20VC: Micron Will Be More Valuable Than Meta | How Export Controls Helped Not Hurt China | Power is the Bottleneck to AI | Why Dario Has Done a Disservice to AI with his Labour Replacement Messaging with Aravind Srinivas, Founder @ Perplexity

Aravind Srinivas discusses Perplexity's 'attack, attack, attack' ethos and its significant impact on Google's search interface.

Aravind Srinivas, CEO of Perplexity, explains his 'thrill of winning' motivation, stemming from a humble background in India where his family's ultimate ambition was simply to secure an engineering job at Google. He believes he has 'nothing to lose' and advocates for an 'attack, attack, attack' mentality, always preferring offense over defense. This mindset drives his entrepreneurial approach.

This aggressive ethos translated into Perplexity's early strategy, with Srinivas personally engaging in public comparisons between Perplexity and Google on social media. While he has since adopted a more measured tone, he does not regret the initial boldness. He views that direct comparison as less relevant now, as Perplexity has expanded beyond its initial answer engine to develop numerous other products like agents and deep research tools.

Srinivas asserts that Perplexity has influenced Google's core search interface more profoundly than internal product managers. He highlights Google's 'AI mode,' noting its striking resemblance to Perplexity's design, including font, citations, inline text, hyperlinks, and suggested follow-ups. He anticipated this adaptation from Google and views it as both good and bad, recognizing it was only a matter of time before the tech giant adopted similar features.

attack, attack, attack. That's, that's, that's my moto.

AI's Frontier is Doing Work, Not Just Answering Questions

The true frontier of AI lies beyond simply answering questions. The focus is shifting towards AI agents that actively perform work for users. There is a growing demand for sophisticated research reports and systems that can autonomously complete tasks, rather than just provide information in a traditional search format.

This shift implies that the underlying AI model itself is not the ultimate product; instead, it functions more as a lead generator for advanced 'frontier products.' The real value comes from how these models are orchestrated and applied to solve complex problems, transforming them into practical tools that execute specific jobs.

The AI landscape is characterized by rapid change, making it impossible for any company to rest on its laurels. Even market leaders like Anthropic or OpenAI, despite their current dominance, must continuously innovate. The field demands constant evolution, as today's leading players can quickly find themselves fighting from behind if they don't adapt.

The frontier in AI isn't about answering questions anymore, it's about actually going and doing work for you.

Advertising is Not a Good Fit for AI Chat

AI chat interfaces are fundamentally incompatible with advertising, according to Aravind Srinivas. Unlike traditional search engines or social media, where users might be in a discovery or browsing mode, AI chat is often used for seeking objective answers. This core difference in user intent makes it difficult to integrate advertising effectively.

Major advertisers like Amazon, Booking.com, and Expedia, who spend billions on platforms like Google, rely on users being in a discovery phase, searching for options to compare. Similarly, direct-to-consumer fashion brands advertise heavily on platforms like Instagram, where users are often browsing casually. AI chat interfaces do not capture this subjective, exploration-based user behavior.

The primary reason AI chat struggles with advertising is that it fundamentally corrupts user trust. When users engage with an AI, they expect accurate and objective information. Introducing advertisements can undermine this trust, making the AI's responses appear biased or influenced by commercial interests, which goes against the expectation of an unbiased answer engine.

It also fundamentally corrupts the trust that people have when they go into a product and they want the accurate answer.

Orchestration Systems and Agent Harnesses Convert Model Intelligence into Value

The idea that the "money is in the frontier" (models) is debatable, as we often overestimate the importance of frontier models for performing basic work. This challenges the common assumption that the most advanced models are inherently where the most value lies.

Greg Brockman of OpenAI has stated that "the model is no longer the product," a notable perspective given his leadership role at a frontier lab. This viewpoint contrasts with some companies that continue to position the model itself as the primary product offering.

Instead, the real value is derived from orchestration systems. These systems take a base model and pair it with an agent harness. Examples like Perplexity Computer illustrate this approach.

