Eventbrite, Databricks and the Search for Durable Software Growth

Harry Stebbings, Jason Lemkin and Rory O’Driscoll discuss reported AI partnerships and valuations, the growth limits of software markets, and the challenge of investing in durable companies. Their company metrics, deal interpretations, security assertions and predictions are not independently verified. The speakers disclose relevant investment and business interests.

Stebbings asks about a reported Thrive and OpenAI partnership. O’Driscoll says subsequent reports of a “code red” and renewed focus on the core product may be more consequential. His description of internal priorities is an interpretation of reporting, not an independently confirmed company plan.

Lemkin views the partnership as an example of a venture investor deepening a relationship with a major winner, while acknowledging he does not understand the equity terms. Stebbings says the arrangement involves OpenAI and Thrive Holdings; the discussion does not establish the full transaction structure.

O’Driscoll thinks the association may benefit Thrive’s profile, but says OpenAI’s advantage is less clear. He mentions a press-release reference to data and speculates about relationships between the principals. His assessment of motives and terms is not verified.

Stebbings cites rumors of a $5 billion Databricks raise at a $134 billion valuation and reported 2025 sales and growth. O’Driscoll compares the suggested price and growth rate with Snowflake, asking how much extra revenue multiple an investor should pay for faster growth. The financing and company figures are not confirmed by the interview.

O’Driscoll argues that compounding can make a growth premium valuable if the higher rate lasts several years. He notes that a thin set of public comparables makes the calculation difficult. Lemkin says accelerating growth at scale is unusual; neither can establish that it will continue.

They discuss reacceleration and the limits of valuation models: extrapolating ever-increasing growth mechanically gives implausible results. O’Driscoll cites rough frequencies of reacceleration without a disclosed dataset. Both emphasize the role of sustained growth and the size of the addressable market.

Stebbings says a later investment in a fast-growing company may be more attractive on a risk- and time-adjusted basis than an uncertain early round. Lemkin agrees high seed prices can be difficult to justify. These are portfolio opinions, not a recommendation or proof of a three- to fivefold return.

Asked whether Databricks and Snowflake can coexist, O’Driscoll describes different origins and overlapping markets. He expects prolonged competition rather than either firm simply disappearing. His comparison and prediction are not an independent technical assessment.

Lemkin discusses new ways to access enterprise data with AI tools and says he cannot predict the resulting changes. The speakers distinguish what might be possible from what is currently deployed, including access controls and enterprise data limitations.

Stebbings asks whether CRM platforms will become databases beneath independent agents or develop their own agent tools. Lemkin points to Salesforce’s investment in Agentforce as evidence of intent, while Stebbings notes that intent does not guarantee execution. Both see an uncertain transition.

O’Driscoll thinks some customers may prefer an agent bundled with their existing CRM, while larger enterprises may combine data from several systems and build custom agents. This is an architectural hypothesis, not a claim that any architecture is inherently secure or suitable for every company.

Lemkin raises security and data residency concerns and describes alleged access restrictions and security incidents involving named software vendors. These accounts are not independently established in the episode; the discussion does not determine causes, scope, responsibility or the status of any incident.

O’Driscoll notes that restrictions affecting newer AI vendors could follow issues involving older software suppliers. The group speculates that incumbents might use security concerns to favor their own products. They do not establish that any named company acted for that reason.

Stebbings contrasts fast-growing AI valuations with reported lower valuations for Eventbrite and PagerDuty, citing a reported Eventbrite acquisition and transaction premium. The quoted multiples and deal status are presented as current news at recording time and are not independently checked.

O’Driscoll says a low-growth public company receiving a premium offer may face difficult board and fiduciary questions. He speculates a buyer might see opportunities to improve assets or add products. His remarks are general investment commentary, not legal advice or evidence of board deliberations.

Lemkin cautions that combining an established software business with a hot AI startup is easier to discuss than to demonstrate in operating results. O’Driscoll recalls evaluating PagerDuty years earlier and argues adjacent products and operational resolution could expand its opportunity; these are proposals, not a confirmed transaction.

Stebbings asks whether the apparent ceiling of a widely used product changes how investors assess market size. Lemkin says some mature software companies have not expanded as far as he expected, calling this a “TAM trap.” The claim is his interpretation of growth, not proof that all software markets are saturated.

O’Driscoll offers a different explanation: many software markets and adjacent categories already have strong venture-backed competitors, leaving less room to expand. He says paying a high price requires a sufficiently large opportunity and uses Zoom as an illustration; his claims are not a complete market census.

