Altman says the 2012 breakthrough in deep learning, especially its improvement with additional compute, convinced him by 2015 that AI could become extraordinarily important and was worth pursuing through OpenAI.
Altman identifies the GPT-4 period as the point when even internal skeptics increasingly concluded that scaling the technology could lead much further than previously expected.
Altman predicts that AI will dramatically expand entrepreneurship and small-business creation by giving individuals access to capabilities such as expert advice, scientific problem-solving, supply-chain management, coding and other work that previously required larger organizations or more capital.
Altman says coding and white-collar agents accelerated AI adoption in 2025 and expects persistent agents that continuously collaborate with users to emerge as another major step.
Altman argues that countries may regulate, deploy and source AI infrastructure differently, but using AI itself is effectively non-negotiable because he expects its economic and societal value to be too large to ignore.
Altman says OpenAI's role should be to provide a general AI engine, while individuals, businesses and societies determine which problems to solve, what questions to ask and what local context matters.
Altman views agriculture, industry, computing and AI not as isolated revolutions but as layers in a continuous process where each technological generation builds on previous scientific, economic and social foundations.
Altman argues that societies need to participate continuously in technological progress because failing to equip people and businesses with AI could leave them unable to create subsequent technological advances.
Altman characterizes OpenAI as seeking a middle course between uncritical optimism and extreme pessimism, acknowledging both AI's opportunities and its risks while trying to remain a reliable global partner.
Altman says OpenAI wants safeguards against catastrophic risks while allowing governments and societies substantial freedom to decide how they use AI, with people remaining central to economic and societal decision-making.
Altman says OpenAI wants a broader ecosystem of entrepreneurs, governments and companies to benefit from AI rather than having one company concentrate the technology's economic value.
Altman argues that AI coding tools can reduce work that once took startups months into dramatically shorter periods, allowing entrepreneurs to build products, test ideas and obtain customer feedback much faster.
Altman compares national refusal to adopt AI with refusing electricity, arguing that AI could become similarly foundational to economic activity, public services and citizens' quality of life.
Altman predicts that within roughly a decade AI will be so routinely embedded in products and services that people will discuss it less explicitly and simply expect systems to be intelligent and capable.
Altman warns that cybersecurity presents an urgent near-term challenge and says major failures in cybersecurity or other safety domains could substantially reduce society's willingness or ability to adopt AI.
Altman warns that scarce or expensive AI could concentrate power and advantage among early investors and wealthy users, undermining its potential to act as an equalizing technology.
Altman expects better chips and algorithms to make AI much more efficient but argues that rapidly increasing demand for intelligence will still require substantially more infrastructure if AI is to remain inexpensive and abundant.
Altman uses token consumption as a rough proxy for intelligence usage and projects enormous further growth, supporting his argument that future AI systems will require much larger infrastructure capacity.
Altman compares AI with the declining cost of electric lighting and argues that making intelligence extremely inexpensive could give people across income levels broader access to education, services, entrepreneurship and creative capabilities.
Altman argues that concern about technological dangers is constructive when it motivates engineers and policymakers to anticipate failures, learn from accidents and develop safeguards.
Altman expresses optimism that society can manage significant AI risks through technical progress, policy and learning, while acknowledging that accidents have happened and will continue to occur.
Altman argues that AI's potential economic and personal benefits are large enough that society should focus on improving safety rather than withholding the technology entirely.
Huang argues that AI represents a fundamental computing platform shift in which computing infrastructure converts energy into economically valuable intelligence rather than merely producing consumer devices.
Huang compares AI tokens with units of energy, arguing that computation produces tokens representing intelligence that can be monetized because they solve useful problems.
Huang argues that AI can raise individual and national capabilities by making advanced computational intelligence accessible through ordinary human language rather than requiring programming expertise.
Huang describes AI as a five-layer stack of energy, chips, infrastructure, models, and data or applications, and says countries need not lead in every layer but should decide strategically where to invest.
Huang says the most important national objective is diffusing AI throughout domestic industries such as education, healthcare, manufacturing, and science.
Huang explains physical AI as an agentic AI system embodied in machines, enabling applications including autonomous vehicles, industrial robots, delivery systems, surgical robots, and automated laboratories.
Huang argues that countries should build local AI infrastructure as a form of digital intellectual capacity supporting researchers, students, industries, startups, and the broader economy.
Huang argues that AI developers are responsible for building systems safely while continuing technological advancement, which he says can itself improve reliability and safety.
Huang says countries face serious downside if fear-driven discussions cause them to miss AI's benefits, and he argues that safety concerns should be balanced with discussion of prosperity and opportunity.