An agent harness is crucial, as it dictates how an agent loop should run by defining all the necessary skills, sub-agents, connectors, and tools it can access. Without this harness, the intrinsic intelligence embedded within a model cannot be effectively captured and converted into valuable, usable output.

Without the harness, you don't necessarily capture and convert the intrinsic intelligence in the model into valuable output tokens.

Perplexity Differentiates by Orchestrating Competing AI Models for Superior User Value

Simply reselling raw model tokens does not constitute a sustainable business; true value lies in grounding models with valuable context, orchestrating them with a robust agent harness, and connecting them to relevant tools and connectors to provide a unified user experience.

Perplexity differentiates itself by not only orchestrating across tools, files, and connectors, but also across competing AI models like GPT and Claude within its own system. This is a unique advantage that model builders like Anthropic or OpenAI cannot claim for their own competing models.

This multi-model orchestration strategy aims to maximize "token value per watt per user," which is identified as the most important metric in AI. By providing the most valuable output tokens with the least amount of power expended, Perplexity generates greater value for end-users and achieves stronger pricing power.

The core idea is that owning the interface that generates highly valuable AI output tokens is crucial, even more so than just building models, which are becoming a utility. This focus ensures sustained value creation at the frontier of AI.

The most important metric in AI is token value per watt per user.

Power Users Propel the AI Token Economy Through Continuous Agents

The AI product layer is not about achieving a billion users; instead, a select group of 'power users' is driving the current AI token economy. These users demonstrate incredibly high spend, with some engineers at major companies like Meta reportedly spending ten million dollars annually on coding tools that leverage AI. Similarly, a single user on Perplexity might spend upwards of ten thousand dollars a month, not on wasteful activities, but to power their business operations.

These power users are not using AI for simple, one-off tasks like deep research. Their significant consumption stems from deploying AI in sophisticated, continuous ways, often described as 'agent loops' or 'cron jobs.' They establish multi-agent hierarchies and complex software architectures where AI continuously monitors events, triggers actions, provides alerts, and automates workflows.

The key differentiator for heavy AI agent users is the implementation of repetitive, always-on tasks. This includes scenarios like identifying the root cause of latency spikes, analyzing inbound emails, or performing continuous monitoring and analysis. This shift from discrete tasks to persistent, automated processes represents the true frontier of AI application, driving substantial economic activity among a niche group of advanced users.

The single biggest differentiation between those who use agents a lot and those who don't is whether they run repetitive cron jobs.

Decreasing Token Costs and the Rise of Autonomous Software Engineers

The discussion considers the percentage of developer salaries being spent on AI tokens, referencing Salesforce's $300 million investment in Anthropic. While a current estimate puts this at 3.8% of developer salaries, it is difficult to predict future spend due to anticipated changes in token costs.

A key prediction is that AI token costs will decrease. Within 12 months, open-source models are expected to become as capable as current frontier models, but at potentially one-tenth of the inference cost. This will significantly alter the economics of AI adoption for developers.

Despite the availability of cheaper open-source alternatives, companies will continue to invest in the latest "frontier" AI models. This is likened to hiring a top-tier engineer like Jeff Dean; the premium is paid for cutting-edge capabilities that enable new, currently unconceived applications, pushing beyond what is currently possible.

This technological evolution is projected to lead to the emergence of completely autonomous software engineers, where AI agents independently develop software. This future state will likely result in companies operating with smaller human software teams, heavily augmented by AI, fostering the creation of more businesses.

My prediction would be agents that are like completely autonomous software engineers.

AI's Ultimate Purpose: Solving Grand Challenges and Understanding the Unknown

The frontier of AI is evolving towards applications with massive, indirect human impact rather than just user count. This includes designing advanced chips, discovering new drugs, developing robotics, and finding cures for diseases like cancer. Companies are already positioning for this shift, with examples like Anthropic acquiring a wet lab, suggesting a future where AI integrates diverse data beyond traditional digital tokens.

