OpenAI Secondaries, Ramp and the Price of AI Growth

Harry Stebbings hosts investors Jason Lemkin and Rory O’Driscoll and Twilio co-founder and former CEO Jeff Lawson. They discuss founder compensation, AI valuations, reported employee liquidity, acquisitions and the incentives shaping venture capital.

The panel begins with the proposed Tesla compensation package for Elon Musk and its trillion-dollar headline. O’Driscoll interprets the performance targets as a board’s decision to keep betting on Musk and pursue very large additional market value. His reading of the proposal is not a statement that the award has been earned.

O’Driscoll says the package includes demanding market-cap, operational and profitability conditions. He interprets the board as paying for both an expanded existing business and entirely new products; whether those conditions will be met is uncertain.

Lawson says that as Twilio’s founder he already held substantial equity and preferred simple compensation for himself and his team. His view is that once pay is perceived as fair, adding many variable conditions can create new disputes over fairness rather than focus people on work.

Stebbings asks whether giant founder packages will become the norm. Lemkin says he sees more ambitious founder incentives in private companies; O’Driscoll expects the Tesla proposal to be an extreme rather than an ordinary package, though he thinks it may raise others’ expectations.

Lawson asks whether the board also fears the downside of Musk leaving. O’Driscoll agrees that the market might assign substantial value to his presence. His hypothetical suggestion of a 75% share-price fall is an opinion about a possible reaction, not a measured result or forecast that it will occur.

Lemkin contrasts the current market with founders who felt they already had enough equity and focused on the team. He sees intense competition for AI talent changing incentives. Lawson worries about founders leaving their own companies for lucrative offers, arguing that a startup’s leader should be committed to its mission.

O’Driscoll offers a different perspective on transactions involving Scale AI and Windsurf: he thinks some offers may have represented attractive exits for shareholders. He raises concerns about employees left behind and says he does not know whether they were fully taken care of. His description of the future of those businesses is a personal opinion, not a verified outcome.

Stebbings notes that investors’ gains can flow to beneficiaries such as universities and foundations. Lawson and O’Driscoll distinguish those financial outcomes from a founder’s relationship with the team; they debate whether a founder has additional responsibilities even when a transaction rewards investors.

Stebbings cites Ramp at $1 billion ARR and Brex at $700 million, with Brex growing 50%, and asks whether the entire market is rising. O’Driscoll says the companies may be doing well without representing every business. The figures are the panel’s claims at the time, not independently verified current results.

O’Driscoll notes that credit-card and interchange economics differ from conventional high-margin software. He says valuation should ultimately reflect risk-adjusted cash flows: growth can justify a premium, but financial-service margins and funding risks still matter.

Lemkin sees AI spending spreading across technology infrastructure and believes B2B companies can benefit from it. Lawson, who says he uses Ramp, adds that customer spending may reflect venture funding and market-share gains, not necessarily durable success for every customer.

Lawson recalls Twilio serving fast-growing mobile companies that paid for infrastructure even when some later failed. The provider could earn revenue during the boom, but had to replace that revenue when customers disappeared. His experience illustrates both the opportunity and the dependence on customer survival.

The panel discusses Sierra at a reported $10 billion valuation and $100 million ARR. Lemkin praises founder Bret Taylor, while O’Driscoll says a large market and strong operator do not by themselves settle whether the entry price compensates an investor for risk.

O’Driscoll believes AI customer support could be a large category, but his suggestion that AI may handle 70–80% of the work is a forecast, not a demonstrated outcome. At a hypothetical 100-times ARR valuation, an investor must still consider competition, execution and the price paid.

Stebbings considers the opportunity cost of a large check in a single company. Lawson notes that a founder’s own commitment is often even more concentrated: building a company can consume nearly all of an entrepreneur’s time and financial attention.

Stebbings cites a reported $100 million Kleiner Perkins investment in Anthropic’s $13 billion round at a $183 billion valuation. The group debates whether late-stage investment is merely buying association with a leader or a rational risk-adjusted strategy. They do not verify the transaction terms.

O’Driscoll says rapid growth can make an expensive round work if it persists, while warning that investors may over-extrapolate and let valuation risk replace other risks. Lemkin questions whether this resembles traditional early venture capital; the distinction is their analysis, not a claim about the firm’s motives.

The panel discusses a reported $10 billion OpenAI secondary sale and its possible effects on employees, housing and entrepreneurship. A secondary transaction transfers shares from existing holders; the speakers do not establish that $10 billion was invested into the company or how much each employee received.

Lawson recalls worrying that Twitter’s IPO would affect housing prices, but says he cannot attribute the housing market’s later movement to that event. He hopes the Bay Area will build enough housing to accommodate new wealth without displacing others, and notes that liquidity may make it easier for employees to start companies.

O’Driscoll observes that a large secondary sale can be understood as diversification by people with concentrated holdings, especially if a company is worth hundreds of billions. His numerical example is hypothetical; private-company sales are simply more visible as a single event than dispersed public-market selling.

Lemkin worries that large AI compensation and liquidity events make it harder for other startups to hire. The panel suggests companies may recruit outside the most expensive talent markets rather than compete for exactly the same workers.

