NVIDIA CEO Jensen Huang

Nvidia was valued at approximately $1.1 trillion at the time of the interview in 2023, making it the world's sixth-most-valuable company. Its position at the center of the emerging AI industry raised two major questions: whether AI would become a multitrillion-dollar market and whether Nvidia could maintain its competitive advantage.

Jensen Huang founded Nvidia in 1993 to transform personal computers through hardware-accelerated graphics. He said Nvidia was the only consumer 3D graphics company when it started, at a time when Windows graphics depended heavily on software rendering.

Nvidia's original graphics architecture turned out to be fundamentally wrong. Its early NV1 and NV2 designs used forward texture mapping, curves and tessellation rather than conventional triangle-based rendering. The architecture avoided traditional Z-buffers but ultimately proved incompatible with the direction of the industry.

Microsoft's introduction of DirectX undermined Nvidia's original technology strategy. Nvidia had developed its own approach to graphics programming, but DirectX established a different industry standard that the company's existing architecture could not properly support.

By 1997, Nvidia faced approximately 30 competitors despite having started as an early pioneer. Huang described the combination of architectural mistakes, rising competition and shrinking financial resources as an existential crisis.

Nvidia had approximately six months of cash remaining when it committed to the RIVA 128 graphics processor. The company could not afford the traditional semiconductor development process of fabricating a prototype, discovering problems, redesigning it and manufacturing another version.

Huang considers Nvidia's decisions during the 1997 crisis among the best in its history. He said the company had previously made numerous mistakes, but the decisions made when its survival was threatened established practices that remain central to Nvidia's engineering approach.

Nvidia abandoned its effort to resist DirectX and instead decided to build the best possible hardware for Microsoft's standard. Huang considered fighting the established software ecosystem a losing strategy and chose to compete through superior performance.

The RIVA 128 was designed as a fully hardware-accelerated 3D graphics pipeline. Huang said Nvidia accelerated the complete rendering process, including transformations, projection and the framebuffer, rather than relying on software for significant stages.

Nvidia deliberately pushed the RIVA 128's physical limits. Engineers incorporated texture caching, expanded bus and framebuffer capabilities as far as practical, used extremely fast memory and designed one of the largest graphics chips attempted at the time.

Nvidia chose an unusually expensive chip design because it wanted competitors to be unable or unwilling to match its performance. Rather than optimizing primarily for manufacturing cost, the company selected a cost structure substantially above what it expected competing manufacturers would accept.

Huang believed the developing PC market contained enough enthusiasts willing to pay for extraordinary performance. Nvidia reasoned that a product offering dramatically faster graphics could attract customers even at a premium price, and Huang said that prediction proved correct.

Nvidia deliberately entered graphics because performance requirements would keep increasing. Huang described 3D graphics as a sustainable technology opportunity: improving graphics would create demand for even better graphics rather than permanently satisfy customers.

The RIVA 128 shipped with support for only eight of DirectX's 32 blending modes. Huang acknowledged that Nvidia should have implemented all 32, but he discovered the omission too late and had to proceed with the hardware already developed.

Nvidia purchased emulation technology from Ikos, a company Huang said was shutting down because it had no customers. Huang contacted the company and offered to purchase its available equipment without requiring additional assurances, giving Nvidia a way to test its chip before physical manufacturing.

The Ikos emulator allowed Nvidia to build and test its graphics software before the chip existed physically. Dwight and the software team developed the software stack against the simulated hardware instead of waiting for silicon.

Early hardware emulation was extraordinarily slow but still economically valuable. Huang recalled watching Windows render at a speed somewhere between roughly a minute and an hour per frame, yet that slow simulation was preferable to risking another costly fabrication cycle.

Nvidia tested its drivers, software stack, available games and VGA applications before committing the RIVA 128 to manufacturing. Huang's confidence in a successful production run came from moving the testing normally performed after fabrication into the period before fabrication.

Nvidia went directly from design completion into production without first receiving and testing a physical prototype. With enough confidence in the simulation, the company committed its remaining resources to manufacturing, marketing and developer relationships simultaneously.

Huang rejected the idea that Nvidia's survival strategy was simply reckless gambling. He argued that a genuine company-defining commitment should involve identifying future risks, moving them into the present and testing as many assumptions as possible before making an irreversible decision.

The 1997 crisis established Nvidia's philosophy of avoiding unnecessary development cycles. Huang argued that fabricating a chip once instead of repeatedly improves cost, engineering efficiency, time to market and competitiveness. He said this remains fundamental to how Nvidia designs products.

