Saam Motamedi on AI Valuations, Series B Risk and Backing Founders

Saam Motamedi says parts of the AI investment market look more exuberant than the 2021 peak, with young companies attracting high valuations and Series B pricing particularly competitive. His assessment is a 2024 market opinion, not a finding that every investment in the period will lose money.

Harry Stebbings speaks with Motamedi, a Greylock general partner who invests in enterprise software, security and AI. Greylock has positions in several companies he discusses, including Figma, Abnormal Security, Upwind and Braintrust, giving the firm a direct interest in their success.

Motamedi grew up in Houston, where competitive debate taught him to enjoy competition and biomedical research showed him what small teams can accomplish. He says those experiences shaped how he works with early-stage founders.

He mostly agrees that venture rewards the energy associated with younger investors, but points to experienced colleagues who remain intensely competitive. His view is about willingness to pursue opportunities, not a rule that age determines investment ability.

He calls the AI market an exuberant bubble, citing seed valuations in the tens of millions and sometimes above $100 million post-money. He also reports seeing companies priced at 100 to 200 times revenue, compared with his contemporaneous estimate of roughly 15 to 20 times forward revenue for the most highly valued public software names.

Motamedi says those comparisons only make sense if private companies sustain unusually strong growth. He cites a reported OpenAI annualized revenue figure as an example of exceptional growth, while questioning how many companies can maintain it. The interview does not independently verify that report or convert an annualized rate into realized revenue.

Stebbings argues that corporate investors may pay for strategic partnerships or cloud consumption as well as financial returns. Motamedi agrees that such motives can distort prices. He also cautions that fast initial revenue can mask uncertain retention, citing interest in early consumer AI applications that later faded.

When a company grows rapidly, Greylock starts by asking why it should not invest, he says. It then examines market dynamics, product-market fit, retention and defensibility. In his view, even a rise from zero to $20 million of revenue in six months is not enough on its own if those fundamentals are weak.

Asked how to distinguish numerous AI competitors, Motamedi says software investors have always faced similar questions: which team can build deep, sticky workflows, earn pricing power and establish distribution? He argues that labeling an application a model wrapper does not settle whether it can create lasting value.

He points to HubSpot and Greylock portfolio company Figma as examples of products that developed useful workflows and distribution. Other software products, he says, generated superficial value and later stalled. His judgment about portfolio companies is not an independent performance analysis.

Stebbings asks whether an OpenAI update could destroy an application startup. Motamedi expects model providers to compete aggressively in foundational capabilities such as general writing and code generation. He sees more room for specialized legal, medical and developer workflows, although he does not claim those markets are free from competition.

On foundation models, he stresses large uncertainty. Greylock is less enthusiastic about pure model-as-a-service providers if competitive gaps narrow and margins fall, but he sees potential value where a model must be tightly integrated with a first-party application. Greylock's historical investments make this an investor's strategic view.

Personal agents, enterprise agents and some coding applications might need close model integration, he says. Customer-service applications may instead benefit more from work around the model. He expects pricing to vary with value and potentially combine per-seat charges with charges for work performed.

Motamedi rejects the prediction that AI means the end of SaaS. He sees a chance for new software businesses to challenge existing systems of record when data models, delivery and pricing, or interfaces change substantially. He cites Salesforce, Workday and ServiceNow as established examples.

He explains a data model using customer and account records in a CRM: users, implementers and adjacent software learn to work within the same structure. A delivery change might resemble the move from on-premises software to cloud subscriptions; AI could create another pricing shift.

His interface hypothesis is that an agent could work alongside a salesperson, navigate systems and request missing information. He predicts that in ten years many salespeople may not need to know the underlying CRM interface. This is a future scenario, not a demonstrated capability or guaranteed outcome.

He imagines an AI application synthesizing email and call information to build a sales forecast rather than depending only on manually entered CRM data. He expects this could be more accurate, but the interview provides no comparative trial establishing that claim. Existing providers may also adopt similar ideas.