The panel turns to reported payments to authors connected with Anthropic’s copyright litigation. O’Driscoll describes his understanding of the court proceedings, while Lemkin criticizes the alleged use of pirated books. Their discussion is not a definitive statement of copyright law or a verified breakdown of a settlement or damages.

The speakers recognize that other questions remain, including disputes over the use and reproduction of creative work. Their views on whether conduct was justified and how future cases may be resolved are opinions; the discussion does not establish the outcome of other litigation.

Stebbings cites a reported Mistral financing at a $14 billion valuation and ASML’s participation. O’Driscoll explains ASML’s role in semiconductor manufacturing equipment but says he does not know the company’s rationale. The reported valuation is not the amount ASML invested.

Lemkin considers why corporations with cash might make strategic investments, including accounting treatment and access to an ecosystem. Lawson and O’Driscoll ask whether an investment actually improves the core business and note that cash tied up in a cyclical industry also has an opportunity cost.

O’Driscoll speculates that some countries and companies want national AI champions for strategic autonomy. He compares this with defense procurement but says he is not convinced sovereignty is generally a sufficient business model for technology. His explanation of ASML’s motives remains speculation.

Stebbings discusses Atlassian’s reported $610 million cash acquisition of The Browser Company. He discloses that Atlassian also acquired Cycle, one of his seed investments, for about $21 million in cash. The participants question whether a work browser is the best answer to Atlassian’s AI challenge.

Lemkin observes that companies and investors can feel pressure to act during a transition, potentially choosing the best available deal rather than an ideal one. Lawson says the value of an acquisition depends on the strategic problem it solves, and the group cannot know Atlassian’s internal reasoning.

Lawson explains that Twilio’s acquisitions were intended to build beyond messaging before that business eventually matured. He says the company could use its customers and revenue as a bridge into new areas, although timing and successful integration were uncertain.

Lawson predicts AI could reduce demand for software sold per human seat, and questions whether a browser acquisition addresses that risk. This is his view as a former Twilio CEO, not evidence that Atlassian will lose a particular share of its revenue or jobs will disappear at a particular rate.

The panel considers the innovator’s dilemma: incumbents with seat-based revenue may hesitate to sell automation that reduces seats, while a new entrant has no existing business to protect. Lawson sees Twilio’s infrastructure model as a different starting point because it was not principally selling software seats.

Lawson says public companies can have capital and customers but may struggle to invest for the long term while addressing immediate growth pressure. He thinks the opportunity to navigate an AI transition is exciting, though he personally prefers founding a company to serving as a hired CEO.

The participants disagree over whether automation revenue will replace software spending one-for-one or capture part of the larger labor budget. Both are hypotheses; the answer matters for startup valuations and for the revenue incumbents may lose.

Asked which public-company leaders might acquire aggressively, Lawson points to Atlassian’s history and discusses opportunities and constraints at Dropbox and Box. He says those companies may need product breakthroughs, but does not claim to know which strategy will succeed.

Lawson describes a product challenge at Twilio: a messaging API specifying sender, recipient and message leaves limited room to add value beyond delivering precisely what the developer requested. He contrasts this with Cloudflare’s position, where features can be offered at an existing point in the customer’s workflow.

Looking back at developer platforms before AI, Lawson groups potential breakout businesses into three types: business development as a service, capital expenditure as a service, and unusually difficult algorithms as a service. It is his framework from experience, not an exhaustive taxonomy.

Twilio, Stripe and AWS can let a developer access relationships or infrastructure that would otherwise require corporate negotiations or large purchases. A working product creates a reason for the business to pay, while the provider handles complexity behind the API.

Algorithm services, Lawson argues, need to solve problems difficult enough that customers cannot easily justify rebuilding them. If a service becomes expensive and its function is reproducible, an engineering team may attempt to create its own version.

They debate whether AI inference fits that model. Lawson notes that open-weight models can be run by customers, while training a frontier model remains difficult and expensive. The competitive outcome depends on capability, operating cost and customer choice rather than a fixed rule.

In a quick-fire round, the panel gives different hypothetical Figma share prices one year ahead and disagrees on whether Canva will list in the next quarter. These are guesses at the time of recording, not investment guidance or confirmed corporate plans.

Stebbings raises reports of fraud allegations involving the former IRL CEO. Lemkin calls for consequences when fraud is proved and argues rapid fundraising can undermine diligence. An arrest or allegation is not a conviction, and the panel does not establish the facts of the case.

O’Driscoll agrees deliberate lying can warrant serious consequences but distinguishes forged documents from ambiguous revenue classifications, saying each case depends on facts and intent. He does not offer a finding about the specific case discussed.

Lawson argues investors also have responsibilities: if they rush to invest without checking basic claims, they increase the chance of being misled. His observation that he had not heard of some alleged fraud companies is a personal heuristic, not evidence of wrongdoing.

O’Driscoll recounts a diligence practice of independently checking a company’s cash with its bank rather than relying only on a statement. He says sophisticated investors who manage other people’s money should exercise judgment and slow down when basic claims need verification.

The panel closes by recognizing that responsibility for proven wrongdoing and the duty of investors to conduct diligence are separate questions. The speakers thank Lawson for joining the discussion.