Sequoia's Alfred Lin and Konstantine Buhler on AI, Investment and the Agent Economy

At Citadel Securities' Future of Global Markets event, Nohshad Shah welcomes Sequoia Capital partners Alfred Lin and Konstantine Buhler for a conversation about investing and technological change. Sequoia is an investor in Citadel Securities and in several companies discussed, an important commercial relationship for understanding their views.

Lin describes becoming a co-steward of Sequoia as an unexpected honor. He says the partnership's principles place founders first, followed by limited partners, the firm, the team and the individual. He wants to leave the organization stronger for future partners and says its next investment matters more than its past record.

He credits Buhler's conviction with helping bring Waymo and Citadel Securities into the portfolio. Lin describes his leadership role partly as creating room for the next generation of partners to make decisions and continue the firm's work.

Buhler recalls meeting Citadel founder Ken Griffin at a New York conference while he was a graduate student studying machine learning. Griffin became a mentor, he says, and their relationship eventually led to a discussion about Sequoia investing in Citadel Securities. Both guests speak positively about a company in which their firm holds a stake.

Asked about investment decisions that were not obvious at the time, Lin describes a process in which a partner writes a thesis, the firm performs diligence, the partners debate and vote, and the sponsor's conviction carries weight. He believes the quality of the discussion comes from partners challenging one another's assumptions.

Lin organizes that discipline around five Cs: clarity, conviction, courage, creativity and commitment. A clear thesis should withstand different frameworks and an extended time horizon. Conviction requires answering partners' questions, seeking answers when possible and admitting what remains unknown.

Courage and creativity matter when a desired investment is difficult to access or the market's terms are unattractive. The final C, commitment, reflects the expectation of working with founders for a decade or longer. Lin presents this as Sequoia's philosophy, not a formula that guarantees investment success.

Buhler offers Kalshi as an example of a decision that looked unusual before regulatory clarity. Lin resists declaring the outcome complete. He says the partnership also had to challenge its own assumption that a mature New York company such as Citadel Securities fell outside its usual business.

On AI investing, Buhler points to Sequoia's history with Nvidia, Google and later model companies. He says the firm tries to look beyond the label: remove the word AI from a business description and ask whether the underlying problem and proposition still make sense.

The partners say disagreement is frequent because a young business can have a wide range of outcomes. They describe continuing the debate until colleagues have heard each other out. Lin adds that Sequoia invested later in OpenAI and Anthropic than it would have liked, illustrating how old frameworks can slow a decision.

Lin says OpenAI's nonprofit origins made its investment structure hard to assess, and that Sequoia waited for conviction about a viable commercial structure and the models' progress. Even then, partners debated the valuation and unclear economics. He describes the firm's eventual investment and later additions, speaking as an interested investor.

Sequoia normally avoids backing direct competitors when it is working closely with a company, Lin says. Over time, portfolio companies told the firm they used both OpenAI and Anthropic, leading it to revisit that rule and invest in both. He presents the decision as an example of adapting old frameworks to new circumstances.

Shah asks where value will accrue across AI infrastructure, frontier labs and applications. Buhler compares AI's potential effect on cognitive work to industrialization's effect on physical work. He predicts major expansion over decades and winners at several layers; the analogy and timeline are forecasts rather than measured outcomes.

Buhler says infrastructure and chips have captured substantial demand, while leading model providers can command a premium. Over the longer term, he is most excited about applications that turn capabilities into useful services, citing autonomous driving as one example.

He names coding, legal work, medical information, cybersecurity and customer support as application categories. Several companies he mentions are in Sequoia's portfolio. His comments about their capabilities and effects on work are investor perspectives, not independent evidence that they will fully automate those jobs.

Lin says the answer to where value accrues changes by time frame. Early in a technology cycle, he argues, chips and foundation models receive investment because the infrastructure and intelligence must first be built. Applications can then help people use capabilities that advance faster than adoption.

He calls the space between technological capability and human adoption the diffusion gap. A founder can build products for professionals who do not want to train or program their own models. Lin sees this as a continuation of software making computing easier to use, even if the product form changes from SaaS or mobile applications to agents.

Shah asks about open-weight models. Buhler says they can provide enterprises with less expensive options and more control, while he believes frontier providers retain pricing power for their most capable models. He cites an estimated share of spending on open models; the figures are estimates and depend on definitions and the period measured.

Buhler alleges that some Chinese model developers have benefited from access to US models outside their terms of service and expects Western open models to improve. This is his allegation and forecast; the interview does not independently establish a general finding about all developers.

