Matt Miller on Building Enduring AI Companies

Joel Hellermark welcomes Sequoia partner Matt Miller, who explains that he lived in Sweden for two years and learned Swedish there. Miller recalls a later visit to Klarna where many engineers did not speak Swedish, reflecting Stockholm’s international technology workforce.

Hellermark asks whether principles for building enduring companies change with AI. Miller says technology waves rewrite some rules, but sustainable competitive advantage remains important.

A company can grow quickly by using a widely available model, but others can use the same technology. First-mover advantage matters, he says, yet founders must also build lasting value in workflows, services and customer experience.

Miller sees opportunities both in applications built on third-party models and in infrastructure that makes the technology usable. Differentiation depends on what customers value and why they continue to use a product.

He points to Character as an example of a company combining models, infrastructure and an engaging experience. He also discusses OpenAI, in which Sequoia invested, as a potentially very large opportunity; these are his investment views in 2023.

Miller acknowledges that OpenAI must navigate the relationship between its mission and building a business. He is optimistic about its potential while saying the structure will need to develop over time.

Sequoia invested early in Nvidia, while Miller later backed UK chip company Graphcore. He says independent AI chip companies have faced a difficult market because of Nvidia’s advantages and substantial research spending.

For a new founder he would generally consider software easier than building chips, while still seeing opportunities in interconnects and technologies that make other processors more useful.

Asked about Nvidia’s advantage, Miller emphasizes Jensen Huang’s focus and persistence. Hellermark adds observations about Huang’s wide span of direct reports and frequent attention to employees’ priorities; these are the speakers’ accounts, not independently audited management data.

Miller would like more founders to address work in specific industries. He cites Harvey as an example of AI assisting legal work and argues that specialized workflows can create value and differentiation.

On incumbents versus startups, he says large model development favors organizations with substantial research capacity and resources. He sees openings for younger companies in tooling, enterprise access and applications tailored to particular users.

Hellermark compares some early AI applications to simple early smartphone apps. Miller agrees that being first can produce revenue, but similar products can quickly arrive and erode margins unless a company builds an enduring advantage.

They discuss whether data and scale are durable moats. Miller thinks larger organizations have advantages in data, people and compute, while startups can benefit from a closer understanding of customers.

He highlights community: users who love a product may recommend it to colleagues and bring it to new workplaces. For him, that relationship can be a durable advantage created by serving users well.

Asked to forecast a decade ahead, Miller expects much more automation and reduced friction, comparing AI’s potential impact with the mobile phone. He makes clear the forecast is uncertain.

He also recognizes the possibility of job displacement and says societies need to bring affected people along as technology changes. He ends on the hope that people can spend more effort on research and higher-level problems.