Stebbings asks whether customers will pay more simply because software has AI features. Motamedi says it is too early to know, but expects stronger pricing power when a product reliably performs valuable work rather than merely adding a chatbot. His AI sales-development example is hypothetical.

On 2024 seed and Series A pricing, he gives broad ranges for a particular cohort: roughly $20 million to $40 million post-money for strong seed teams and $80 million to $150 million, sometimes $200 million, post-money at Series A with early product-market fit. He says some seed companies were newly incorporated and had little or no product or revenue.

Motamedi considers some high seed prices rational under a power-law strategy. For a sufficiently large fund, he argues, the difference between a $20 million and $30 million entry matters less than access to a truly exceptional company. Stebbings agrees that access matters but argues that investors may not be compensated for early risk at those prices.

Motamedi uses Upwind, a Greylock portfolio company, as an example of paying for an experienced team. He says it raised a $26 million seed round and its founders had previously built a business that reached $100 million of ARR and was sold to NetApp for about $500 million. These are his case-specific claims, not proof that large seed rounds are generally low-risk.

Stebbings calculates a potential entry valuation from standard dilution assumptions. Motamedi says that particular valuation was lower and agrees that investors need a ceiling. He distinguishes making an exception for an outstanding team from treating every expensive deal as an exception across the whole portfolio.

Greylock's 17th fund was a $1 billion pool, he says. For core early positions, it considers ownership of roughly 20% to 25% or more adequate, though it does not achieve or require that in every deal. Motamedi says the firm sometimes commits more capital before conventional proof to reach those positions.

Stebbings argues too much early capital can undermine urgency and creativity. Motamedi agrees overspending is a risk but says the effect depends on the founder: he has seen urgency in large rounds and its absence in small ones. The same financing amount can be an accelerator or a temptation to spend poorly.

Asked how a large fund maintains standards on small checks, Motamedi says partner time is the scarce resource. Each Greylock partner makes relatively few investments, and he describes spending 90 days working with founders before one $6.5 million investment. He says other decisions must sometimes be made much faster.

He estimates that around 80% of Greylock's investments begin with a relationship before the company exists. Small initial checks can grow through later rounds, he says, citing portfolio companies including Figma, Rubrik, Abnormal Security and Discord. These are his descriptions of the firm's process, not audited portfolio statistics.

Stebbings raises signaling risk: could an outside investor interpret a seed backer's refusal to lead the next round as a warning? Motamedi says he has not seen it matter in Greylock's portfolio. Stebbings counters that a strong firm brand could itself help portfolio companies raise, so those examples do not disprove signaling.

Motamedi cites Cresta, a Greylock investment, as an example of evaluating a company on its own merits after another firm led the seed. He says he did not know why the earlier investor did not lead the A. His anecdote illustrates his own decision process, not a general test of signaling effects.

At each round, he says Greylock asks whether a new investor can add meaningful help. If a founder finds someone who can materially improve the business, Greylock may accept a smaller ownership share; if an incoming investor adds only capital, he argues an existing investor may be a better option. That position also serves Greylock's interest in maintaining ownership.

The firm holds significant reserves both to follow successful companies and support them through setbacks, Motamedi says. A small initial position may become a much larger cumulative investment over time. Reserves can fund defensive inside rounds as well as offensive follow-ons.

He objects to treating $1 million in annual recurring revenue as a universal Series A threshold. ARR can indicate product-market fit, but most companies that reach $1 million do not reach $100 million. He begins with the founder and market, then uses ARR as a secondary piece of evidence.

As an example, he says Wiz raised its Series A at a $500 million valuation without ARR, backed by an experienced founder in a changing cloud-security market. Greylock invested later in growth rounds, he says. His assessment is retrospective and does not establish that every pre-revenue company at a high valuation is attractive.

Stebbings prefers less crowded, unglamorous markets. Motamedi agrees that finding an overlooked market can be a route to a strong investment, but says paying more than competitors for a company can also be contrarian if the judgment is right. Neither approach removes the risk of being wrong.