Huang sees development speed as a form of product performance. A technically superior chip delivered too late can lose its competitive value, making engineering efficiency and release timing strategic considerations rather than merely operational ones.

CUDA emerged from Nvidia's earlier experiments with Cg and higher-level GPU programming. Before committing to general-purpose GPU computing, Nvidia was already investigating how developers could express complex calculations through abstractions above the graphics hardware.

Nvidia experimented with using GPUs for scientific and medical workloads before CUDA became a major platform. These included CT image reconstruction and image processing, providing early evidence that graphics processors could perform valuable work beyond rendering.

The parallel structure of programmable shaders convinced Huang that GPUs could become general-purpose processors. Their ability to execute large numbers of threads simultaneously suggested a computing architecture fundamentally different from conventional sequential processing.

Nvidia committed enormous engineering resources to CUDA before demand for machine learning became obvious. The interviewers estimated that the platform represented approximately 10,000 person-years of engineering work, reflecting how far ahead of the eventual commercial market Nvidia invested.

AlexNet's breakthrough in computer vision became a decisive moment in Huang's understanding of AI. He said its effectiveness appeared to leapfrog approximately 30 years of computer-vision research, forcing Nvidia to investigate why the approach worked and whether it could scale.

Nvidia had already been studying computer vision and the ImageNet competition before AlexNet appeared. Huang said many established vision algorithms were poorly suited to CUDA, making AlexNet especially interesting because it suggested a more natural fit between deep learning and GPU computing.

Huang interpreted deep neural networks as potentially universal function approximators. Because they could learn relationships from large datasets and support increasingly complex models, he reasoned that their applications might extend far beyond computer vision.

The ability to scale models and datasets became central to Nvidia's early AI investment thesis. Huang believed increasing training data, network depth and computational resources could continually improve model performance, making accelerated computing increasingly valuable.

Huang distinguished predictive computing from traditional scientific modeling based on causation. Conventional algorithms often begin by understanding the physical causes of a phenomenon, while machine learning can produce commercially useful predictions without fully explaining those causes.

Consumer preferences offered an early demonstration of machine learning's economic potential. Huang cited predicting toothpaste preferences, condiment purchases, music, movies and social-media recommendations as examples where accurate predictions could be valuable even without understanding the customer's underlying reasoning.

Huang believed predictive AI could eventually transform scientific computing as well as commercial software. He extended the argument to weather forecasting and other complex systems, where learning predictive relationships from data could complement or replace aspects of manually constructed algorithms.

The possibility that machine learning could transform nearly every software category changed Nvidia's long-term hardware strategy. Huang reasoned that if software were increasingly developed through learned models, computers and processors would need architectures optimized for that approach.

CUDA's original scientific community gave Nvidia an established network of researchers before the AI boom. Huang said CUDA had effectively democratized supercomputing by allowing researchers to access powerful parallel processing without relying exclusively on traditional supercomputing installations.

CUDA was already used across numerous scientific disciplines before AI became its defining commercial application. Huang identified molecular dynamics, medical imaging, seismic processing, weather simulation and quantum chemistry among its uses.

Nvidia approached leading AI researchers directly to understand their computational needs. Huang described building relationships with Yann LeCun, Andrew Ng and Geoffrey Hinton, among others, and regularly attending AI conferences.

Huang met Ilya Sutskever through Nvidia's engagement with the AI research community. Those relationships allowed Nvidia to follow the development of deep learning before it attracted widespread commercial attention.

Huang also encountered Ian Goodfellow's early generative adversarial networks when their results were extremely primitive. He recalled an early GAN producing an approximately 32-by-32-pixel, blurry image of a cat, yet saw the possibility of dramatic improvement through scaling.

Research relationships gave Nvidia continuous feedback on where AI hardware and software needed to improve. Huang said Nvidia's early role was not simply supplying chips but developing systems and software that allowed researchers to advance their work.

Huang was not involved in founding OpenAI in 2015. He nevertheless knew several people associated with its founding, including Elon Musk, Peter Thiel and Ilya Sutskever, and said some early OpenAI personnel subsequently joined Nvidia.

Huang personally delivered Nvidia's first DGX system to OpenAI. Nvidia had initially developed and used the first version of DGX internally before providing the system to the AI research organization.

Nvidia's early support for OpenAI reflected a long-term belief in research rather than an immediately proven commercial market. Huang acknowledged that early AI systems had limited practical usefulness but believed successive improvements could make them extraordinarily capable.