There is no mathematical limit to the economic value that advanced AI systems, like AGI or ASI, can create. As posited by Elon Musk, a post-AGI economy could produce an abundance of energy and labor, fundamentally changing the meaning and value of money.

Humanity will not run out of grand challenges or frontier problems to solve, largely because our fundamental purpose has always been to understand the unknown. This innate curiosity drives exploration into areas like subatomic particles, quantum physics, black holes, and the origins of the universe, ensuring a continuous supply of complex problems for AI to address.

I don't think we'll run out of things to solve at the frontier. I think we're always gonna be creating, like, like why did people even- We want to understand the universe.

Hybrid AI orchestration is crucial for affordable, private 24/7 agents, benefiting from the entire AI stack.

Running continuous 24/7 AI agents presents a significant "orchestration problem" due to competing objectives of intelligence, accuracy, privacy, and cost. While large data centers can maximize intelligence and accuracy, they lead to high centralization costs and privacy concerns. Conversely, purely local solutions offer privacy and cost benefits but may lack frontier intelligence.

The solution involves a hybrid approach that orchestrates across both local and server-side models, grounded in valuable personal context. This means utilizing local compute on devices for sensitive data and general tasks, and only engaging powerful server-side frontier models when necessary. A world-class harness that intelligently routes tasks can make even average models perform well, ensuring 24/7 AI functionality without prohibitive costs or privacy risks.

Companies positioned as orchestrators, rather than solely frontier model builders, are best placed to capture long-term economic value by maximizing "token value per watt per user." Perplexity, for example, thrives because its product improves with advancements at any layer of the AI stack. Better chips from NVIDIA, improved models from Anthropic, or enhanced devices from Apple all contribute positively to Perplexity's system and user experience.

The broader challenge for AI infrastructure lies in the slow physical build-out time for data centers, particularly securing adequate power. New, more powerful chip generations like Blackwell and Rubens will demand even more resources, exacerbating this bottleneck and highlighting the value of efficient orchestration and local compute to manage demand.

The one whose product or business benefits from other people's progress at any layer of the stack.

AI Infrastructure Bottlenecks Shift Value to Memory and CPU Suppliers

Companies with strong vertical integration in AI infrastructure, managing complex software layers, compute, and cooling to produce frontier output tokens, are being valued highly by the market. This explains why infrastructure companies often have higher P/E ratios than even large tech companies like Meta, despite Meta's significant infrastructure investments. For Meta, CapEx spend primarily correlates with ad accuracy, yielding a modest revenue bump, whereas for dedicated AI providers, it directly fuels their core products.

A significant bottleneck in AI is memory, specifically High Bandwidth Memory (HBMs). This scarcity has driven HBM prices up fivefold, directly benefiting suppliers like Micron. It's suggested that Micron's market valuation could potentially surpass Meta's within the next 6-12 months, as memory continues to be a critical constraint in AI model performance and deployment.

Beyond memory, CPUs have emerged as another bottleneck, largely due to the rise of AI agents. Agent loops and harnesses primarily run on CPUs, handling tasks like data munging, file downloads, and website hosting, even as GPUs generate the tokens for frontier models. This increased demand for enterprise CPUs is creating a new competitive advantage for companies like AMD and Intel.

Building and operating data centers is an operationally intensive and challenging endeavor. Success hinges on resourcefulness in securing power from low-cost natural resource areas, fast deployment times, and reliable service. The difficulty in replicating this work, including securing supply and planning ahead, makes data center infrastructure a valuable layer, where even established AI firms might struggle compared to specialized providers.

Whatever is the bottleneck will command the price.

Data center businesses require software orchestration and depend on open-source AI models for long-term sustainability.

Merely renting GPU server racks offers limited value and low margins due to the substantial operational complexities involved in data center management, including securing permits, managing power, dealing with supply chain bottlenecks, and handling physical issues. Companies focused solely on capacity rental struggle with the high total cost of operations.