He says Greylock incubated Abnormal Security and invested in several rounds before repeatable product-market fit was clear. He cites its announced $100 million-plus ARR and growth above 100% at the time. Those are company-reported operating claims, and Greylock is an investor.

For later-stage decisions, Motamedi says market dynamics become even more important. An exceptional founder may pivot, but it becomes harder to redirect a company built around a small market after it has scaled. Stebbings notes successful counterexamples, including HubSpot's expansion into CRM.

Would he back a great founder in a market he dislikes? Motamedi says it is one of Greylock's hardest recurring debates. He is more willing at seed if the founder is exploring a promising broad area but has chosen the wrong initial niche; even then, judging whether the founder is truly exceptional is difficult.

He believes some of Greylock's biggest missed seed investments came from rejecting an outstanding founder's first idea. Sometimes the firm was right that the idea would change, he says, but wrong not to build a relationship before that change.

He disputes the statement that growth rounds have disappeared. For a small group of companies with strong growth and attractive markets, he sees intense competition. He thinks new growth funds at established early-stage platforms replaced some retreating investors while chasing fewer qualifying companies.

Motamedi says Series B pricing may be frothier than in 2021 and questions whether the vintage as a whole will make money unless public multiples expand or growth proves unusually durable. He explicitly expects some outstanding individual Series B investments. This is a forecast about a cohort, not a verified result or a claim that every deal fails.

He rejects the idea that investors must always accept prevailing deal terms. Independent thinking can be costly if wrong, he says, but merely playing the market on the field would also have been expensive in several past periods.

He identifies Glean as a miss: Greylock knew founder Arvind Jain from its Rubrik investment but did not try to invest in Glean's B round because enterprise search had a poor history. He now believes the team was well placed to use generative AI to change search. This is retrospective judgment, not evidence that the investment has produced realized gains.

A second miss was Codeium, whose founder Varun had been a Greylock fellow. The original seed idea concerned model inference, an area Motamedi says the firm did not appreciate sufficiently at the time. These experiences shifted his seed-stage default toward backing founders he considers exceptional even when their initial idea is uncertain.

Of sourcing, selection and service, he considers sourcing the hardest and most important: an investor cannot choose a company they never encounter. He acknowledges Stebbings' point that some successful investors and founders do not require extensive post-investment service.

For younger investors seeking to grow within a firm, he advises thinking beyond personal deal credit and developing a reason founders would choose to work with them. Stebbings distinguishes this from the incentives of someone using a role primarily as a stepping stone; Motamedi agrees those incentives can differ.

He suggests building a long-term area of expertise, meeting the next generation of operators before they start companies and becoming useful in that field. Such specialization can improve both sourcing and the ability to win a founder's trust.

In quick-fire answers, Motamedi says investors can help before product-market fit by discussing customer segmentation, attending meetings and giving feedback. He praises Elad Gil's work across stages and notes that they have worked together at Greylock portfolio company Braintrust.

He recalls Braintrust founder Ankur Goel describing specific developer needs and a plan to address them. Greylock invested $5 million, he says. His praise for the business is that of an investor, not a verified measure of its eventual return.

He thinks venture incubations often fail when firms take too much ownership, attract the wrong founders or cannot help. He cites Greylock's historical work with Palo Alto Networks and Workday, and its incubation of Abnormal Security, as examples he considers successful; past successes do not prove the model works generally.

He has changed his mind about specialized AI models: a focused model combined quickly with a strong application workflow may be defensible even if general models improve. He cites ElevenLabs, in which Greylock had not invested, as an example of a company building such a workflow.

He wishes he had appreciated sooner how much large markets matter. In his view, great companies can start in a narrow niche while deliberately expanding into larger adjacent markets. Market size is an investment judgment, not a guarantee that expansion succeeds.

His final suggestion is that founders should reference-check prospective investors and ask company leaders what would have been different without that investor's involvement. He says he has backed excellent founders who did many references and others who did none, so he does not treat the number of checks as a reliable quality signal.