Summary

Meta’s $14.8 billion Scale AI transaction was framed as a purchase of relevance, talent and strategic knowledge—not an operating business worth the headline price. Meta received a 49% non-voting stake while the cash flowed to existing investors as a special dividend, leaving it exposed to an $800–900 million revenue base the panel expects to shrink. Harry’s read: with Meta behind in AI and the price below 1% of its market cap, Zuckerberg could simply “roll the dice”; Rory agreed there was no supporting DCF beyond “somebody paper the file.”
Scale’s competitive conflict is already redistributing hundreds of millions of dollars across the human-data market. Handshake’s Garrett said demand tripled within a week, with delivery capacity—not customer interest—the constraint; he called audience access “the only durable moat in the entire human data business.” Jason declared Scale “a dead man walking instantly,” while Rory sharpened the cause: not merely losing a founder, but asking OpenAI, Anthropic and other labs to trust a supplier whose 49% owner is their direct competitor.
The deal creates unusually immediate venture liquidity, but it does not solve the system-wide LP shortfall. Unlike IPO holdings distributed after lockups and potentially over 36 months, Scale’s $14.8 billion was already cash; Excel was said to have made $2.5 billion, while Paige Craig’s pre-seed produced a 1,000x return. Rory nevertheless put even $20 billion against “a couple of trillion” of venture NAV: roughly 1%, or “a good start” rather than a reset.
The founder-replacement debate produced a strong default—keep the founder—but not an absolute rule. Jason cited an analysis in which 90% of B2B IPOs still had founder-CEOs and said he would accept a more certain $10 billion Discord outcome with Jason Citron over gambling on a replacement to reach $20 billion. Rory called CEO replacement “open heart surgery” with “a one in three chance of dying,” yet defended boards that act when fiduciary duty demands it: the statistically correct button is “never change,” but no director can press it blindly.
Ramp’s $16 billion round works as low-dilution capital and a recurring momentum signal. The $200 million financing meant roughly 1% dilution, while Ramp’s card-plus-software model may require substantial balance-sheet funding; Rory estimated that $1 billion of revenue could imply roughly $4 billion of financed receivables. Jason’s objection was valuation quality—private markets often price 20–40% gross-margin fintech revenue like 80% gross-margin SaaS—but Harry persuaded Rory that frequent rounds also manufacture attention and reassure customers that “the energy” is not leaving the product.
OpenAI’s most consequential negotiation may turn on an undefined word: AGI. Rory initially gave Microsoft the leverage because the relationship is non-existential for it; Jason argued Microsoft has “no leverage” if profit sharing and related rights expire when AGI is reached. Rory’s memorable resolution was that “we’re gonna hit AGI when Microsoft and OpenAI litigate what AGI is,” because potentially half a trillion dollars of value now depends on a term the industry uses “loosey-goosey.”
Chime’s 50% debut was treated as evidence that the IPO window is open, though it may matter more for the $200–500 million revenue flotilla than for Databricks or Stripe. Rory described attractive pricing and first-day pops as effectively having “bribed the buying public to like IPOs again,” reversing memories of 30–40% average losses and occasional 90% drawdowns from the prior cycle. Databricks can already behave like a public company and file abruptly, but Rory could still imagine the year ending without either Databricks or Stripe choosing to list.
Enterprise incumbency only helps where an incumbent owns the relevant installed base—and the panel sees new AI vendors exploiting every gap. Rory argued Dropbox could ship a perfect Glean equivalent yet remain disadvantaged because it lacks an enterprise-wide customer base; Jason said he was “starting to lose confidence in the old guard.” In the closing bets, Rory and Jason still leaned heavily into equities and moved to “yes” on a Chinese model temporarily reaching number one: “There’s no world where we’ve got all the smart people and they don’t.”

Deep dive

1. Meta paid for AI relevance, not Scale’s cash flows

Rory’s level-set: Scale spent a decade progressing from basic labeling—“Is this a stop sign?”—to recruiting domain experts and PhDs for post-training reasoning. That produced an $800–900 million business serving roughly five or six frontier-model providers.
The transaction put about $14.8 billion into Scale for a 49% non-voting interest, then allowed investors to withdraw the money through a special dividend while the CEO and other key executives moved to Meta. It did not repurchase Meta’s shares or leave the cash funding Scale.
Harry’s explanation was blunt: Zuckerberg was behind with Llama, needed to demonstrate that Meta remained an AI front-runner, and spent less than 1% of market capitalization to acquire talent and a message. “Honestly, he doesn’t give a shit.”
Jason compared the transaction with Salesforce’s $750 million purchase of Quip to recruit Brett Taylor and Google’s roughly $400 million Bebop deal for Diane Greene. Rory’s broader point: seven companies can write enormous checks that consume only a quarter’s free cash flow—Meta was worth about $1.7 trillion and generated roughly $15 billion quarterly.