To achieve sustainable profitability, data center businesses must develop software orchestration on top of their physical infrastructure, mirroring the strategy of companies like AWS. This allows them to capture higher software margins. Examples include Nebia, Fireworks, and BaseTen, which focus on AI model inference and hosting open-source models.

The business model for these AI inference and hosted model companies critically depends on the continued strength and competitiveness of open-source AI models. If the performance gap between open-source and frontier models widens significantly (e.g., beyond 12-18 months), these companies might lose their market differentiator and be reduced to renting capacity solely to a few dominant providers.

While model selection and routing products exist, their core business value often extends beyond simple model routing. For instance, OpenRouter's main appeal lies in providing model fallbacks, ensuring API key rate limits, and securing reliable capacity, addressing common developer challenges.

If open source models stop to actually be good, where the gap between them and the frontier is like more than twelve months or like fifteen months, eighteen months, then I don't- I don't think these companies really have a business model.

AI Model Routers Prioritize Reliable Token Supply Over Cost Optimization

AI model routers like OpenRouter primarily solve an infrastructure problem by providing a reliable token supply from various providers, including Bedrock, Azure, and OpenAI. Their core offering ensures consistent access to AI model endpoints for developers.

The value proposition of these routing services is not to optimize costs by dynamically choosing between different models, such as GPT or Claude, for a given prompt. Instead, they address crucial operational needs, like ensuring API tokens do not go to specific regions or managing multiple inference providers when developers lack the bandwidth to do so.

While this business model, which often involves charging the listing price and retaining the difference as margin, may not yield high gross margins, the demand for such reliability is evident. OpenRouter, for example, routes trillions of tokens monthly, highlighting the significant need for guaranteed token supply in the AI ecosystem.

Looking three years ahead, power is anticipated to remain a significant bottleneck for AI infrastructure. Aravind Srinivas suggests that public resistance to building new data centers stems less from concerns about water or power consumption—which he claims are often misrepresented—and more from deeper societal anxieties about job losses and increasing wealth inequality.

It's routing, not at the level of like, oh, like deciding which model is cheap for what task, it's more like a reliable token supply.

Public resistance hinders data center development for AI growth.

Public resistance to data center development stems from various apprehensions, including environmental concerns, fears about rising grid prices, and resentment over wealth inequality. These sentiments are channeled in different ways, leading to significant opposition against the necessary infrastructure for advancing AI.

This resistance directly impacts the development of crucial AI infrastructure. Currently, 40% of planned data centers are not being built because of public opposition, creating a substantial power bottleneck. This physical barrier slows down the expansion required to support growing AI demands.

Developing data centers involves tackling complex physical challenges beyond typical software problems. It requires managing supply chains, securing permits, ensuring power supply, and reducing lead times. These issues demand considerable capital, strong connections, and often political assistance to overcome, making them much harder to solve than building a SaaS application or improving marketing.

While some countries may see this as an opportunity and offer more favorable regulations for data center construction, the fundamental challenges remain. The difficulty in navigating public sentiment and the intricate logistical requirements mean that public resistance will likely continue to be a significant bottleneck for AI progress.

right now forty out of a hundred aren't being developed because of public resistance.

US Export Controls Spur China's Vertically Integrated AI Development

US export controls, specifically on high-end GPUs like Nvidia and high-bandwidth memory (HBM), are inadvertently pushing Chinese AI companies to develop highly efficient, vertically integrated architectures. DeepSeek, for example, is building its models using the Huawei stack, not Nvidia, leading to significant innovations born out of necessity.

DeepSeek's architectural advancements focus on memory efficiency, such as optimizing the KV cache to be small enough for SSD hosting, thereby reducing the need for HBM during inference. They are also developing distinct architectures for inference and storage, and have made improvements to attention layers and training algorithms to minimize interconnect capacity usage. This integrated approach encompasses their entire stack, from model architecture to hardware, chips, and fabs.