Huang favors precise regulation focused on actual, practical harms rather than hypothetical risks, while allowing early-stage AI technology to continue advancing.
Huang argues that Nvidia's general-purpose architecture is valuable because it supports many AI models and workloads, making expensive AI infrastructure more adaptable and durable as technologies change.
Huang predicts that systems commonly described as artificial general intelligence will emerge within the next couple of years and says current AI is already approaching that threshold.
Huang argues that even highly capable general AI will not automatically solve company problems because organizations must provide context, purpose, relevance, access, and an operating environment that enables productive work.
Huang distinguishes jobs from individual tasks, arguing that AI will automate skills and activities such as typing and talking while human jobs continue to be defined by purpose, context, and meaning.
Huang recommends using off-the-shelf AI wherever practical while also developing proprietary intelligence, arguing that companies and countries should not outsource all of their intelligence.
Huang rejects the idea that AI will eliminate all employment and instead expects people to become significantly more capable through AI augmentation.
Huang expects AI-enhanced capabilities to increase human and institutional ambition while compressing the time required to accomplish major projects.
Lutnick summarizes the discussion by arguing that countries need infrastructure and data centers to harness AI domestically so their citizens and economies can benefit.
Lutnick says wider access to AI-enabled intelligence should encourage countries and individuals to pursue larger ambitions and higher growth aspirations.
Musk argues that innovation is faster when new technologies are generally permitted by default rather than restricted until explicitly approved.
Musk says economic systems often support established companies too heavily and should instead take active steps to help startups grow.
Musk argues that countries should take a forward-leaning approach to emerging technologies because technologies such as AI and robotics can generate substantial productivity gains.
Musk estimates that AI could increase global economic output by roughly 20% to 30%, equivalent to around $20 trillion to $30 trillion annually.
Musk predicts that AI will become capable of performing essentially any task that is purely digital rather than requiring direct physical manipulation.
Musk predicts that AI software development could reach a level where human programmers can no longer compete with AI at writing software.
Musk says a general-purpose humanoid robot's usefulness depends on the combined quality of its AI software, onboard AI chips, and electromechanical dexterity.
Musk expects humanoid robot production to accelerate dramatically once robots begin helping manufacture additional robots.
Musk predicts that within ten years there could be at least one billion humanoid robots, each producing roughly five times the output of a human, making their combined output greater than that of all humans.
Musk says electricity supply is becoming a constraint on AI growth and cites an expected power shortfall of at least 15 gigawatts for AI chips in 2027.
Musk argues that countries can capitalize on AI infrastructure demand by expanding electricity generation and offering that capacity to AI data-center operators.
Bek says Sequoia operates like a competitive sports team where investors actively pursue opportunities, perform individually at a high level, and ultimately win together.
Bek argues that investors should revisit previous judgments when the environment changes, particularly because exponential AI progress can make earlier assumptions obsolete.
Bek says he remains ownership-focused because he makes relatively few investments and intends to spend substantial time helping each company with customers, hiring, and company building.
Bek says Sequoia's historical fund reviews repeatedly show that its best investments tend to be companies where the sponsoring partner held the strongest conviction.
Bek says Sequoia is experimenting with combining asynchronous written feedback with investment committee discussions to capture both slower independent thinking and faster group debate.
Bek views forceful internal disagreement as useful information for the investment sponsor and argues that investors need enough courage to take meaningful risks.
Bek says unusually strong consensus can itself be dangerous, so Sequoia may assign a devil's advocate and develop a premortem before making the investment.
Bek says he opens up personally in founder meetings so founders will reciprocate, allowing him to move beyond the company pitch and understand the individual's distinctive strengths and motivations.
Drawing on Don Valentine's framework, Bek says investors should separate whether they personally like a founder from whether that founder's distinctive strengths can produce investment returns.
Bek argues against rigid founder pattern matching and instead evaluates how far an individual has traveled and whether their historical trajectory suggests continued development.
Bek highlights Alfred Lin's distinction between exceptional operators and exceptional founders, warning that impressive employment histories can cause investors to overestimate founder potential.
Bek argues that businesses should increasingly design for AI agents as customers because agents will delegate and eventually make more purchasing and operational decisions on people's behalf.
Bek rejects the simplest view that agents will eliminate brand effects, arguing that model training creates biases that influence which products and providers agents select.
Bek sees AI agents as creating a parallel economy with categories and commercial infrastructure specifically designed around machine customers.
Bek argues that enterprise software can remain defensible because factors such as data gravity, controls, and accumulated trust create switching costs even when agents can easily compare alternatives.
Bek argues that investors need not choose categorically between AI infrastructure and applications because their opportunities can materialize on different timelines and both can support large durable businesses.