2. Scale’s conflict has opened a land grab in human data

Every major Scale customer must now assess whether its sensitive post-training questions are visible to a supplier “owned, controlled, dominated, infiltrated—pick your word” by a competing model builder. Jason connected that reaction to Anthropic cutting off Windsurf.
Garrett reported that Handshake’s demand tripled in the week after the announcement and said “hundreds and hundreds of millions of dollars” were seeking reallocation. The binding constraint was whether alternative providers could deliver enough volume quickly.
Frontier work is broadening beyond text into audio, tool use and agentic trajectories, while demand concentrates in science, finance, law and medicine. Customers balance quality, volume and speed; Garrett’s central claim was that “the only durable moat” is access to an audience, which Handshake has through one million employers.
Asked whether Scale would still exceed $100 million of revenue in a year, Garrett said yes—but also confirmed major spend migration. Jason predicted it could retain perhaps $200 million of its $800 million for 24 months; Rory agreed the competitive ownership signal, more than one executive’s departure, makes an independent recovery difficult.

3. Scale delivered instant liquidity through a nearly irreversible structure

Harry emphasized the unusual feature for LPs: cash returned immediately, without an IPO lockup or staged distribution. Jason contrasted that with Chime, Circle and other public holdings that venture funds may distribute over 36 months.
Rory said liquidity “has gotta help,” but contextualized the scale: rounding the transaction up to $20 billion still represents only about 1% against a couple of trillion dollars of venture NAV. LPs typically recommit only after dollars actually arrive.
The panel described the structure as a clever regulatory “get-around”: the deal closed, executives changed employers and Meta acquired a 49% non-voting stake rather than a conventional operating asset whose unwinding would be simple. If the true assets were people and market perception, judicial unwinding would be exceptionally awkward.
The immediate winners were unmistakable. The panel credited Excel and Daniel Levine with roughly $2.5 billion, while Paige Craig’s pre-seed reportedly returned 1,000x; Rory expects competitors to respond through selective insourcing or smaller acquisitions, not another replica of Meta’s arrangement.

4. Keeping the founder is the correct default, not a religion

A tweet discussed on the show alleged that Benchmark might remove Discord’s Jason Citron over a delayed IPO. Rory’s starting position was strongly founder-friendly: replacing a founder costs years, creates enormous work and is usually a poor investment decision even before moral considerations.
A CEO surprised by dismissal represents a board failure, Rory argued; directors should define success, communicate concerns and let the founder participate in any transition. Even a consensual change is “open-heart surgery” with “a one-in-three chance of dying.”
Jason said 90% of the B2B IPOs he analyzed still had founder-CEOs, with almost every departure being elective. For Discord, he would prefer a relatively certain $10 billion IPO under Citron to replacing a generational founder for a chance at $20 billion: “If the founder CEO is out, I’m out” in B2B.
Rory retained the fiduciary exception, citing the controversial Uber transition: a director responsible for $10 billion of third-party capital cannot refuse every removal on principle. If forced to choose one universal default, however, he would press “never change” because the statistics favor it.

5. Ramp is financing its balance sheet and its aura

Ramp raised $200 million at a $16 billion valuation—roughly 1% dilution—while Harry contrasted Brex around $10–11 billion and Mercury at $3.5 billion with approximately $500 million of revenue. Jason saw minimal cap-table cost in repeating such rounds.
Rory characterized Ramp as a valuable hybrid of card economics and software, competing with Amex plus the accounts-payable functions of SAP and Oracle. At an estimated $700–800 million of revenue and still-high growth, $16 billion felt lofty, particularly if growth slows.
Jason’s concern was that private ARR multiples insufficiently distinguish revenue quality: fintech may ultimately trade around 2–4x revenue while high-gross-margin software deserves more. “Everyone gets the same ARR valuations,” even when gross margins range from 20–40% to 80%.
Capital still has an operating use: Ramp pays merchants before cardholders settle, potentially financing roughly $4 billion of receivables at $1 billion of revenue under Rory’s assumptions. Harry added that constant fundraises create media velocity and category confidence; Rory conceded it “galls” him, but admitted it is smart. Perplexity’s move from a $15 billion first tranche to $18 billion showed the same deal heat.

