Reasoning models will significantly improve self-driving cars, moving beyond perception and planning to handle novel situations by breaking them down into known components.
• The history of self-driving cars is described in four eras: smart sensors, perception/world model/planning, end-to-end models, and future end-to-end reasoning models.
• Robotics, particularly human robots and multi-embodiment robots, are expected to see rapid advancements, benefiting from foundational technologies developed for self-driving.
• While incumbents like Tesla and Waymo lead, the vast scope of 'everything that moves will be robotic' suggests ample room for startups and niche applications, especially in vertical solutions that ensure 99.999% reliability for industrial use cases.

Energy Demand and Growth for AI (54:06 - 58:49)

• The massive energy demand for AI is acknowledged, with Huang advocating for all forms of energy (natural gas, nuclear, wind, solar) to support industrial growth.
• He credits President Trump's 'drill, baby drill' policy for preventing the US from ceding the industrial revolution to other nations.
• The demand for AI is seen as a primary driver for climate innovation, spurring investment in new battery companies, solar concentrators, and sustainable energy solutions.
• Huang criticizes 'doomer' messages for potentially scaring people away from necessary AI investments that enhance safety, functionality, and productivity.

2026 Outlook: US-China Relations and Geopolitics (58:49 - 1:04:43)

• Huang expresses optimism for improved US-China relations in 2026, advocating for a nuanced strategy that recognizes both countries as adversaries and partners.
• He argues that decoupling is naive due to the deep interdependence between the two economies, emphasizing the need for both nations to invest in their own independence while managing their relationship productively.
• The US export control policy is seen as grounded in national security, technology leadership, and national prosperity, allowing American companies to maintain leadership.
• China's internet growth has historically benefited American chip companies (Intel, AMD, Micron) and contributed significantly to open source, which in turn benefits American startups.

Is There An AI Bubble? (1:04:43 - 1:16:20)

• Huang refutes the idea of an AI bubble, explaining that the fundamental shift from general-purpose to accelerated computing is driving demand, irrespective of generative AI.
• He argues that if chatbots didn't exist, NVIDIA would still be a multi-hundred-billion-dollar company due to this underlying shift.
• The demand for AI capacity is immense and widespread across all industries (AV, digital biology, financial services, robotics), not just chatbots.
• The world's $100 trillion GDP, with $2 trillion in annual R&D, is increasingly shifting towards AI-powered approaches, necessitating vast supercomputing infrastructure.
• Every startup, researcher, and company is 'dying for capacity,' indicating a genuine, global shortage rather than a bubble.
• He dismisses studies claiming AI isn't useful in enterprises, suggesting they overlook the rapid innovation happening in startups and the need for proper change management.

Conclusion (1:16:20 - 1:16:20)

• Huang concludes by expressing immense pride in the industry's tremendous strides in 2025, particularly in technology advancements and the intact scaling laws that link compute to intelligence.
• He anticipates an extraordinary next five years, with 2026 being particularly incredible, as innovations rapidly diffuse across sectors.

"If the marginal cost of AI is going to go down significantly, that one AI is going to be monitored by millions of AIs." — Jensen Huang

"In the case of AI over the course of 10 years is probably 100,000 to a millionx... I would not be surprised." — Jensen Huang

"Coding is one of the tasks... The purpose of a software engineer is to solve known problems and to find new problems to solve." — Jensen Huang

"The chat GBT moment the generative AI moment all of that stuff is coming together for for um digital biology." — Jensen Huang

"Everything that moves will be robotic." — Jensen Huang

"Without energy, there can be no new industry." — Jensen Huang

"AI was is probably the biggest driver for sustainable energy ever." — Jensen Huang

"Decoupling is naive and the idea of decoupling... is just not not it's not based on any common sense." — Jensen Huang

"If generative AI... if none of that existed today Nvidia would be a multiundred billion dollar company and the reason for that is because as you know the foundation of computing is shifting to accelerated computing." — Jensen Huang

"Everybody is dying for capacity." — Jensen Huang