Bek says improving AI capabilities allow companies to move from copilots that assist workers toward autopilots that complete workflows and charge customers for completed outcomes.
Bek argues that outcome-oriented AI businesses can potentially capture spending traditionally allocated to human services rather than competing only for the smaller software-tool budget.
Bek says moving from copilot to autopilot does not require eliminating humans entirely; companies can progressively increase AI's role while retaining humans for judgment-intensive decisions.
Bek invokes Jevons paradox to argue that AI-driven productivity can make building software cheaper and stimulate more creation rather than simply reducing demand for software engineering.
Bek says he is reluctant to invest in traditional services businesses planning to transform into software companies because they may struggle to attract the frontier technical talent required.
Bek says intensity is essential because building a very large business is sufficiently difficult that founders need exceptional drive to persist.
Bek says Sequoia partners can publicly hold substantially different AI theses and argues that distinctive investors should express strong individual perspectives rather than converge on a bland house view.
Bek is particularly excited about increasingly capable AI contributing to scientific discovery, especially in life sciences and biology, where he believes the potential impact could be transformative.
Sottiaux says OpenAI tries to preserve a bottom-up culture where people can quickly test and ship ideas while balancing that freedom with simplicity, performance, and overall product coherence.
Sottiaux argues that organizations need conviction, rapid user feedback loops, and a willingness to redirect resources toward promising new ideas even when doing so disrupts established priorities.
Sottiaux describes the desired personal agent as one that understands a user's goals, daily activities, and surrounding context, reacts when needed, acts proactively, and minimizes the need for users to manage underlying agent mechanics.
Sottiaux argues that computers designed around human workloads will become a constraint for increasingly capable models, which can potentially operate across far more concurrent applications and therefore require greater computational resources.
Sottiaux says AI products should manage information and interruptions around human attention, with sufficiently fast models enabling users to remain in a continuous creative flow rather than manually coordinating many parallel agents.
Sottiaux distinguishes personal agents from full automation, describing systems that could handle tasks such as performance optimization, regression repair, and vulnerability patching with humans involved mainly for high-risk approvals.
Sottiaux says merging Codex and ChatGPT reflects a future in which the same underlying multimodal agent handles both technical and nontechnical tasks while its interface adapts to each individual's needs.
Sottiaux expects future AI interactions to extend beyond text toward ambient experiences that can understand surrounding context and support natural voice-based interaction.
When asked about Anthropic, Sottiaux says his focus is on building capable and efficient models, distributing powerful technology broadly, and accelerating toward OpenAI's own values rather than closely tracking competitors.
Sottiaux says Codex usage resets began as a way to compensate users when the product broke or performed below expectations, reflecting a broader principle of materially making up for poor experiences.
Sottiaux says more capable models are being used to re-engineer inference and serving infrastructure, producing substantial cost and speed improvements that OpenAI aims to pass on to users.
Sottiaux says recursive self-improvement can include models improving the inference stack, kernels, cloud-agent systems, products, and other infrastructure that determines how effectively the models themselves can be used.
Sottiaux says alignment and safety become increasingly important as model capabilities grow, and describes a training pause as necessary for teams to understand and harden the system before restarting.
Sottiaux says OpenAI uses ultra-fast inference for high-stakes situations such as incident response, where reducing latency has immediate operational value.
Sottiaux says very fast AI can support shared canvases, rapid image and prototype generation, and real-time steering, allowing users to evaluate ideas by creating them rather than waiting on lengthy generation cycles.
Sottiaux predicts that continuing gains in token efficiency, inference hardware, and system performance could make today's ultra-fast speeds close to the default within one or two years, while premium higher-compute tiers remain available.
Sottiaux argues that continued improvements in model efficiency will make capabilities that are currently expensive much cheaper, supporting OpenAI's stated goal of broad access and practical utility.
Altman says AI technology can improve much faster than people, organizations, and markets change their habits, making the economic transition slower than many technologists expected.
Altman argues that many necessary AI technologies already exist, but product design has not yet produced a seamless interface that fundamentally changes how people use computers.
Altman says OpenAI's highest-leverage work is developing smarter models and making enough compute available to run them efficiently and broadly.
Altman describes compute expansion as requiring coordinated work across chips, fabrication, power, finance, policy, supply chains, partnerships, and logistics.
Altman compares AI research with startup investing, arguing that a small number of unusually successful ideas can matter far more than the rest and justify high-risk, high-upside bets.
Altman says strong founders and researchers often think differently from the prevailing consensus and maintain conviction in ideas that may look wrong but have exceptional upside if correct.
Altman argues that using AI to advance areas such as physics, mathematics, and medicine could ultimately matter more than automating existing human tasks.