While these export controls offer a short-term advantage for the US, potentially creating a 12-month gap in frontier model development, there is a risk of converting China into a more potent long-term competitor. By forcing Chinese companies to innovate at the physical layer, the controls enable them to develop robust domestic capabilities. China also holds advantages in building data centers rapidly, facing fewer issues with power, permits, labor, and expertise.

Many believe the US still underestimates China's capabilities, especially if AI involves not just digital aspects but also physical infrastructure like fabs, robots, chips, and energy management. In response, there's significant investment in US fabs, including TSMC's Arizona facility, Intel's efforts, and Elon's TerraFab. To maintain competitiveness, the US needs to seriously invest in physical infrastructure and counter misinformation about facilities like data centers.

But there is a chance that because of that, they now get really good at the physical layer. And one advantage they have is they can actually build data centers a lot, lot faster.

AI Empowers Small Teams to Build Efficient, Multi-Billion Dollar Companies

Aravind Srinivas strongly criticizes the prevalent "AI doom" messaging, particularly the fearmongering about widespread job loss. He views this narrative as contradictory, especially from entities that also express difficulty in building data centers quickly. He argues that there needs to be consistent communication that focuses on the positive potential of AI.

Instead of fear, Srinivas advocates for a narrative centered on how AI enables a new wave of entrepreneurship. He envisions individuals and small teams, potentially as few as 20-30 people, building highly efficient multi-billion dollar companies. This shift could propel trillions of dollars in new GDP by allowing more people to pursue their ideas using AI agents to perform tasks traditionally requiring a larger workforce.

Companies like Perplexity are demonstrating this efficiency, operating with 400 people while aiming for significant valuations. This model suggests that even smaller teams of 40 people could build billion or two-billion dollar companies. Perplexity supports this vision with initiatives like the "billion dollar build," offering $1 million in compute credits to credible teams, mirroring efforts like Sam Altman's tokens for YC companies.

Srinivas believes fostering thousands of such companies, even those valued in the hundreds of millions, would dramatically increase new GDP and spread wealth more broadly. He prefers a future with many entrepreneurial groups building valuable companies over a few monolithic corporations, encouraging more people to become entrepreneurs.

Let's talk about how to enable that, let's talk about how to build that and create a more positive future together instead of, oh, like ninety percent of the jobs are gonna be gone.

Embracing AI as a Non-Native and Managing Token Budgets with Hybrid Inference

Aravind Srinivas encourages individuals who are not AI-native to explore artificial intelligence driven by curiosity, rather than solely using it to enhance existing tasks they might find unfulfilling. He clarifies his earlier stance on job dissatisfaction, emphasizing that many people genuinely dislike their work, regardless of their economic standing. AI now presents an unprecedented opportunity for individuals to build significant companies with minimal initial resources, potentially allowing a small team of one or two friends to create a billion-dollar venture.

The conversation also addresses the challenges of sustained demand for AI resources, particularly concerns about "token maxing" and the rising costs of inference. Srinivas advocates for a hybrid inference approach, where some computing runs locally without token costs, complementing server-side solutions. The expectation is that future companies will rely on orchestrators to manage complex token budgets, allocating resources across different models and tasks, rather than handling this intricate process themselves.

While Google possesses significant advantages in becoming a low-cost token producer due to its full-stack ownership, Srinivas notes their current lag in developing advanced coding models. The discussion highlights the rapid increase in "agent traffic" surpassing human traffic online, which he believes indicates a future with greater personal agency. This shift could fundamentally alter how people interact with the internet, potentially impacting traditional website design and advertising models.

For the first time in history, you can get started on an idea with like one or two other friends and maybe have a real genuine shot at building a billion dollar company.

The Internet Will Bifurcate into Objective and Subjective AI-Driven Realms

The internet is expected to split into two distinct realms: objective and subjective. Transactions and decisions based on objective judgments, such as purchasing a microphone based on technical criteria, will increasingly be handled by AI agents. This marks a significant disruption where agents will dominate objective decision-making.

