Tom Hulme on Angel Returns, Venture Discipline and AI Investing

Tom Hulme describes venture investing as a portfolio of emotional highs and lows, and distinguishes helpful investors from those who interfere without understanding. The GV managing partner joins Harry Stebbings to discuss angel returns, investment discipline and company building.

Hulme recalls being bullied at school and says the experience taught him perseverance and empathy, though he would not wish it on anyone. He and Stebbings discuss the discomfort of direct feedback; Hulme believes strong founders listen carefully to the market even when they disagree with an investor.

After selling a company, Hulme began angel investing as an experiment to discover whether he enjoyed helping founders and whether he was any good at selecting them. He found he liked supporting strategy and design, while noting that venture’s long feedback loop makes skill hard to judge.

He divides investors into those who contribute useful expertise, passive investors who stay out of the way, and people who interfere without adding value. His rough estimates of how common each type is are impressions, not measured industry data.

The two discuss deal structures and valuation incentives. Hulme believes convertible financing can postpone a priced round and maintain paper marks, while liquidation preferences and IPO ratchets can change outcomes for employees and investors. He advises founders to assess the whole set of terms and preserve options, not just pursue a headline price.

Stebbings worries that too much funding can push a company into premature expansion. Hulme says the use of the money matters: adding costs before product-market fit can reduce adaptability and slow learning. More experienced founders may be less prone to change behavior simply because capital is available.

For roughly 27 pre-2015 angel investments, Hulme reports about 4.5 times DPI and 24–25 times TVPI. DPI measures distributions relative to invested capital; TVPI also includes remaining value on paper. These are his approximate portfolio figures, not a realized 24-fold cash return.

He cautions that low entry valuations from that period are not necessarily repeatable and admits he would have ranked his future winners poorly at the outset. He concluded that, for his own angel portfolio, following every pro-rata opportunity would have reduced performance, since some highly marked companies later failed.

His durable winners, he says, built on fundamentals and compounded over a decade. He cites GoCardless, which he invested in at YC, and contrasts it with investments whose acquirers later failed, including Fab and Jawbone. Those experiences inform his view, rather than prove that every popular deal fails.

Hulme says secondary liquidity was not available to him in some earlier investments. When opportunities arise, he often suggests founders or early investors consider selling a modest portion to reduce regret, while recognizing that circumstances and available markets vary.

Among his mistakes were following momentum, relying on other angels’ diligence and not spending enough time with founders. A famous fund and a well-known executive can create an appearance of validation even if neither has made a large commitment relative to their resources.

He asks founders about their product partly to understand how they think, measure success and investigate the market, even if the product later changes. He likes asking why this team has an advantage, why the timing is right and what might go wrong; candid concern and willingness to seek help are useful signals to him.

Hulme believes an idea alone is rarely enough: competitors may have had it already or copy it, so execution matters. Timing matters too. He cites companies that survived while waiting for their market and warns that spending heavily before demand arrives can end that opportunity.

He rejects a fixed dollar price for learning venture investing. His suggested minimum is several investments observed over several years, starting with small checks that a person can afford to lose. He also cautions against assuming check-size differences accurately express conviction.

He distinguishes founders investing personal money from founders raising a side fund. Their operating experience may help portfolio companies, but he worries that managing other people’s capital is effectively a second company and can distract from the first.

Hulme frames venture work as four activities: sourcing, selecting, supporting and selling. Selling includes convincing founders to accept investment, communicating with limited partners and helping recruit executives. He says maintaining sourcing skills matters even as a portfolio grows.

He considers sourcing his relative strength, helped by curiosity and a trusted network. Supporting is harder for him because he empathizes deeply with founders’ setbacks; he compares a good board member to a shock absorber who helps moderate both exuberance and distress.

Hulme argues that networks create opportunities that can later look like luck. He also acknowledges that access to venture is uneven: people may have a chance to succeed once inside but never learn how to enter the industry in the first place.

An optimistic culture needs a deliberate space to examine the downside. He describes pre-mortems in which a team imagines how a decision failed and invites objections. In his view, optimism and explicit risk analysis are complementary.

Investors can underestimate winners because they struggle to imagine successive growth curves. Hulme’s team considers both the current plan and the options it might open, then tests the downside. He says he can describe a bull and bear case for every investment and does not claim complete certainty.

Stebbings argues an exceptional team can turn a modest initial market into a much larger opportunity. Hulme agrees that talent creates options but notes the difficulty of sustaining a market opportunity and reaching that scale; the eventual outcome is uncertain.

Both worry about limited IPO and acquisition liquidity. Hulme says exits matter not only to investors but to employees and founders who may use proceeds and experience to build new companies. He particularly values founders who have worked inside rapidly scaling businesses.

He challenges the idea that every venture investment must individually return the entire fund. Concentrated power-law portfolios have worked historically, he says, but other approaches may also work. The essential discipline is to choose a coherent strategy and avoid abandoning it for one exciting deal.

