Alfred Lin on Building in the AI Era: The Diffusion Gap, Founder Conviction, and Fire and Ice

Stanford AI Club opens its 2026–2027 Speaker Series by welcoming Alfred Lin of Sequoia Capital. The introduction highlights companies he has worked with, including Airbnb, DoorDash, Citadel Securities, Kalshi, and Zipline, along with his earlier roles at Zappos. Lin’s background in mathematics and statistics provides a starting point for a conversation about careers, founders, and AI.

The moderators ask how growing up in Taiwan and moving to New York shaped Lin. He recalls two commercial-banker parents who valued education and ranked acceptable careers: professor first, then doctor, then lawyer, with business near the bottom. Even a business career was supposed to mean working with large companies rather than small ones.

Lin chose almost the opposite path, partly because his parents had themselves taken a major risk. They left successful lives in Taiwan to give their children different opportunities in the United States. He encourages the students to recognize their own favorable position for taking risks during a major technological transition, with a Stanford education as something to fall back on.

He urges students to notice things in the world that feel slightly wrong. Those observations can be the beginning of an idea. Every problem has a solution, and every solution creates further problems; the work is to find a sequence of problems one wants to keep solving, become exceptional at it, and develop an economic engine that supports the effort.

LinkExchange, the company Lin joined after leaving his Stanford PhD program, connected websites in a way others did not initially consider important. It later sold to Microsoft for $265 million, then one of Microsoft’s largest acquisitions. Lin uses it as an example of an apparently small or strange problem becoming a substantial business.

At Tellme Networks, the problem was automating calls to 800 numbers. There was no ready-made cloud infrastructure for the service, so the company built the systems it needed itself. At Zappos, the objection was that people would not buy shoes online because they needed to try them on. Shipping shoes to customers’ homes created a way to try them there.

Asked whether ambitious students should join a startup, found a company, or work in a larger organization, Lin says there is no way to know a single correct answer. His recommendation is to maximize learning and work with very good people. That principle has served him both early in his career and when choosing founders to support as an investor.

Great companies are not overnight successes, Lin emphasizes. They often require a decade or longer. He points to Jensen Huang’s long commitment to building GPUs: the first twenty years of the company looked very different from the later period that made Huang widely celebrated. Students should look for work they could remain interested in for decades.

Lin illustrates the many possible paths with an unrealized business idea from high school. He and his father assembled clone PCs from parts costing around $1,000 and could sell them for roughly $2,000, when an IBM PC cost much more. He asked his parents to give him the money intended for college so he could start a company instead.

His parents insisted on college. Lin applied only to the few institutions they were willing to pay for, hoping he would not get in and could recover the business funding instead. The plan failed: he attended Harvard and later accepted an NSF-supported opportunity at Stanford. He came west because the technology companies he followed made the region seem like the place to learn.

Life is path-dependent, he says, but that does not mean young people are trapped by their early choices. He originally expected to finish his PhD and join a hedge fund. His parents would have predicted finance for him and entrepreneurship for his younger brother; their actual paths reversed those expectations. Other friends finished their doctorates and built successful lives at Google.

The moderators ask whether mathematics, computer science, and technical ability still matter when AI makes coding and mathematical work more accessible. Lin answers in terms of holding opposing ideas in tension. Some tasks can become commoditized at the same time that a deep understanding of how the underlying systems work becomes more valuable.

He points to the demand for people who understand models deeply enough to get exceptional results from them. Their work may no longer resemble traditional coding, but their understanding still matters. He then asks technically strong students to study humans as well: founders build for people, and economic value ultimately flows through people and the institutions they create.

For students who want to found a company but have not found their problem, Lin says the search can be a journey rather than an immediate requirement after college. He suggests asking which problems are interesting, why they matter, and how valuable solving them would be. Creating value, delivering it, and capturing some of it are three essential parts of company building.

Working with people one enjoys and learns from can help answer those questions. Lin’s first companies grew from such relationships. Students may meet future collaborators now, build something on the side, work together elsewhere, and only later start a company. He sees Stanford as a particularly valuable environment for forming those connections and testing shared interests.