Conversely, transactions involving subjective judgments, like selecting furniture based on room aesthetics, will likely remain ad-based. The human element of personal preference and taste will continue to drive these subjective choices, keeping traditional advertising relevant in these areas.

This shift implies a new essential skill. With AI agents handling objective tasks, the ability to formulate and ask better questions becomes paramount. Instead of executing tasks, the focus shifts to defining the right problems and leveraging agent capabilities effectively.

A key question to consider is: assuming you have a large amount of agency available, such as 10,000 agents and ample compute credits, what would you do to further your goals? This thought experiment highlights the need to strategically think about deploying vast AI resources.

the defining skill of the area is asking better questions.

Perplexity aims to foster global curiosity and empower agency through its product

Aravind Srinivas outlines Perplexity's mission to foster global curiosity, viewing the product as a tool that helps users continuously ask follow-up questions. He believes the current level of agency in the world is insufficient and seeks to enhance it through their platform.

The discussion touches on the concept of exponential growth, inspired by Elon Musk's challenge to transform a ten-year task into a ten-month one. This framework encourages thinking about how to achieve ambitious goals, like growing a fund 10 to 100 times faster, assuming ample resources.

Aravind sees milestones like becoming a "trillion dollar company" as motivational, emphasizing that any company, regardless of its humble beginnings like SK Hynix selling dried fish, has the potential for immense growth. The core focus remains on the mission of increasing curiosity and agency.

Our mission beyond any level of capitalism is to make the planet more curious. The product is always intended to helping people ask the next question.

Empowering Individuals with AI to Address Wealth Inequality

The discussion addresses concerns about growing wealth inequality, specifically how the accumulation of wealth in the hands of a few might be exacerbated by new technologies. The guest, Aravind, counters this by advocating for a wider distribution of artificial intelligence benefits as a solution.

A compelling example illustrates this point: an Uber driver in San Francisco used AI tools, guided by one of Aravind's YouTube interviews, to build a web application. This entrepreneurial endeavor generated more passive income than his traditional Uber driving, allowing him to reduce his driving hours and pursue a passion for coding new apps.

This story highlights that individuals with agency and a positive future outlook can leverage AI to create new opportunities and improve their economic standing. Aravind stresses the importance of communicating these positive narratives about AI to inspire others and counteract pervasive negative discussions about wealth inequality.

for the person with agency and a positive outlook for the future, anything is possible.

Enterprise SaaS companies adapt to the AI era through acquisitions while frontier AI model providers must constantly innovate.

Enterprise SaaS companies face a continuous need to reduce costs and produce new value to remain relevant. Companies like Salesforce have successfully adapted by consistently acquiring new technologies, similar to IBM's strategy with purchases like Red Hat and HashiCorp, which extended their business lifespan despite brand relevance challenges.

For new AI companies, the focus is on rapid top-line growth. They manage costs and work towards future profitability by training proprietary models and post-training on open-source solutions. This approach aims to reduce reliance on expensive frontier model tokens for existing products, reserving frontier models for designing new capabilities and experiences.

The trend of enterprises fine-tuning open models to create tailored solutions incentivizes cost reduction, posing a significant challenge to large frontier model providers. These providers must consistently innovate and introduce new capabilities every six months to maintain their leading position and relevance in a highly competitive market.

The AI field is characterized by intense competition and high stakes, meaning even current leaders cannot afford to be complacent. The rapid growth of companies like Anthropic, achieving significant valuations in short periods, underscores the volatile nature where today's winners could easily lose their advantage tomorrow.

Frontier model providers will only remain relevant if they remain at the frontier. If for six months you're not seeing a new capability, it's bad for them.

Velocity as an AI Moat and the Return of Industrial Infrastructure

Aravind Srinivas argues that early-stage companies often mistakenly obsess over identifying a moat in their first year or two. He believes the true competitive advantage is velocity. Moving fast is a form of humility, requiring constant engagement with the world and questioning of assumptions. This rapid iteration is the only shot at success.

