NVIDIA: OpenAI, Future of Compute, and the American Dream | BG2 w/ Bill Gurley and Brad Gerstner

The discussion highlights the transformative power of AI and the accelerating pace of technological advancement, projecting it as an industrial revolution. This revolution is driven by new scaling laws in AI, moving beyond single-shot inference to a "thinking" process involving research and learning. This evolution is significantly increasing compute requirements, necessitating massive infrastructure buildouts.

A pivotal development discussed is the substantial partnership and investment between a major technology company and an AI research firm. This collaboration aims to build dedicated AI infrastructure, enabling the AI firm to scale as a hyperscale company. This strategic move addresses the dual exponential growth in customer base and computational demand driven by AI's improving capabilities and reasoning.

The conversation also delves into the future of computing, emphasizing the shift from general-purpose computing to accelerated and AI computing. It explores the immense market opportunity in AI infrastructure, the challenges of keeping pace with technological advancements, and the competitive landscape, including the development of specialized chips. The discussion underscores that AI is not just about better models but about building and scaling the entire AI ecosystem.

• AI is driving an industrial revolution with new scaling laws, moving from single-shot inference to "thinking" AI that requires vastly more compute.
• A significant partnership and investment are detailed, focused on building dedicated AI infrastructure for a growing AI firm to become a hyperscale company.
• The shift is from general-purpose computing to accelerated and AI computing, creating massive opportunities and demanding continuous innovation.
• The annual release cycle for hardware and extreme co-design are critical for keeping pace with AI's exponential growth.
• Discussions touch on national security, sovereign AI capabilities, the global AI race, and the importance of attracting and retaining top talent in the US.

Related Video Summary

• Nvidia Investing $100 Billion in OpenAI
• Internet EXPLODES w/ New Jensen Interview

The AI Industrial Revolution and New Scaling Laws [0:00]

• OpenAI is projected to be the next multi-trillion dollar hyperscale company.
• Over 40% of current revenue is from inference, which is about to increase by a billion times due to "chain of reasoning."
• This era is described as an industrial revolution, with AI time feeling like a hundred years in just one year.
• A key prediction from a year and a half ago was that inference would not just 100x or 1000x, but 1 billion times.
• There are now three scaling laws: pre-training, post-training (AI practicing a skill), and inference.
• The new way of inference involves "thinking" before answering, which includes research, ground truth checking, learning, and then generating an answer.
• Confidence in the 1 billion times growth of inference and the resulting intelligence levels is higher this year due to agent systems and AI being a "system of language models."
• AI is now multimodal, generating incredible video content.

The new way of doing inference, which we appreciate, is thinking. So think before you answer.

Partnership with OpenAI and Infrastructure Buildout [3:23]

• A significant deal with OpenAI is announced, where the company will be a preferred partner and invest $100 billion over time.
• This deal involves building 10 gigawatts of capacity, potentially leading to upwards of $400 billion in revenue for Nvidia if OpenAI uses their hardware.
• The AI firm is expected to become the world's next multi-trillion dollar hyperscale company, with both consumer and enterprise services.
• Investing in this firm before it reaches that scale is considered one of the smartest investments imaginable.
• The partnership involves several projects: continuing Microsoft Azure buildout, OCI buildout with SoftBank, and a new initiative to help OpenAI build its own self-built AI infrastructure.
• This new partnership focuses on working directly with OpenAI at the chip, software, systems, and AI factory levels.
• OpenAI is experiencing two exponentials: customer growth driven by improving AI and use cases, and computational growth per use due to "thinking" inference.
• These compounding compute requirements necessitate building out extensive projects.

So first of all that I'll answer that last question first and then I'll come back and present my way through. I think that OpenAI is likely going to be the next multi-trillion dollar hyperscale company.

The Shift to Accelerated Computing and AI Infrastructure [10:43]

• General-purpose computing is over; the future is accelerated computing and AI computing.
• The world's computing infrastructure, worth trillions, needs to be refreshed with accelerated computing.
• Moore's Law is effectively dead, leading to the shift from general-purpose to accelerated computing.
• AI's first use cases are already widespread in search, recommender engines, and shopping, now utilizing GPUs instead of CPUs.
• Hyperscale computing is transitioning from CPUs to accelerated computing and AI, representing hundreds of billions of dollars.
• Beyond mundane tasks, AI and accelerated computing unlock new applications and opportunities.
• AI supercomputers and factories will generate tokens to augment human intelligence, which contributes significantly to global GDP.
• An analogy of an employee augmented by AI is used: a $100,000 employee with a $10,000 AI becomes twice as productive.
• This productivity increase is being implemented across all employees and software engineers within the company.
• The total market for AI infrastructure is estimated at $400 billion annually, with potential for a 4-5x increase.
• Data center power is expected to increase 10x, correlating directly with revenue, as token generation doubles every few months.
• Performance per watt must increase exponentially to keep pace, driving innovation in hardware.

The first point and this is the laws of physics point. This is the most important point that general general purpose computing is over and the future is accelerated computing and AI computing.