GV has a long time horizon and focuses on absolute returns rather than near-term IRR, Hulme says, because Alphabet does not impose the same time pressure. That institutional position shapes GV’s strategy; he still advises founders and early investors to consider partial liquidity when appropriate.

Discussing Hopin, he considers the difference between durable product-market fit and demand created by pandemic restrictions. He says founder liquidity can be reasonable, but raises a separate question about how other employees are treated; he says he lacks specific knowledge of their outcomes at that company.

Stebbings sees speculative pricing in generative AI. Hulme believes capital is flowing toward AI while other companies struggle, and predicts that some valuations will prove difficult to justify. Both are investors commenting on a market in which their firms participate.

Hulme compares training foundation models to building power stations that can quickly become obsolete as competitors build similar capacity. He worries about rapid commoditization and says GV had invested in AI infrastructure and applications but not foundation models at that time. The analogy and outlook are his investment thesis.

He acknowledges that an investor could profit from momentum and secondary sales even if long-term fundamentals are uncertain. Drawing on Clay Christensen’s framework, he sees much generative AI as a sustaining innovation that improves existing products rather than necessarily rebuilding entire industries; this is a contested prediction.

He expects cloud providers to supply models as part of a broader utility-like offering. He also speculates about Microsoft’s motives in the Inflection transaction, without inside evidence. Asked about OpenAI at a $90 billion valuation, he says he would struggle to invest given competitive and durability concerns.

For a more defensible model business, he would look for distinctive capabilities such as useful memory or agency. At the application layer, he cites GV-backed Synthesia as an example of building distribution and enterprise functionality around synthetic video. His favorable assessment carries GV’s investment interest.

He asks whether AI founders can commercialize, iterate and create value, not only whether they have impressive research credentials. His numerical guesses about investment losses across models, applications and incumbents are speculative, not measured loss rates.

He believes existing distribution and data may give incumbents an advantage in areas such as legal work and coding, citing Microsoft’s products. This assessment should be read in the context of his position at GV, an Alphabet investment firm, and his broader market thesis.

Hulme sees fear of missing out as a driver of poor decisions in easy-money markets and fear of looking foolish as a force that can freeze investing after a downturn. He says a disciplined process should ask whether a particular investment is sound in the current moment.

He describes GV’s $100 million investment in Stripe during 2020 as an example of investing amid fear. He recalls a roughly $32 billion valuation but is unsure whether it was announced. He expects Stripe to grow substantially over a decade; Stebbings questions whether the prospective return compensates for opportunity cost. This is a forecast, not a realized return.

Asked whether founders need industry experience, Hulme says it depends. He contrasts GV-backed Lemonade, whose founders brought in insurance specialists, with Currencycloud, whose founder had deep fintech experience. Outsiders must be willing to learn and recruit missing expertise.

He defines execution speed as choosing and answering the important unanswered questions efficiently. A first version needs enough quality to generate meaningful feedback, not perfection; activity should produce evidence that guides the next step.

He argues early business customers should often be charged because free use may not reveal whether a product is valued, though consumer markets can differ. The goal is fast, accurate feedback from customers representative of the intended market.

Hulme worries about cultural debt: habits of mistrust, excessive hierarchy or weak accountability can become hard to reverse. He believes some companies used major restructuring to reset their culture, but the appropriate decision depends on the organization.

Remote work has different outcomes depending on how deliberately it is designed, he says. He cites GV-backed GitLab as a remote-native example with deliberate asynchronous practices, while describing difficulties he observed in companies abruptly forced remote during the pandemic. These are observations, not controlled comparisons.

Hulme thinks in-person work offers early-career employees valuable informal feedback and shadowing, while acknowledging cost can limit attendance. Stebbings adds that hybrid teams benefit when colleagues coordinate the days they are together; they do not present a universal study of productivity.

In quick fire, Hulme praises the board of Nothing and predicts roughly $600 million in revenue for the year. As a board member, he has an interest in the company; the figure is a forecast from the interview, not a reported result.

He compares attention to climate risks with attention to artificial general intelligence and speculates that forecasts of imminent AGI correlate with fundraising needs. He offers this as his impression, not evidence that particular researchers are insincere; GV and Alphabet have their own AI interests.

He recalls GV’s investment in Neuralink and his impression of its founders’ pitch and technical ambition. His enthusiasm reflects an investor’s perspective and does not establish medical efficacy or future outcomes.

Hulme says advances in computer vision, models and components have made him more open to robotics investment, and argues for working with the military amid geopolitical risks. These are his judgments and forecasts, not established outcomes.

Parenthood has led him to place more weight on innate personality, though he calls his analogy to investing a stretch. He believes investors can support founders but cannot fundamentally remake them; this is a personal view, not a scientific conclusion.

He closes by saying GV has backed Snyk founder Guy Podjarny again in a new, then-unannounced venture. Hulme presents his confidence in a founder he already knows, and thanks Stebbings for the conversation.