Huang followed AI research papers closely to measure the pace of progress. He recalled important developments initially appearing approximately every three months, compared with new research being published virtually every day by the time of the interview.

The hosts identified scaling as an important surprise in the development of large language models. They said models became substantially more accurate, useful and lifelike as parameter counts passed roughly 10 billion and then 100 billion, even without the fundamental algorithmic breakthroughs some researchers had expected.

Huang was particularly impressed by the simplicity of self-supervised language-model training. His early reaction to BERT was that predicting masked words from existing text provided a clever way to create training problems without manually labeling enormous datasets.

Huang viewed language models as systems that compress information, including patterns of reasoning, from human text. Because written language contains examples of logic, explanations and common sense, he considered it plausible that sufficiently capable models could learn some reasoning abilities from reading.

Huang regarded emergent AI capabilities as remarkable but not necessarily inexplicable. He believed some unexpected behaviors could be understood through the reasoning patterns encoded in training data, while acknowledging that their appearance remained astonishing.

The hosts explained how machine learning became commercially valuable at Google and Facebook through controlled experimentation. New models were deployed to selected users through A/B testing, allowing engineers to measure product improvements before wider releases and reward research that produced measurable business results.

Huang's management structure departs substantially from conventional corporate hierarchies. The interviewers described him as having more than 40 direct reports, while Huang explained that Nvidia deliberately avoids a military-style chain of command.

Huang organizes Nvidia around the architecture of the products it builds. He compared the company to a computing stack containing architecture, chips, software and higher-level modules, with people and teams responsible for particular components.

At Nvidia, leadership responsibility is intended to reflect expertise rather than formal title. Huang said the person best qualified to manage a particular function or module should lead that work, regardless of their position in a conventional hierarchy.

Huang believes an organizational chart should reflect how a company's products are actually created. He questioned why companies making fundamentally different products often adopt nearly identical management structures.

Nvidia uses the principle “mission is the boss” to organize work across departments. Teams and resources are assembled around concrete objectives rather than being limited by departmental boundaries.

Nvidia's missions are specific engineering and product deliverables rather than only broad corporate ambitions. Huang gave examples including building Hopper, developing its systems, preparing CUDA for Hopper, developing cuDNN and delivering cloud products.

Huang described Nvidia's internal collaboration as resembling a neural network. People from different areas connect according to the needs of a mission, creating a flexible structure that cuts across formal reporting relationships.

Nvidia deliberately distributes important information to different levels of the organization simultaneously. Huang described a robotics meeting attended by new college graduates, three vice presidents and two executive staff members, all of whom participated in discussions and learned decisions at the same time.

Huang believes controlling privileged information should not be the basis of managerial authority. In his structure, executives must justify their positions through reasoning, judgment and helping colleagues succeed rather than through exclusive access to decisions.

The downside of Nvidia's relatively flat information structure is greater pressure on leaders. Without the advantage of controlling information, managers must continually demonstrate their competence and value to the organization.

Nvidia's unusually rapid product-development cadence was highlighted as a competitive distinction. The hosts compared its ability to deliver major technology generations in rapid succession with the difficulty of imagining other large technology companies launching flagship products at a similar pace.

Huang studies successful businesses without attempting to copy their management systems. He said he reads business books for useful experiences and principles, then asks how those lessons apply to Nvidia's circumstances, capabilities and objectives.

Huang actively studies competitors and unrelated industries for useful ideas. He believes companies should examine external innovations rather than ignore them, while developing their own responses instead of adopting someone else's strategy wholesale.

Nvidia's move into data centers began long before the recent AI boom. Huang traced the decision back approximately 15–17 years, when he recognized that tying graphics hardware to a desktop computer and monitor would eventually limit Nvidia's addressable market.

The original strategic insight was to separate where computing happens from where its results are experienced. Huang reasoned that rendering and processing could occur remotely while images and other outputs were delivered to users through connected devices.

An Nvidia engineer demonstrated the technical foundation for remote graphics by capturing a framebuffer, encoding it as video and streaming it to another device. Huang said that demonstration helped lead to Nvidia's cloud-gaming efforts.

GeForce NOW was Nvidia's first data-center product, according to Huang. It moved graphics processing away from the user's local machine, although the company understood that networking latency and physical distance would make cloud gaming a difficult long-term project.

Nvidia's second data-center application involved remote graphics for enterprise customers. This placed Nvidia GPUs inside corporate data centers and expanded the company's experience beyond consumer gaming.