Internally, even companies developing cutting-edge AI products can improve their own product adoption. Srinivas envisions a future where his company operates as an AGI, with an autonomous AI running various divisions with human oversight. He predicts that this shift will quickly normalize and feel like having highly effective "10X engineers" managing different aspects of the business.

If given unlimited capital, Srinivas would invest in building data centers on Earth. He sees this as a return to the industrial age, where forefathers built essential infrastructure like oil pipelines, steel bridges, and factories. He emphasizes that AI requires similar large-scale, cost-efficient infrastructure buildouts to scale effectively.

Velocity, in my mind, like moving fast is a way of expressing humility, because you're, you're constantly making contact with the world and trying to question your assumptions all the time.

Perplexity's growth strategy centers on orchestration and impact, while new AI labs must prioritize differentiation

Aravind Srinivas emphasizes the need for significant differentiation among new AI research labs, or "neo-labs." He suggests that many current labs are generic and will struggle to succeed. True neo-labs, in his view, are spinouts from larger organizations, are highly verticalized, or make foundational bets, such as challenging existing architectures like transformers or building models specifically for robotics.

Perplexity aims to achieve a trillion-dollar valuation by focusing on accuracy and establishing itself as a robust "orchestration layer." This involves seamlessly integrating across various devices, chips, AI models, tools, files, and connectors. The company's consistent goals since its inception have been accuracy and orchestration, which they believe will drive their long-term growth and market position.

Srinivas clarifies that while ambitious valuation goals are exciting, they are not his primary motivation. He states that the process of building and the desire to create meaningful impact are the true drivers. He finds it challenging to be motivated solely by wealth, instead prioritizing the tangible influence and contribution his work can have.

It's hard to get motivated by wealth, you wanna get motivated by impact.

Elon Musk's Limiting Problem Focus and Jensen Huang's Relentless Intensity

Elon Musk, despite his public persona, demonstrates an exceptional ability to concentrate on the single limiting problem or bottleneck in a venture, completely zoning out other issues. This laser-sharp focus allows him to dedicate all attention to resolving the most critical obstacle, a skill many entrepreneurs find challenging to replicate, as it requires ignoring even seemingly important distractions.

Jensen Huang, on the other hand, is characterized by an extreme, truth-seeking intensity. He operates with a mentality that his company is constantly '30 days away from going out of business.' This urgent mindset persists even though NVIDIA is a multi-trillion-dollar company with dominant technology, pushing him and his team to continually strive for innovation and avoid complacency.

The approaches of both leaders highlight a shared dedication to overcoming challenges and an aversion to resting on past successes. Musk's focus on bottlenecks and Huang's constant fear of failure drive them to achieve monumental goals, exemplifying a relentless pursuit of progress that is crucial for sustained high performance in competitive industries.

He operates with that mentality that he could be thirty days away from going out of business, that is what it takes to be Jensen Huang.

True entrepreneurial spirit prioritizes making impossible things happen and continuous impact over achieving generational wealth.

Many entrepreneurs pursue ventures primarily to accumulate generational wealth, aiming to retire and avoid future work. This perspective, however, overlooks the deeper purpose of entrepreneurship. Children raised with trust funds but without seeing their parents' continuous effort may lack the inspiration and drive to multiply their inherited wealth, as they didn't witness the process firsthand.

Influential figures like Elon Musk embody a different approach, driven by a long-term vision to accomplish seemingly impossible feats, such as colonizing Mars. Their motivation extends beyond personal enrichment to making significant impacts. Similarly, Jensen Huang's expressed desire to "die on the job" reflects an attitude of relentless dedication to one's work and mission.

The message emphasizes the importance of always being engaged in meaningful activity. While the entrepreneurial path is inherently challenging, true leaders persevere through difficulties. They are driven by an intrinsic desire to create and solve, understanding that sustained contribution, rather than an endpoint of wealth accumulation, defines a truly impactful career.

I think it's pretty hard, but you do it despite that.