DoorDash illustrates that process. Its cofounders brought together business and engineering perspectives and graduate and undergraduate experiences. They worked on several ideas before finding the problem they all wanted to solve. Lin contrasts Stanford’s latitude for student projects with an example he recalls of a similar student venture encountering administrative barriers at Harvard.

The moderators ask how he distinguishes a transient AI feature from an enduring company, especially when a later model release might absorb the feature. Lin starts with the founders’ explanation of why their ambition is larger than the first implementation. A company consisting of two people and an idea rarely begins with much defensibility.

Large companies could in principle copy the feature, but they are distracted by other priorities, including infrastructure spending. Lin therefore does not want founders paralyzed by a lack of an initial moat. The important next questions are what follows the first useful feature and how it can grow into a suite of products supported by an inspiring vision.

Airbnb’s early form was fragile: a listing service for staying with strangers, with prospective guests waiting for a host to answer an email. It did not resemble the established company people recognize today. Brian Chesky nevertheless held a different vision of travel, and communicating that vision helped attract others to a journey that would take years.

Lin remembers his father traveling internationally and choosing the same hotel chain and identical room layout everywhere. Predictability was the benefit: he even knew where an unused hair dryer would be stored. For Chesky, that same uniformity missed what travel could offer—local people, a sense of belonging, and places where a family could live together rather than simply occupy hotel rooms.

A feature for one person can be a flaw for another. Lin encourages founders to observe those differences closely. Airbnb’s value was not necessarily a downstairs conference room; it could be a kitchen or living room that let a family spend time together. Even as AI commoditizes technical capabilities, these human needs and connections do not disappear.

Asked why AI is the largest megatrend of his career and whether it resembles earlier bubbles, Lin first puts it in historical perspective. PCs, the internet, broadband, cloud, and mobile each seemed like the dominant transition of their period. Every generation has faced an overwhelming set of new tools and the expectation that it will accomplish more with them.

He distinguishes earlier revolutions from the current one. Industrial machinery replaced human and animal muscle power, while the internet and mobile connected people through communications. He describes AI as a computational transition involving machines that can reason, arriving after communication networks have already connected billions of people. That existing distribution helps explain its speed and potential reach.

Even so, Lin warns against assuming a technological megatrend transforms society immediately. When he left graduate school, he predicted that Amazon would kill Walmart. Decades later, he notes, Walmart had become much larger, and Amazon had only recently exceeded its revenue. Recognizing a technology’s direction is different from predicting how quickly institutions and customers will change.

The persistent gap is between what technology makes possible and what ordinary people adopt. Humans are the slowest-moving part of the system, he argues. Founders can work on the distance between a model’s capabilities and what customers actually use it for. Lin expects that distance to remain substantial even as the technology improves.

Technology spreads through increasingly useful abstractions. Earlier programmers worked close to machine instructions; later generations built layers that made computing accessible without that knowledge. Lin sees the same kind of work ahead in AI. Technologists can underestimate how much effort it takes to make an underlying capability usable throughout the population.

The moderators ask where the largest practical gaps lie as models advance on benchmarks. Lin says models are already more powerful than many applications make use of. He rejects the idea that software or the application layer has simply lost its value. Whatever the layer is called, somebody must connect a capability with a human who can derive an economic result from it.

He names examples in law, medicine, marketing, and finance, including Harvey, OpenEvidence, and Profound. Professionals in those fields generally do not want to descend into the model and program it directly. An application can translate the capability into the professional’s context, workflow, and expectations, which is where Lin sees substantial value being created.

Another opportunity is helping organizations own their intelligence. A pharmaceutical company, for example, may want to adapt an open-weight model to its own data rather than put everything into a small number of dominant services. Lin expects demand for tools and specialized systems that give customers more control over knowledge, models, and how they are applied.

He points to drug-discovery work, developer tools such as LangChain, and open-weight model builders as different responses to that opportunity. The gap between capability and adoption keeps generating work. A solution to one problem introduces others: giving models the ability to investigate cybersecurity issues, for instance, creates new questions about containment and control.

The moderators ask whether an application provider might go further and deliver the underlying service itself—for example, moving from tools for lawyers toward performing legal work. Lin says those possibilities are open. He describes Harvey’s evolution in models and customer groups as a reason to think more ambitiously about the scope of an AI company.