Nvidia's subsequent data-center products combined GPUs with CUDA to create supercomputing systems. This progression established hardware, software and operational capabilities that later became essential for AI infrastructure.

The shift to data centers removed the assumption that Nvidia could sell only one GPU per individual computer user. Computing resources could instead be centralized, shared, expanded and accessed remotely, substantially increasing Nvidia's potential market.

Nvidia's data-center business was not originally synonymous with AI. Huang emphasized that cloud gaming, enterprise graphics and scientific supercomputing preceded the AI explosion and helped prepare Nvidia for it.

Huang believes CEOs must position companies near future opportunities before their exact commercial form becomes clear. He compared this to standing near a tree before an apple falls: a company does not need to predict everything perfectly if it has already developed the ability to seize an opportunity.

Huang also believes companies should generate revenue from intermediate technologies while preparing for larger future markets. Each stage must help sustain the business, allowing it to maintain investments until the anticipated opportunity arrives.

Huang considered remote computing practical despite unavoidable speed-of-light limitations. He discussed approximate network-latency figures in the range of 70–120 milliseconds, while acknowledging uncertainty about the precise numbers, and argued that eliminating avoidable delays could make remote computing useful on a global scale.

Nvidia acquired Mellanox because Huang believed data-center computers required far more than advanced processors. He reasoned that networking, distributed computing, infrastructure and security distinguish data centers from individual PCs or mobile devices.

Huang saw Mellanox as providing capabilities that Nvidia lacked but would need to become a complete data-center computing company. The acquisition extended Nvidia beyond processors into the systems connecting large numbers of processors.

Huang identified a fundamental difference between conventional cloud computing and AI training. Conventional hyperscale computing often virtualizes one machine among many users, whereas a large AI training job may distribute one workload across millions of processors.

Huang described AI training as nearly the inverse of traditional hyperscale computing. The shift changes the importance of interprocessor communication, making networking performance central to the effectiveness of the complete system.

Conventional Ethernet was adequate for many established data-center workloads but was not necessarily optimal for distributed AI training. Huang cited applications such as Hadoop and search queries as suited to commodity networking, while large models spread across many processors require different networking characteristics.

Mellanox's high-performance networking and InfiniBand expertise matched Nvidia's supercomputing requirements. Huang believed technologies developed for tightly connected supercomputers were especially well suited to distributed AI workloads.

Nvidia already had substantial working relationships with Mellanox in high-performance computing before the acquisition. This familiarity strengthened Huang's confidence in the strategic combination.

Huang placed considerable value on Mellanox's Israeli engineering team. He described the team as world-class and said Nvidia had approximately 3,200 employees in Israel at the time of the interview.

Huang considers acquiring Mellanox one of the best strategic decisions of his career. The hosts reported hearing similarly strong assessments from people interviewed during their research into Nvidia.

Huang's acquisition strategy reinforced his preference for preparing for future markets rather than reacting after demand becomes obvious. Mellanox placed Nvidia closer to the networking requirements of emerging AI systems and gave it the capability to benefit as those requirements expanded.

Nvidia deliberately targets what Huang calls “zero-billion-dollar markets.” These are markets that do not yet exist commercially but that Nvidia believes will eventually become large enough to justify sustained investment.

Huang prefers entering markets before conventional competitors recognize their potential. He argued that when a new industry finally emerges, a company that has been developing the necessary technology for years may face few competitors with equivalent capabilities.

Nvidia entered automotive computing because Huang predicted cars would become increasingly software-defined. He reasoned that software-heavy vehicles would eventually require exceptionally powerful onboard computers.

Automotive executives initially questioned the reliability of computer-driven vehicles. Huang recalled one chief technology officer warning that cars could not tolerate the equivalent of a Windows blue screen of death, illustrating resistance to the idea of software-defined automobiles.

Huang maintained that reliability concerns did not invalidate the long-term transition to software-defined cars. Looking back roughly 15 years after Nvidia entered the automotive field, he believed the industry's development had substantially supported his original prediction.

Huang identified PC gaming, professional workstations, supercomputing and AI as examples of Nvidia entering markets before their full potential became obvious. He argued that Nvidia repeatedly invested in new computing categories years before they became established businesses.

Huang claimed Nvidia technology had become pervasive in professional design workstations. He presented this as an example of how redefining a computing category early could ultimately produce a dominant market position.

Nvidia had been investing heavily in Omniverse despite its small commercial market. Huang described it as another zero-billion-dollar opportunity and estimated that it had roughly 40 customers at the time; Amazon and BMW were mentioned during the exchange.