The Economics of AI Infrastructure and Market Growth [14:54]

• The AI infrastructure market is estimated at $400 billion annually, with a projected 4-5x increase.
• Global GDP is expected to accelerate due to AI providing billions of co-workers.
• The demand for AI infrastructure is driven by AI generating tokens, which requires factories, and this revenue can have significant gross margins.
• The math suggests that $5 trillion in annual capital expenditure could align with the AI infrastructure needs.
• The total addressable market (TAM) is projected to be a 4 to 5x increase over the current $400 billion.
• One company plans to increase its data center power by 10x by the end of the decade, aligning with Nvidia's revenue correlation to power.
• Token generation is doubling every few months, necessitating exponential improvements in performance per watt.
• Revenue per watt is essentially revenue in this future.
• The historical GDP growth, which was stagnant for 2,000 years, accelerated with the industrial and digital revolutions, and AI is poised to do the same.

So if you told me that on an annual basis the capex of the world was about $5 trillion, I would say the math seems to make sense.

Addressing Market Skepticism and Future Projections [18:54]

• Skeptics discuss potential "glut" or "bubble" scenarios, contrasting with optimistic projections from industry leaders.
• One estimate projects $100 billion in AI revenue for 2026 (excluding certain segments).
• The goal is to reach at least a trillion dollars in AI revenue by 2030.
• The argument is made that hyperscalers' entire revenue base is already AI-driven, making the transition from CPUs to GPUs for AI near certain.
• The probability of achieving a trillion dollars in AI revenues by 2030 is considered near certain, as the transition is already well underway.
• A "glut" scenario is deemed extremely unlikely until general-purpose computing fully converts to accelerated computing and AI.
• The annual release cycle for hardware and the co-development with the supply chain are seen as increasing the competitive moat.

But when you think about that, it sounds a little bit like prisoners dilemma.

Hardware Innovation: GPUs vs. ASICs and Co-Design [33:15]

• An annual release cycle for hardware has been adopted, with significant upgrades like Grace Blackwell in 2025 and Vera Rubin in 2026, followed by Ultra and Fineman.
• This annual cadence is enabled by AI, allowing for faster development and scale.
• The token generation rate is increasing exponentially, doubling the compute requirements, necessitating annual performance increases to drive down costs.
• From Kepler to Hopper, there was a 100,000x performance increase; between Hopper and Blackwell, a 30x increase is expected.
• Future releases are projected to bring further significant gains.
• The focus is on breaking down the entire problem at the system level and changing chips, software, and systems simultaneously, a practice called "extreme co-design."
• This extreme co-design involves optimizing the model, algorithm, system, and chip together.
• Nvidia's strength lies in its integrated approach, from CPUs and GPUs to networking and switching, creating an AI factory.
• The competitive moat is increasing due to extreme co-design, the scale of customers, and the robust supply chain.
• The debate between GPUs and ASICs is discussed; while ASICs can be cheaper for niche workloads, Nvidia's strength lies in building comprehensive AI infrastructure systems.
• Nvidia views its business holistically, looking at the entire AI infrastructure and diverse workloads, rather than just GPUs.
• The programmability of CUDA is crucial for iterating on AI models and algorithms.
• The concept of "customer-owned tooling" is mentioned in the context of large-scale ASIC businesses.
• Nvidia's strategy includes opening up its platform to allow other companies to plug in their specialized products.

The transistors aren't really helping us very much, right? Moors law is largely the density is growing up but going up but the performance is not.

Geopolitics, National Security, and the Global AI Race [58:48]

• Sovereign AI capabilities are becoming crucial for nations, viewed as existential to national security and economic security, comparable to nuclear power in the 1940s.
• AI is fundamentally different from nuclear power because everyone needs AI, while not everyone needs atomic bombs.
• AI is described as "modern software," requiring democratization to ensure every country can participate.
• To participate in AI, countries need to encode their history, culture, and values within it.
• It is recommended that countries utilize open models like OpenAI, Gemini, Grok, and Anthropic, while also dedicating resources to building their own AI capabilities for various applications.
• Every country needs AI infrastructure, just as they need energy and communication infrastructure.
• The role of AI in the US government is highlighted, with individuals understanding the complexity of technology and policy.
• The US government's pro-energy stance is seen as crucial for AI development.
• The narrative around China's technological advancement and the perceived mistake of "unilaterally disarming" by pushing Nvidia out of China is discussed.
• The potential for Huawei to accelerate due to monopoly profits in China is raised.
• Export controls and tariffs on chips to China are mentioned, alongside potential Chinese responses.
• The competitive relationship with China is acknowledged, with the understanding that Chinese companies aim to succeed and operate in an open market.
• The importance of the Chinese market to Nvidia is emphasized, despite current guidance excluding it.
• The idea that competing in China, rather than decoupling, could actually increase the US's probability of winning the global AI race is presented.
• The notion of "China hawks" is critiqued, arguing that hindering the flow of technology is not patriotic.
• The importance of "the American dream" and the right to rise is a recurring theme, emphasizing the need to attract and retain global talent.
• The potential negative impact of high H-1B visa fees on attracting talent and accelerating investment outside the US is discussed.
• The concern over a precipitous drop in Chinese AI researchers wanting to come to the US is highlighted as an early indicator of future problems.
• The analogy of a sports team recruiting top players is used to illustrate the importance of attracting talent for continued success.
• The idea that "China hawks" might be hurting the US by discouraging talent and competition is debated.
• The concept of decoupling is seen as the wrong approach, with a focus on competition and deal-making favored.
• The president's "bring it on" attitude and belief in America's ability to compete are emphasized.