Lin compares that expanded ambition with children who can now make and play games quickly through coding tools, rather than first learning an entire programming language. Their first success gives them the desire and ability to build more. He sees a similar opportunity for founders to keep extending what people can accomplish with these systems.

The discussion turns to Lin’s idea of “fire and ice.” Fire means passion, vision, and the urge to solve an important problem; ice means composure, clear strategy, and execution. A company may begin with something creative that nobody has seen before, but it soon needs to build a product, deliver it, and persuade customers to pay.

Creating value without getting it into other people’s hands is insufficient. Companies that begin with abundant fire need the cold discipline of operating and management. That discipline can then support further creativity by supplying resources and capacity for another idea. Lin presents the two forces as mutually enabling parts of the entrepreneurial journey.

At Zappos, the company began with shoes and expanded into handbags and clothing. Amazon followed a longer version of a similar pattern: establish discipline and profitability in books, then move into music and additional categories. In Lin’s account, each functioning business supported the next expansion until the company became the everything store.

Sometimes vision and operating detail coexist within one founder; sometimes they are distributed across a team. Lin describes himself as the ice alongside Tony Hsieh’s fire at Zappos. He encourages founders and managers to know which tendencies come naturally to them and which qualities they need from a complementary partner.

A moderator asks about circular investment in AI, using the example of Nvidia funding a company that then purchases Nvidia GPUs. Lin says a system supported only by circular money would be problematic, but circulation itself is normal. He suggests understanding how money moves through the economy before treating every loop as evidence that value is fictitious.

He illustrates the point with a simplified banking example: a deposit supports lending, and the borrowed money eventually enters another account and can support further activity. His concern would be a situation in which all incoming money immediately flows back out with no cushion. He sees Nvidia’s ecosystem investments partly as an effort to help customers build on its hardware.

Lin says the greater issue is leverage. Companies that once produced cash are taking on debt as they expand data-center spending. Debt can magnify outcomes in either direction, and he regards the amount of leverage in the system as more important to examine than the mere existence of circular transactions. He still welcomes scrutiny and caution about the overall system.

Audience questions begin with the qualities Lin looks for when founders raise large amounts before reaching product-market fit. He says he tries to understand the founder’s journey and what brought them to the problem. Outliers may fail to fit a familiar pattern. Their history of taking risks and the conviction behind their idea can matter more than a conventional profile.

Conviction is not the same as passion or a persuasive presentation. For Lin, it includes having investigated a problem deeply enough to form a novel, compelling insight. He uses Tony Xu’s experience of restaurant work alongside his mother: Xu understood the mismatch between kitchen capacity and dining-room capacity from lived experience, not only from a theoretical delivery-service proposal.

Asked how finance and operating backgrounds shape investors, Lin distinguishes early-stage investing from growth investing. The former puts more weight on founder quality; the latter emphasizes business quality. Investment banking or private equity can teach a great deal about the economics of a business, while operating beside a founder can build a different kind of understanding and empathy.

He cautions that advice often reflects the particular experience of the person giving it. His own ability to help founders comes partly from having lived through operating problems. Asked for his best mentor, he resists choosing only one: his father’s lesson was that there is something to learn from everyone, and different problems call for different teachers.

An audience member asks which inherited assumptions the next generation should reconsider. Lin says earlier organizations were built around the need for many people to write and control software. AI may change both the work and the structure required to coordinate it. He does not claim to know the final organizational model, but expects existing management systems to need revision.

Earlier technology shifts changed jobs without necessarily making companies smaller. Now an individual may be able to take an idea through development, launch, and revenue generation. If a large company contains many people with that end-to-end capability, Lin asks, what should the organization look like? He expects more decentralization than the inherited command-and-control model assumes.

Some familiar frameworks may also be weaker than their popularity suggests. Lin questions the story around the lean startup by pointing out that seed-stage funding amounts kept rising during his time at Sequoia. His broader advice is to be wary of frameworks that become persuasive because they tell a neat story, rather than because their limits have been understood.