Huang characterized Nvidia's usual market-development horizon as approximately a decade. The company may invest for many years before an anticipated application develops into a meaningful revenue opportunity.

Huang believes companies that create new markets can develop strong competitive advantages by supporting an entire ecosystem. If developers, customers and complementary businesses build around a company's technology, their combined investments make the platform more difficult to displace.

Huang prefers building networks of successful partners to thinking defensively about economic moats. He argued that the goal should be enabling other businesses to succeed alongside Nvidia rather than concentrating on protecting an isolated competitive position.

Nvidia was conceived internally as a platform company from its beginning, even when it appeared externally to be a graphics-chip manufacturer. Huang said its original Unified Device Architecture was intended to describe a broader family of accelerators rather than only individual graphics products.

Nvidia's original business plan was effectively to create a game console inside the PC. By providing common hardware and programming interfaces, Nvidia hoped developers would create software specifically for its platform.

Developer relations was one of Nvidia's earliest organizational priorities. Huang said one of the company's first employees focused on working with developers because Nvidia understood that its hardware needed a software ecosystem.

Nvidia originally developed an architecture and programming approach called DirectNV. The company hoped developers would program directly for Nvidia hardware, creating a platform relationship before Microsoft's DirectX became the dominant interface.

Microsoft's introduction of DirectX in the mid-1990s forced Nvidia to reconsider its platform strategy. Microsoft's existing developer relationships gave it an advantage that Nvidia could not realistically overcome through its original proprietary approach.

Huang acknowledged that Microsoft handled parts of the graphics-platform transition correctly while Nvidia made serious mistakes. He resisted the suggestion that Nvidia had been an equal competitor to Microsoft, emphasizing the enormous difference between their positions at the time.

CUDA eventually allowed Nvidia to build a computing platform on top of a graphics business already supported by DirectX and OpenGL. Graphics applications paid for the GPUs and created an installed hardware base that could subsequently support general-purpose computing.

Huang made architectural compatibility across Nvidia's accelerator platform a nonnegotiable principle. He argued that developers cannot rely on a computing platform if each hardware generation requires a fundamentally different programming approach.

Huang estimated Nvidia had approximately 250 million to 300 million active CUDA GPUs installed worldwide at the time. He emphasized that architectural compatibility across the platform was central to the value of that installed base.

Huang described architectural compatibility as Nvidia's one absolute rule. He said almost everything else inside the company could be negotiated, but breaking compatibility would undermine the developer platform Nvidia had spent decades building.

Huang does not believe successful companies can be explained through a simple checklist of correct decisions. He said founders can make many intelligent decisions and still fail, or make numerous mistakes and succeed when favorable circumstances and market conditions come together.

Huang credited both preparation and external circumstances for Nvidia's survival. The RIVA 128 reflected deliberate engineering and strategic choices, but the rapid emergence of demand for accelerated 3D gaming also contributed to its commercial success.

John Carmack initially favored software rendering rather than dedicated graphics acceleration. Huang recalled that Carmack's work on Doom achieved extraordinary performance using CPUs and highly optimized software.

Doom demonstrated how much performance could be extracted from conventional processors by an exceptional programmer. Huang acknowledged the quality of Carmack's implementation while emphasizing that such performance required unusually specialized engineering talent.

Carmack's move toward OpenGL and hardware acceleration for Quake helped validate Nvidia's original graphics thesis. Huang said that without this shift, it was unclear what consumer application would have created the necessary early demand for Nvidia's products.

Huang credited John Carmack and Tim Sweeney with creating the first major consumer applications for accelerated 3D graphics. Quake and Unreal helped establish the commercial market for dedicated 3D graphics hardware, and Huang expressed gratitude to both developers.

Huang believes developers create strategic opportunities that hardware manufacturers cannot invent alone. A strong developer ecosystem exposes a platform company to ideas and applications originating outside its own engineering organization.

Huang identified AI safety as a necessary condition for broader deployment. He distinguished between physical safety in robotics and autonomous vehicles and information-related risks in generative AI.

Nvidia has invested in functional and active safety for robotics and autonomous driving. Huang said an important design question is determining when humans must remain involved in a system's decisions and when automation can operate safely without continuous intervention.

Huang identified bias, misinformation and creators' rights as major concerns for information-generating AI. He argued that developers must consider the reliability and origins of AI outputs rather than treating every available dataset as unrestricted material.

