Bessemer’s Byron Deeter on AI Growth, Venture Scale and Liquidity

Harry Stebbings interviews Bessemer Venture Partners partner Byron Deeter about AI, software investing, venture fund strategy and liquidity. Deeter discusses Bessemer investments including Anthropic, Perplexity, Canva, Intercom and ServiceTitan, so his assessments and return forecasts carry financial and professional interests.

Deeter says he is energized by the pace of AI development after a difficult period of market declines and layoffs. He believes the technology could produce major companies and economic change. Those are optimistic forecasts, not established outcomes.

Stebbings worries that impressive AI demos quickly become ordinary and defensibility is disappearing. Deeter instead emphasizes rapid iteration: he expects strong teams to take advantage while weaker companies are passed more quickly. He believes valuable businesses can emerge at both model and application layers, though that is not guaranteed.

On margins, Deeter says he cares about the future margin profile more than demanding mature gross margins immediately. He cites software businesses that invested heavily before improving economics. Future margins depend on pricing, costs and execution; a low current margin alone does not prove they will improve.

He says Bessemer has invested nine-figure sums in each of Anthropic, Perplexity and Canva while holding less than a traditional venture ownership target. Stebbings points out the dilution and opportunity cost of backing capital-intensive firms. Deeter accepts the risk and says the investment case depends on much larger future outcomes.

The host gives a hypothetical example of dilution reducing an investment’s multiple. Deeter says Bessemer believes Anthropic could become a generational company and suggests a much higher multiple might be possible, while acknowledging such businesses could still go to zero. Bessemer is an investor; private marks and hypothetical multiples are not realized returns.

Stebbings asks whether investing outside the largest AI deals still makes sense given concentrated financing. Deeter expects strong venture opportunities around the major labs as well as within them. His projections of large funding totals and many high-multiple outcomes are estimates, not proof of future returns.

Asked whether vertical SaaS is dead, Deeter says Bessemer believes AI can create another growth cycle. He points to industry data, supply-chain collaboration and marketplaces, and compares the opportunity with payments added to software platforms. Bessemer has interests in several examples he discusses.

Stebbings asks whether the shift favors an established company over a tiny startup. Deeter concedes incumbents may have distribution, data and platform advantages because AI often builds on existing cloud systems. He still believes well-executing challengers can win, a prediction rather than a conclusion about every market.

He praises portfolio companies Canva and Intercom for adapting at scale. Deeter recounts a company using Intercom’s Fin for customer support and reporting more automated resolutions and higher satisfaction. This is an investor’s anecdote about a portfolio product, not independent evidence of typical performance or employment effects.

Stebbings asks whether AI software can reach labor budgets rather than just technology budgets. Deeter believes it already addresses services and labor spending, citing accounting, legal and medical workflows. The potential size and distribution of that spending remain uncertain.

In health care, Deeter cites Abridge and argues AI could help clinicians spend less time on documentation. Stebbings asks how a strong incumbent such as Epic could respond. Deeter favors open access and predicts broad benefits, but the interview does not establish clinical accuracy, safety or patient outcomes; patient data access also requires appropriate protections.

Stebbings cites workforce reductions alongside revenue growth and asks if this means companies will do more with fewer people. Deeter favors framing the change as helping workers focus on higher-value tasks, while acknowledging disruption and uncertainty around entry-level work. His emphasis is a view, not a measurement of net jobs created.

Stebbings pushes back that young workers may face a severe reduction in opportunities, not merely better tools. Deeter draws on earlier technological transitions to argue new work can emerge. He agrees disruption is coming and predicts society will adjust positively; the historical analogy cannot establish the speed or distribution of future employment outcomes.

Deeter imagines AI helping education, medicine and research, while saying he does not subscribe to a completely work-free utopia. His examples are potential future applications, not evidence that particular systems are effective, safe or equitably available.

The host asks whether the old triple-triple-double growth standard is now too slow. Deeter says some exceptional AI businesses have reached $100 million in roughly 18 months, while a broader group has taken around four years. He stresses these are rare profiles; figures and selection come from Bessemer’s analysis, not all startups.

He cites Anthropic’s progression through reported revenue milestones and a possible future $10 billion outcome. Bessemer invests in Anthropic, and the prospective milestone is a company-related forecast rather than audited realized revenue. Strong early growth does not establish durable profits.

Stebbings asks whether founders should raise as much as possible and ignore margins. Deeter says no: efficiency still matters and businesses ultimately need sustainable free cash flow. He discusses Bessemer’s “Rule of X,” which weights growth relative to efficiency for a mid-stage comparison; it is an analytical model, not a universal valuation law.

Deeter expects technology and hardware improvements to continue reducing costs and extending capabilities. He sees more suppliers alongside Nvidia and believes scaling trends will persist. These claims are forward-looking and sensitive to technical and economic constraints.

Stebbings says crowded markets reward companies that raise aggressively and attract attention. Deeter argues compelling products can spread through users and that product-led growth can support rapid scale better than relying only on a large sales force. He cites Perplexity, a Bessemer investment, as an example; that is an interested assessment.

The host challenges Bessemer’s pricing discipline in expensive rounds. Deeter says the firm pays market-clearing prices but still seeks a plausible route to sustainable unit economics. He acknowledges the tension between short-term discipline and backing unusually ambitious founders.

He recalls passing on Tesla because he could not see its early Roadster economics working. In hindsight, he says he underestimated the founder’s ability to create later opportunities, while noting the importance of outside financing. This is an investor’s retrospective interpretation, not a complete causal history of Tesla.