Lin distinguishes open weights from traditional open-source software and argues that lower-cost models can help entrepreneurs improve margins. He also expects competition from open models to pressure closed providers to lower costs. That is his view of the competitive mechanism, not a guarantee about future pricing.

Lin separates creating value, delivering it and capturing it. A founder can build something useful yet struggle to reach the right customers or sustain a business model. He says open distribution can help delivery while making monetization harder, and that defensibility often develops over time rather than appearing fully formed at the earliest stage.

Shah turns to consumer agents and asks how distribution, advertising and transactions might change. Lin expects the business model of AI to differ from earlier technology waves, while human desires for content and connection persist. He says trust is a central barrier before people will hand agents personal information and payment authority.

Buhler defines an agent simply as AI that both produces an answer and takes action. He describes an agent economy in which systems interact and potentially transact with one another. He says consumer adoption has surprised him, though he acknowledges these experiences are still early and imperfect.

He compares automated trading at Citadel Securities with a future in which agents coordinate logistics or other transactions. The comparison is an analogy from an investor in Citadel Securities; it does not imply that every trading system and future consumer agent works the same way.

Both partners expect agent adoption to increase computing demand. Buhler says scaling inference will require advances in hardware and architecture; Lin uses the idea of computing resources assigned to active agent tasks to illustrate the potential scale. Their numerical examples and demand projections are illustrative, not established capacity requirements.

Buhler explains how the firm translated its computing thesis into an investment search. In addition to parallel GPU work, he expects agent workflows to need sequential processing, memory and CPUs. He says Sequoia recruited a team with Apple Silicon and Qualcomm experience to build a company focused on CPUs for agent workloads; his bottleneck assessment is an investment thesis.

Lin recalls an internal memo about the return of the CPU and says the idea preceded the eventual company by one or two years. He uses it to illustrate how the partnership tries to discuss a possible change before an investment opportunity appears.

Shah asks how the firm balances the risk of paying too much against missing a future winner. Lin distinguishes early and growth-stage decisions: for very early companies, he says picking the right founder and outcome can matter more than a relatively small valuation difference; for mature, highly valued companies, he expects a tighter assessment of possible returns.

Lin says investors can fail to update their expectations as the scale of successful companies changes. He uses the increase in the largest public-company market values over his career as context for reassessing early-stage opportunities. His argument is a venture investor's framework, not a recommendation that valuation is irrelevant.

An audience member asks how Sequoia decides investment size. Buhler says selecting the company is the first priority and that the partnership also values meaningful ownership because it commits considerable time to a relatively small number of businesses. He says gaining additional ownership can become harder once other investors pursue the same company.

Lin emphasizes the firm's preferred language of long-term partnership with founders. The partners say they seek to help companies develop rather than think only about buying and selling stakes, while acknowledging that investment returns remain how limited partners are paid.

Asked which parts of AI are overlooked or overrated, Lin highlights physical AI and space-related communications and infrastructure as potentially underrated. Buhler also names biotechnology and health. Both are discussing investment opportunities they see, including areas in which Sequoia has interests; they are predictions, not evidence that any particular investment will succeed.

On the overrated side, Buhler questions consumer applications with fast revenue growth that solve only a one-off problem and do not become a default for an important workflow. Lin similarly warns about businesses without durable defenses. Their judgments depend on future competitive conditions.

Buhler imagines applying substantially greater computational capability to disease prevention, treatment and healthspan. He is optimistic about biotechnology as a business as well as a scientific opportunity. The conversation does not establish that any specific medical result or treatment will follow.

An audience question turns to cryptocurrency and blockchains in an agent economy. Buhler says digital payments could fit agent transactions, but treats the choice among blockchain, cryptocurrency and other digital money as an implementation question rather than the end goal.

Lin imagines decentralized ledgers playing a larger role when agents transact across parties that do not already trust each other. He offers a possible clearing mechanism, not a demonstration that blockchain is necessary or sufficient to establish trust between agents.

Asked for an underappreciated opportunity in 2030, Buhler returns to the application of intelligence to large challenges: disease, materials and food production. He imagines large coordinated projects using many machines to pursue breakthroughs. These are aspirations and investment forecasts, not promises of medical or scientific results.

Lin describes himself as a technology optimist. He imagines AI multiplying the cognitive resources available to address problems and cautions against treating technology companies as rival sports teams: in his example, both Apple and Microsoft became important businesses. He expects many opportunities rather than naming one guaranteed winner.

The conversation closes with thanks to the partners. Their discussion combines firm-specific investment philosophy, examples from companies in their portfolio and speculative views of future technology; none of it establishes a particular investment outcome.