Nvidia partnered with Getty Images and Shutterstock as an alternative to relying exclusively on indiscriminate internet scraping. Huang presented these relationships as examples of building generative AI systems around commercially appropriate access to creative material.

Huang opposed allowing increasingly autonomous AI systems to retrain and modify themselves freely after deployment. He advocated collecting and curating data, training controlled model versions, testing them and validating their behavior before releasing updates.

Huang expects humans to remain involved in supervising advanced AI for a substantial period. He considered human oversight especially important as AI systems acquire greater independence and the ability to take consequential actions.

Huang cited aviation as a useful model for designing safe autonomous systems. Autopilot, multiple pilots, air-traffic control, redundancy and diverse safety mechanisms demonstrate how automation can operate within structured systems of human oversight.

Huang predicted AI would probably create more jobs overall than it destroys, although he acknowledged uncertainty about timing. His reasoning was that productivity improvements increase profitability and economic activity, allowing businesses to pursue additional opportunities.

Huang argued that productivity gains do not automatically mean companies will need fewer employees. A business with no plans to expand might reduce staffing, but a business with new ideas can use higher profits to hire people and enter additional markets.

Huang identified drug discovery, transportation, retail, entertainment and technology as industries with substantial room for expansion. He believed these sectors contain enough unmet needs and unrealized ideas to absorb additional productive capacity.

Huang acknowledged that net job growth would not protect every individual worker from displacement. Even if AI increases total employment, particular people may lose jobs as companies change their operations and required skills.

Huang warned that workers may be displaced by other people who use AI rather than by AI operating independently. His practical advice was for individuals to learn AI tools and use them to improve their own productivity.

Huang similarly encouraged companies to adopt AI to improve productivity and profitability. He argued that stronger businesses could subsequently expand their operations and employ more people.

Huang expected AI to be particularly useful in industries suffering from labor shortages. Automation could help constrained businesses increase output and resume growth despite difficulty finding sufficient workers.

The hosts connected Huang's employment argument to what they called the Moritz corollary to Moore's law. The idea, associated with Sequoia investor Michael Moritz, is that cheaper computing expands the number and size of markets technology companies can address.

Michael Moritz reportedly recognized this principle while studying Sequoia's historical investment returns. Concerned about how future funds could outperform an earlier fund containing Cisco, he concluded that falling computing costs would open increasingly large economic opportunities.

Huang agreed that greater productivity generally leads humanity to attempt more rather than settle for less. He attributed this to continuing human ambition and the availability of new problems worth solving.

Huang said he had never read a science-fiction book despite his enthusiasm for futuristic technology. His favorite television series was Star Trek, which also influenced the naming of conference rooms at Nvidia.

Huang has strong personal memories associated with his Toyota Supra. He recalled getting engaged to Lori around Christmas and then totaling his new Supra during their drive home, an accident he said was neither his fault nor the car's.

Huang was being driven in a Mercedes-Benz EQS at the time of the interview. He cited security and other practical reasons for no longer driving himself regularly and said the EQS was an excellent vehicle.

Huang identified Clayton Christensen's business books as among his favorites. He valued their intuitive explanations and their usefulness in thinking about technological disruption and business strategy.

Huang also praised Andy Grove's books. He said he had read many business books over the years and considered Grove's work particularly valuable.

Don Valentine was one of Nvidia's formative early investors. Huang remembered Valentine as grumpy but endearing, with a demanding personality that concealed genuine concern for the companies he supported.

Valentine jokingly warned Huang that he would kill him if he lost the investment. Huang recalled the remark from the period when Valentine decided to back Nvidia, describing it as characteristic of the investor's blunt manner.

Valentine maintained a personal connection with Nvidia after investing. When newspapers such as the Mercury News published favorable articles about the company, Valentine sometimes wrote short congratulatory messages directly on the clippings and mailed them to Huang.

Huang said his younger self would have disagreed with his present belief that there is enough time to accomplish important work. He now believes effective prioritization and deliberate sacrifices make time manageable, even when responsibilities are enormous.

Huang warned against allowing Outlook or a calendar to control personal priorities. He believes leaders should choose what deserves their attention instead of automatically letting scheduled meetings determine how they spend their days.

Huang arrived late to the interview because he had been absorbed in another activity. He used the incident to illustrate how he prioritizes important work rather than treating every calendar commitment as the ultimate authority.

Huang's greatest professional fear is letting Nvidia's employees down. He said employees join a company because they believe in its ambitions and often adopt those ambitions as their own.