Asked about selling private stakes, Deeter says he has become more open to secondary liquidity as companies stay private longer. He argues some limited partners and emerging funds need capital returned and that selling need not signal trouble. Secondary sales can provide liquidity but their price and consequences depend on the transaction.

He hopes IPO activity will improve and notes that private growth and public-market valuations should be compared with appropriate growth and liquidity adjustments. Whether markets reopen on the timetable he expects remains uncertain.

Deeter says Bessemer’s Cloud 100 list has over $1 trillion in combined private market value and believes the leading marks are credible. The sum reflects private valuations, not cash proceeds or independently verified exit values, and Bessemer owns stakes in some of the companies.

Stebbings questions whether some of that value is hype and asks why Canva has not listed sooner. Deeter says its founders take a long-term view and expects a public listing eventually, without announcing a date. Bessemer has invested hundreds of millions in Canva, a material interest in his assessment.

Deeter names StubHub, Canva and Anthropic among Bessemer’s largest investments and says the firm has expanded its ability to invest hundreds of millions. The amounts describe capital invested, not fund profits or distributions.

He says it can be psychologically difficult to invest again at a much higher price after an earlier stake has appreciated on paper. Bessemer brings in another team to reassess whether a new investment has sufficient upside. An earlier mark does not ensure a later round will earn a return.

Stebbings admits he has hesitated to follow on after a sharp valuation increase. Both say a higher price need not automatically make the next investment unattractive, but they also recognize that new capital must earn its own return. Their examples are investment judgments, not recommendations.

Stebbings raises the danger of investors pushing too much capital into a company. Deeter agrees overfunding can distort spending even if strong investors and adequate financing can help. He recommends evaluating how capital will be used and whether the company can execute its plan.

Asked about missed follow-on investments, Deeter names Procore and ServiceTitan and says Bessemer underestimated how payments could expand their markets. He contrasts these with Twilio, where he says the firm doubled down more aggressively. Hindsight about portfolio winners does not prove a follow-on would always have been profitable.

He says Bessemer owned about 28% of Shopify and Twilio at their IPOs and less of Procore and ServiceTitan. These are historical ownership statements, not net fund returns, and the interview does not give a complete accounting of cost, dilution or distributions.

Bessemer requires scenario analysis in its investment recommendations, Deeter says, though past upside cases often underestimated eventual outcomes. Stebbings questions whether unreliable market-size forecasts harm decisions. Deeter acknowledges the problem but still looks for exceptional teams and plausible adjacent markets rather than abandoning the exercise.

Stebbings says Bessemer sold Shopify too early. Deeter corrects him: the firm distributed shares to limited partners, who could then decide whether to sell, although he says many sold and the firm wishes it had held longer. A distribution of stock is not the same as a sale or a guaranteed cash return.

On evergreen funds, Deeter sees benefits in staying invested longer but says public-market management and fund economics differ. Some limited partners want capital returned so they can manage their own allocation. He treats hold duration and liquidity as trade-offs rather than claiming one structure is always superior.

Stebbings asks whether venture is now simply a game of scale, contrasting a specialist luxury brand with a mass retailer. Deeter rejects the retailer analogy but agrees scale matters. He compares Bessemer’s multi-stage, multi-sector global platform with specialist firms and expects both models to have opportunities.

Stebbings favors deep specialization in areas such as health care and cybersecurity but is skeptical of many thematic funds. Deeter largely agrees and warns against continuing to deploy into a sector simply because a fund is dedicated to it. Both are describing investment preferences; specialist and generalist performance varies.

Deeter recalls being warned that his first venture investment would be poor and says the warning proved right. He argues investors need enough opportunities across cycles to learn, while Stebbings says spreading investments over time can reduce exposure to a single overheated vintage. Diversification does not guarantee returns.

On spinouts, he sees a period when investors sought a reset and predicts some new firms will build durable platforms. He also thinks established firms can benefit from networks and persistence, citing a historical estimate he cannot recall precisely. That rough figure is not an independently established forecast for particular funds.

Deeter expects liquidity from several sources: private equity, strategic acquisitions, IPOs and secondary sales. He sees opportunities for private equity to improve the efficiency of mature software companies. The scale and timing of each exit route remain uncertain.

He predicts strategic buying and IPO activity could recover, and expresses approval of what he perceives as a more permissive antitrust stance. These are a venture investor’s market and regulatory opinions, not a prediction of any specific approval or a legal finding.

Asked whether Tiger helped or hurt the ecosystem, he says it did both: it brought capital and ambition but also some overfunding and weak governance. Stebbings expects better returns than critics suggest, and Deeter agrees some winners could offset losses. Their portfolio-return scenarios are speculative and not audited fund results.

In quick-fire questions, Deeter credits colleagues for sourcing and selecting unusual opportunities. Asked to choose investment firms as a limited partner, he favors a mix of established and newer approaches and says prior platform experience can matter. The interview does not establish future fund performance.

He says his own style emphasizes teams and intuition and admits that he sometimes misses details. After selling his prior business, he considered whether to retire but chose to keep working because he enjoys helping founders. His personal motivations and approach do not remove the need for diligence.

Deeter says he now sees a much larger potential AI market and predicts multiple trillion-dollar companies. Asked whether Anthropic will exceed $1 trillion within three years, he says yes and notes Bessemer has bought shares. This is an explicitly interested, highly uncertain valuation forecast, not a realized return or investment advice.