Simple narratives are powerful because people like clear answers. They work up to a point, then obscure the nuance that matters when circumstances change. Lin encourages students to inspect their own frameworks and understand the boundaries at which they break. Nobody reasons from first principles all the time, so awareness of those boundaries becomes especially important.

He offers his own changing view of rising investment check sizes. Larger checks and higher valuations initially suggested worsening returns to him, but the potential outcomes were growing as well. He compares the largest public company when he joined Sequoia with Nvidia’s much larger value around the time of the talk, arguing against treating the upper bound of outcomes as fixed.

In that framing, founders may need more capital to compete and may have a much greater distance to travel to build a legendary company. The available human time has not expanded at the same rate. Lin treats that as another reason for ambition and productivity, while keeping the observation tied to the scale of businesses in the environment he describes.

An audience member asks how a founder or investor wins other people’s support. Lin’s first answer is a vision compelling enough that others want to join. At the beginning it will not persuade everyone, and he does not treat universal agreement as necessary. The task is to find the community that wants to work toward that particular vision.

Asked where an AI-native hedge fund could find a durable edge, Lin begins with the market rather than the technology. Existing funds already use machine learning to price many things, so outperforming them directly is difficult. He would look for instruments or areas those firms are not focused on, where a new model might address a neglected pricing problem.

A doctor in the audience asks about regulated markets and powerful institutions controlling important data. The question uses OpenEvidence and access to medical publications as an example. Lin advises an early startup against trying to break an established oligopoly head-on. Understanding how to work around or with the powerful parties may offer a better path to serving customers.

He describes OpenEvidence as building a product physicians wanted: an evidence-based conversation grounded in a published body of work. In his account, that preference distinguishes the product from simply asking a general chatbot a question, even if an answer were similar. Partnerships and access to relevant material can be part of building the product rather than merely obstacles to it.

Powerful incumbents are difficult to attack because they are powerful, not only because a market is regulated. Lin points to DoorDash’s early focus on smaller restaurants before it worked with major chains. A startup can begin in a segment a large company does not prioritize, develop strength there, and expand later. Incumbents cannot do everything, which leaves room for new companies.

Asked how to spend the first ninety days as a student founder, Lin begins with people. He would find collaborators he wants to go to battle with for the next decade and work together to test compatibility. The question is not only whether an idea looks promising, but whether the group can keep building through a long and difficult journey.

The DoorDash founders combined complementary business, marketing, product, and engineering abilities. Lin also points to the different contributions of Airbnb’s founders. Working together helped the teams discover both their shared interests and their ability to cooperate. He thinks students sometimes focus so heavily on choosing a problem that they neglect whether the right people have come together.

Lin recalls asking his father about his smart college friends. His father’s concern was that they had too many options. Constantly comparing and optimizing among attractive alternatives can prevent someone from committing to the one or two things a company needs. A company requires a core idea and vision, and giving the next decade to it can feel frightening.

An audience member asks what makes Lin exceptional as an operator or investor. He begins by saying he is not that good, noting how often investment decisions turn out wrong. The distinction he identifies is his relationship to pain, risk, and failure. He feels relatively comfortable taking risks and remaining outside other people’s expectations.

He describes learning early to be his own person rather than relying on others’ approval. Good test results did not make him feel like a conventional good student, and he often did not fit in. An entrepreneurial path may require being an outsider for a long time. He sees value in processing signals from the world while still forming an independent judgment.

Lin wants mistakes to become sources of learning. He distinguishes mistakes one learns from from mistakes one does not learn from, and warns against the latter. Even expensive tuition can be useful if the lesson is absorbed. The conversation’s final rapid-fire questions then move from business judgment to more personal stories.

His first job was a paper route. Asked about a near-death experience during a company’s launch, he instead recalls an actual traffic incident while driving to Tellme Networks. On the first rainy day, a car cut him off; when he swerved to avoid a rear-end collision, his own car spun around and ended up facing the opposite direction on the highway.

The final question asks about an unusual place to sign a term sheet. Lin recalls signing one in a bar. Asked how many drinks were involved, he says the founder had three and he had none. The session closes with thanks and applause after a discussion that repeatedly returned to risk, judgment, and choosing people for a long journey.