Huang feels responsible for employees' careers, families and opportunities to build successful lives. He wants the people who helped create Nvidia to experience the financial and professional rewards that he has enjoyed.

Huang described himself as someone who rarely changes jobs or abandons commitments. He said that without Chris and Curtis persuading him to start Nvidia, he believes he would still be working at LSI Logic.

Huang retains considerable affection for LSI Logic, where he worked before founding Nvidia. He described himself as deeply committed to the company and said many colleagues from that period remain his friends decades later.

Huang began his semiconductor career at AMD before moving to LSI Logic. His early experience exposed him to different approaches to chip design and shaped the philosophy he later brought to Nvidia.

LSI Logic transformed Huang's understanding of semiconductor engineering by demonstrating the power of higher-level design abstractions. He learned that transistors, logic gates and chip functions could be described through higher-level languages rather than designed entirely through low-level manual processes.

Huang became fascinated with high-level chip design at approximately 21 years old. He saw how compilers, optimization tools and software abstractions could dramatically increase the productivity of semiconductor engineers.

Huang regarded LSI Logic's contribution to computer design as revolutionary. He believed its approach changed how chips and systems could be built and helped enable the broader evolution of the computing industry.

Huang connected LSI Logic's work to subsequent developments at Synopsys. He regarded the rise of sophisticated electronic design automation as a crucial part of the transition toward more abstract, productive hardware development.

Harvey Jones played an important role in Huang's exposure to high-level design. Huang credited Jones with the terminology and ideas surrounding that approach, said he knew him from the early Synopsys era and identified him as a member of Nvidia's board.

Huang's time at LSI Logic placed him in contact with influential technology figures. He mentioned relationships involving Chris, Curtis, Andy Bechtolsheim and Jon Rubinstein, among others, as important connections formed during that period.

Huang now wants similar abstraction breakthroughs to transform digital biology. He suggested that biological systems could eventually be represented and manipulated through higher-level languages, potentially using large language models.

Huang sees a recurring pattern in technological progress: making complex systems programmable at higher levels of abstraction. He connected advances in chip design, software development, machine learning and his ambitions for digital biology through this principle.

Huang admitted that he did not know how to write a conventional business plan when Nvidia was founded. He later concluded that elaborate financial projections are often less valuable than founders believe because early forecasts are so uncertain.

Huang believes a useful business plan should clearly identify the unmet need a company intends to serve. He emphasized understanding what problem will matter, why the need is likely to emerge and how the company plans to address it.

Huang believes a startup's proposed solution must be sufficiently difficult to reproduce. If a market becomes attractive and competitors can easily duplicate the product, the original company may lose the opportunity it identified.

Huang distinguished fundamental strategic insight from learnable business skills. Pricing, product positioning, sales and go-to-market execution matter, but he considered them easier to learn than identifying a valuable future market and developing a genuinely defensible solution.

LSI Logic chief executive Wilfred Corrigan personally helped Huang secure early investment. Huang said Corrigan was pleased with his work and called Don Valentine to recommend investing in him, giving Nvidia an important introduction.

Sequoia Capital became one of Nvidia's original venture investors. Huang credited the firm's early backing as important to the company's creation and longevity.

Sutter Hill Ventures was another founding venture investor that maintained a long relationship with Nvidia. Huang emphasized the unusual continuity of the company's original financial supporters.

Tench Coxe and Mark Stevens continued serving on Nvidia's board decades after the company's founding. Huang identified their continued involvement, representing the early Sutter Hill and Sequoia relationships, as an unusually enduring arrangement.

Huang said Nvidia's longstanding investors contributed more than money. He credited them with advice, encouragement, institutional continuity and emotional support during periods when the company faced severe uncertainty.

Huang believes Nvidia became an exceptionally successful investment for its early venture backers. He suggested that it may have been among Sequoia's best investments, while emphasizing the decades-long relationship rather than only the financial returns.

Asked what company he would start if he were 30 years old again in 2023, Huang replied that he would not start one. His answer reflected the difficulty of building Nvidia rather than a belief that new technological opportunities had disappeared.

Huang said building Nvidia turned out to be approximately a million times harder than he expected. He described the accumulated challenges, vulnerability, embarrassment, shame and suffering as far greater than anything he could have anticipated in 1993.

Huang believes entrepreneurial inexperience can be an advantage because founders do not fully understand the difficulty ahead. He suggested that if aspiring founders knew everything they would endure, many would decide not to begin.

Huang continues to motivate himself by asking, “How hard can it be?” He described this as a necessary mental strategy for taking on enormous challenges despite knowing, from experience, that they will probably be far harder than expected.