The question for new AI products, the panel says, is whether they can tap labor budgets or deliver enough value to justify higher prices. Lemkin gives examples of software pricing, while O’Driscoll warns that multiple suppliers might compete away those premiums. Future pricing and adoption are uncertain.

O’Driscoll suggests companies consider a second product before reaching the limits of the first market. He describes a portfolio example of new products layering into growth without naming it. This is an investor’s account, not proof that diversification always succeeds.

They discuss seat-based software pricing and a reported Workday concern about declining seats. Lemkin expects companies to increase revenue per employee and says this may pressure seat counts; his company-specific efficiency estimates and employment forecasts are not independently validated.

O’Driscoll distinguishes seat pricing, usage pricing and charging for value delivered. He expects AI could make a per-seat model less relevant for some automated work, though other products may still price per employee served or active user. Measuring delivered value could be difficult and pricing outcomes remain open.

Stebbings asks whether startups must deliver both high growth and efficiency. Lemkin distinguishes revenue per employee from bottom-line profitability and says some rapidly growing portfolio companies can attract capital even when investors are less focused on immediate earnings. His account reflects a subset of companies.

O’Driscoll separates mature companies focused on cash flow, model developers spending heavily on compute, and AI applications that may grow quickly with small teams. These are broad categories, not a census, and their reported margins and growth profiles differ across firms.

He observes that all three categories in his framework may hire fewer people than past technology cycles, while maintaining longer-term optimism about adaptation. Lemkin is concerned about employment. Their near- and long-term labor forecasts are uncertain and do not measure actual job losses.

The group discusses a new Google app-building competitor and whether model providers can use distribution to challenge startups. Lemkin says the version he tried lacked features he wanted but notes the speed with which competitors appeared. His product assessment reflects an individual experience at the recording date.

They debate whether large model providers will enter many specialized applications. Lemkin suggests wealth management and estate planning as possible areas where a startup might have room; O’Driscoll discloses an investment in Range and thinks model suppliers may instead prioritize their core products. This is speculation, not a verified roadmap.

O’Driscoll describes a potential AI-assisted wealth-management service for people who cannot justify expensive bespoke help. The guests identify a possible business opportunity, but the claim that AI can reliably handle specific legal, tax or medical tasks is a hypothesis, not evidence of professional accuracy or safety.

Lemkin warns that a lower-cost service still needs a realistic addressable market and asks how many customers would pay its suggested price. O’Driscoll agrees that customer segments matter, and discusses Wealthfront and Betterment as examples of businesses that may take a long time to compound.

Lemkin describes frustrations with wealth managers, trusts and tax referrals; O’Driscoll says integrating services could be valuable. Their criticism and customer experiences are personal accounts, not an independent assessment of named institutions or financial advice.

O’Driscoll argues that some financial businesses take 10–15 years to build because lower fees require more assets and less wealthy customers accumulate capital over time. He contrasts long compounding with rapid venture markups; this is an investment thesis rather than a guaranteed result.

Stebbings argues venture investors face pressure to remain relevant through rapid rounds, media attention and prominent follow-on investors. Lemkin says limited partners may accept a different path if returns are strong. Their claims about fundraising incentives reflect the speakers’ experience and are not a systematic survey.

O’Driscoll says quick valuation increases may show momentum but do not resolve whether a durable large company can be built. He favors conviction in the ultimate business over filtering solely for speed. His reported follow-on financing statistic is not accompanied by a full dataset or method.

When Stebbings asks whether a focused investor can compete for obvious AI winners, O’Driscoll says one can pursue both highly competitive deals and differentiated opportunities. Competitive rounds may require paying market prices; both strategies have selection and valuation risk.

Lemkin describes his own interest in using AI to reduce friction in wealth and estate management after a frustrating personal experience setting up trusts. He speculates about market size, and O’Driscoll expands on possible service models. Their examples are business ideas, not evidence that automated legal or tax advice is reliable.

The speakers discuss capital discipline in less fashionable markets: a founder may need to demonstrate progress before raising more money. O’Driscoll says capital may follow proof, while acknowledging the risk of failing to prove the model. This is a framework, not a guarantee of funding.

In a hypothetical choice at cited valuations, O’Driscoll picks Lovable, arguing the application layer might capture more value if the category succeeds. Lemkin picks Supabase because he prefers the defensibility of a difficult database problem. Stebbings is an investor in Lovable, a relevant interest; these preferences are not recommendations.

Lemkin acknowledges that open-source database technology can be replicated but says operational reliability and switching costs can be significant. He wants more defensibility in a fast-changing market. The discussion closes with the speakers thanking one another and their audience.