Nobody needs atomic bombs. Everybody needs AI.

Reindustrialization, Economic Growth, and the Future of Work [31:14]

• The shift to an annual release cycle for hardware is driven by AI's exponential growth in token generation and customer use.
• Increasing performance annually is crucial to counteract rising costs due to the end of Moore's Law.
• Extreme co-design is the method used to achieve significant performance gains across systems, chips, and software.
• The supply chain is geared up for massive buildouts, with companies providing three-year visibility.
• The competitive moat is growing due to advanced co-design, extreme scale, and a robust supply chain.
• The argument that Nvidia's systems are a better bet than cheaper ASICs, even if given for free, is based on total operating cost, power efficiency, and intelligence output.
• The performance per watt is twice as high for Nvidia's systems, enabling customers to produce twice as much revenue.
• The transition to accelerated computing and AI is seen as transformative for America, with reindustrialization creating new jobs and opportunities.
• AI is positioned as the ultimate equalizer, closing the technology divide by allowing users to interact with AI using natural language.
• The idea that AI will lead to mass job destruction is countered by the premise that humans will continue to have new ideas and create new work.
• Intelligence is not a zero-sum game; more intelligent individuals lead to more ideas, more problems to solve, and more work and jobs.
• The next few decades are expected to see economic growth, new job creation, and transformation of all jobs.
• The acceleration of change driven by AI is likened to 20,000 years of progress in the 21st century.
• The fusion of AI and mechatronics/robotics is anticipated, leading to AI companions and embodied AI in various devices.
• Understanding the complexity of biology and creating digital twins for healthcare is another predicted advancement.
• The advice is to "get on the train" of accelerating technological advancement rather than trying to predict its exact future path.
• Leaders like those in tech and government are crucial for navigating this pivotal time.
• The concept of "Invest America" aims to provide every child with an investment account at birth to participate in capitalism and share in the country's growth.
• Reindustrializing America, encouraging companies to build domestically, and reskilling the workforce are seen as transformative.
• AI's role as an equalizer and its ability to bridge the technology divide are highlighted.
• While some tasks may be eliminated, many new tasks will be created, and jobs will be transformed, not necessarily destroyed en masse.
• The belief is that greater intelligence leads to more ideas, more problems to solve, and ultimately, more work and jobs.

The idea that that we no longer uh uh uh make it only that you get a PhD or you go to you know one of the great schools and only in that way can you build a great life right and and deserve to have a great living uh we've got to change all that doesn't make any sense we love craft right I love people who make things with their hands and and we're now we're now going to go back and build things build magnificent incredible things

Immigration Policy and Talent Acquisition [17:10]

• The $100,000 H-1B visa fee is presented as a "great start" to address illegal immigration and abuse of the H-1B system.
• The "American dream" and the country's singular brand of offering opportunities for people to realize their dreams are central to the discussion.
• The importance of legal immigration and the need for a pragmatic solution to the immigration crisis are emphasized.
• The high fee might tilt the playing field in favor of large companies, making it harder for startups.
• There's a concern that such policies could lead to increased investment outside the US.
• The need to recruit the world's best and brightest is critical for the country's success, and policies should not sacrifice this brand.
• The loss of top AI researchers from China to the US is a significant concern, indicating a potential problem for future talent acquisition.
• The idea of "China hawks" is criticized, with the argument that destroying the pipeline of the "American dream" is not patriotic.
• Competition with China is necessary, but it should not come at the expense of alienating or mistreating individuals.
• The importance of fostering a welcoming environment for international talent and implementing a strategic plan for recruitment is crucial.
• The current situation regarding immigration is acknowledged as not ideal, and there's a need to move towards better solutions.
• The concern that foreign students might choose to study elsewhere or leave the US after graduation is a significant worry.
• The concept of "KPIs" (Key Performance Indicators) is used to describe the desire of smart people to come to and stay in the US as indicators of future success.
• The argument is made that allowing American technology to proliferate globally, including in China, is in the best interest of both countries and maximizes US economic and geopolitical influence.
• The belief that wisdom and truth will prevail in resolving complex issues like those with China is expressed.
• The idea that the US should not "decouple" from China but rather focus on competition and deal-making is presented.
• The historical success of American technology spreading globally is contrasted with current protectionist sentiments.
• The importance of American technology as a national treasure and an essential part of trade is emphasized.
• The belief that the US can win the global AI race by competing globally, including in China, and by attracting and retaining talent is a core argument.
• The positive outlook on President Trump's approach to China, emphasizing deal-making rather than confrontation, is shared.

America has one a singular brand reputation that no country in the world has. And no country in the world is in a position or in the horizon to be able to say come to America and realize the American dream.