6. OpenAI must court Washington while renegotiating Microsoft

Harry described the Pentagon’s $200 million OpenAI contract as its largest award to one provider. Rory welcomed defense procurement moving toward technically advanced companies such as OpenAI, Palantir, SpaceX and Anduril instead of relying exclusively on “old fuddy-duddy companies.”
Jason relayed Sam Altman’s view that the amount was immaterial to OpenAI but treated the deal as strategically necessary: a platform with its market share must remain neutral, avoid becoming Republican or Democratic, and—excluding rogue states—“sell to everybody.” Political alignment itself is a competitive risk.
On Microsoft, Rory initially argued that ambiguity favors the party with leverage, and Microsoft can tolerate failure because OpenAI is not existential to it. Jason countered that Microsoft has “no leverage” if reaching AGI terminates the profit-sharing framework.
Rory accepted AGI as the contract’s decisive ambiguity: “We’re gonna hit AGI when Microsoft and OpenAI litigate what AGI is.” Jason floated a possible reorganization leaving Microsoft near 33%, Sam Altman at 6–7% and Microsoft with a longer tail of IP and relationship rights—the relationship matters more than cash Microsoft does not need.

7. IPO pops are reopening the market by retraining buyers

Chime rose 50% on debut, and Harry said all 2025 IPOs were up except SalePoint. Rory’s dry verdict was that the market is roaring because “there’s been none, and some is better than none.”
Closed windows reopen when good companies price attractively. Investors scarred by 2021 offerings remembered losing 30–40% on average and as much as 90%; successful new issues reverse that conditioning, effectively “bribing the buying public to like IPOs again.”
Harry argued that current demand gives Databricks and Stripe new evidence supporting a listing. Rory conceded the marginal improvement but distinguished them from the $200–500 million revenue “flotilla”: those smaller candidates are dusting off plans, while the giants could always list and may still decline the burden.
Jason noted that Databricks already held an analyst summit covering revenue, growth and operating metrics, suggesting it could file without further visible preparation. He also defended conservative IPO pricing: VCs realize value through later distributions, so a strong aftermarket can outweigh incremental dilution.

8. Gusto’s $900 million ARR validates an enormous old market

Gusto’s tender valued it at $9.3 billion on approximately $900 million of ARR. Rory emphasized the obvious but load-bearing fact: payroll is vast, recurring and universal, with around half of American workers employed by SMBs.
ADP and Paychex provide unusually supportive public comparables. The panel cited Paychex near a $55 billion market capitalization and 10x revenue, with ADP around 15x; modern, faster-growing payroll vendors can therefore justify substantial private valuations.
Rory’s honest postmortem on passing early was that he underestimated the founder, customers’ willingness to switch and investors’ readiness to reprice the opportunity rapidly. “I was wrong on that decision. Being wrong sucks.”

9. Incumbents win through distribution only where they possess it

Jason expected open LLM APIs, open-source chat frameworks and straightforward retrieval systems to let established vendors catch up. Instead, “they’re still too slow,” leaving him “starting to lose confidence in the old guard.”
Rory reframed Dropbox versus Glean as a distribution problem, not an engineering contest. Systems of record can bundle agents into installed accounts; independent agents can sit across every system. Dropbox lacks a large enterprise-wide base, so even a perfect Dash product would meet Glean from “ground zero” outside existing customers.
Salesforce throttling access to its records when customers try to run them through Glean struck Rory as “pathetically lame.” He expects the outright barrier to soften, perhaps into an API-connectivity fee, because the stronger response would be winning through a better product.
Jason estimated Slack near $2.5 billion of revenue and growing in the teens versus Salesforce’s 7–8%, making the acquisition financially defensible. But Slack’s startup-era ambition as an open developer hub is gone; at Salesforce’s roughly $30 billion ARR, smaller customers can protest lock-in, but “we don’t matter.”

10. The closing bets separated forecasts from actual positioning

Jason and Rory leaned yes on Apple announcing some U.S. iPhone assembly, while treating meaningful domestic production as nearly impossible. Rory’s cynical implementation was to make one phone domestically and end the political question; Tim Cook’s straight-shooter reputation was the main reason to doubt the announcement.
With the market odds read on air as roughly 70% for a positive S&P year, Rory preferred betting no at that price—yet held about 75% in equities. Jason was already “100% equity” personally, reinforcing their rule that an investor’s position matters more than an unpriced opinion.
Both finished at yes on a Chinese model reaching number one temporarily during the year. Rory changed his mind in real time: odds below one in five ignored several aggressive, well-run Chinese companies, and temporary benchmark parity would not equal commercial displacement of OpenAI. “There’s no world where we’ve got all the smart people and they don’t.”

Vol.65 AI’s New Era: Is Google Back?Episode 145 - June 13, 2025