Huang applies that same attitude to current initiatives, including Omniverse. Even after decades of experience, he deliberately approaches ambitious new projects with the confidence needed to attempt them.

Huang indicated that he had no immediate interest in retiring. He said he continued to enjoy his work immensely and believed he was still contributing value to Nvidia.

Huang identified family, friends and loyal colleagues as essential to surviving the emotional demands of entrepreneurship. He argued that unwavering personal support becomes especially important when a company faces prolonged uncertainty or public failure.

Several Nvidia colleagues had remained with Huang for approximately 30 years. He specifically mentioned Chris, Jeff Fisher and Dwight, along with other long-serving colleagues, as people who had supported the company since its early period.

Other colleagues had remained at Nvidia for more than 25 years. Huang mentioned Jonah and Brian among the experienced employees whose continued commitment contributed to his sense of stability and support.

Huang said his closest colleagues had never given up on Nvidia or on him personally. He regarded that loyalty as one of the most important reasons he could endure the company's repeated crises.

Huang also credited his family with unconditional support through Nvidia's difficult periods. He emphasized the importance of returning home to people who remain proud of the founder and the company regardless of short-term performance.

Huang credited early supporters including Jim Gaither, Tench Coxe, Mark Stevens, Harvey Jones and Bill Miller for remaining committed to Nvidia. He described their loyalty and encouragement as essential components of the support system surrounding the company.

Nvidia endured multiple public-market collapses exceeding 80%, according to the interviewers. They estimated that the company had experienced two or possibly three such drawdowns during its history, illustrating the extraordinary financial volatility surrounding its development.

Huang recalled Nvidia trading at a market capitalization of approximately $2 billion to $3 billion during one difficult period. He associated that period with the company's expensive commitment to CUDA and related long-term investments.

Huang described severe share-price declines as personally embarrassing and emotionally difficult. Employees, investors and outsiders questioned the company's strategy, while he had to continue defending decisions whose commercial results might remain uncertain for years.

Huang distinguished employee doubt from disloyalty. He said employees naturally had questions during difficult periods, even when they continued to support him and remained committed to Nvidia's strategy.

Huang acknowledged that many forces affecting Nvidia's stock price were outside his control. Nevertheless, he said a major market decline was difficult to endure because public criticism and financial losses made the consequences of long-term strategic decisions highly visible.

Huang believes founders need unusually strong conviction to survive prolonged periods of market disapproval. He argued that without a deep belief in the long-term opportunity, the financial and emotional pressure becomes almost impossible to sustain.

Huang rejected earlier predictions that Nvidia could never become larger than a $1 billion company. He said those predictions reflected an overly narrow understanding of its potential market rather than a permanent limitation on Nvidia's technology.

Huang believes a company's ultimate scale depends on the market it actually serves, not the category in which it originally operated. Limiting Nvidia's opportunity to semiconductor sales ignored the possibility of expanding into computing platforms, infrastructure and intelligent services.

Huang sees AI as the industrial production of intelligence and economically useful work. He argued that this is a far larger category than producing chips because the output is not merely hardware but the performance of tasks that otherwise require human or organizational effort.

Huang estimated the economic opportunity associated with producing intelligence and useful work in the trillions of dollars. He believed this could support technology companies substantially larger than those built around traditional semiconductor markets.

Huang illustrated the difference between hardware markets and AI-service markets through automobiles. A chipmaker's opportunity depends on the number of cars and chips each vehicle requires, while an AI system capable of providing chauffeur-like assistance could capture value from the transportation service itself.

Huang suggested that making autonomous driving assistance available to every car owner would create a much larger market than selling vehicle processors. The economic value would increasingly come from providing driving capabilities rather than the physical computing equipment alone.

Huang estimated that expanding from semiconductor products into AI-enabled capabilities could increase Nvidia's addressable opportunity by roughly a thousandfold. He presented the figure as an illustration of the scale difference between selling hardware and serving the broader market for intelligent work.

Huang expects some future technology companies to become much larger than earlier generations because their products address larger economic activities. In his view, AI changes the potential scale of technology businesses by allowing software and computing infrastructure to participate directly in work traditionally performed by people.

Huang's central long-term principle is to identify a future economic need, develop the capabilities to serve it and remain positioned for the market's emergence. Nvidia's progression from graphics to CUDA, cloud computing, networking and AI illustrates how earlier investments can prepare a company for opportunities that are not yet commercially obvious.