Key Takeaways

  • Rory O'Driscoll framed the investment choice across three distinct tiers: Baseten at $30 billion (up 2x since June), Cognition at $70 billion to $80 billion (up from $48 billion in September), and Serval at $3 billion (up from $1 billion in December).
  • Harry Stebbings favored Serval because its $3 billion price tag still leaves room for a 10x cash multiple when measured against ServiceNow's $110 billion market cap.
  • Jason Lemkin warned that vertical AI apps often fail to expand market size, cautioning that founders and investors mistakenly assume software can charge three to four times legacy contracts.
  • Mega-funds often avoid smaller vertical winners because capital deployment constraints favor lower-multiple asset accumulators over capital-efficient software bets.

The Disagreement

During a mock investment committee on 20VC, Stebbings, Lemkin, and O'Driscoll evaluated three wildly priced AI startups. The options spanned infrastructure, developer tools, and vertical workflow automation.

O'Driscoll laid out the board: “You were basically asking Jason to pick between base 10 at 30 billion, up 2x from June, an inference company. Cognition at 70 to 80 billion, up from 48 billion in September, and Serville at 3 billion, up from 1 billion in December.” (play from 43:21)

Stebbings immediately took the side of Serval. In his view, software investing comes down to the math on exit multiples. ServiceNow sits at a $110 billion market cap in enterprise IT service management. If Serval takes real share from ServiceNow, a $3 billion entry point can actually return 10x to 15x your money. Stebbings noted that founders Jake and Tatiana built software with enterprise adoption that is “better than I've ever seen.” (play from 45:04) The only downside for mega-funds is check size: “The challenge though that we have with Servil that we don't have with cognition or base 10 is if we're an AUM play just putting cash out the door, they only raised 150 million.” (play from 45:31)

Lemkin pushed back directly on the core assumption behind vertical AI valuations. He questioned whether an AI product in ITSM actually makes the market bigger or simply fights for scraps of existing enterprise budgets.

“The just my question to the IC is does it expand the TAM?” (play from 46:02) Lemkin asked. “Am I bounded for real, right? Not for handwaving. Am I in essence bounded by a subset of ServiceNow's TAM today?” (play from 46:06)

Lemkin argued that investors are sleepwalking into a pricing trap. “I think in vertical we're going to see the biggest tears because a lot of folks are going to make investments assuming that especially for like an SMB or an enterprise you can charge three to four times the size, and we're going to find out you know what you can charge on one the size.” (play from 49:18)

Who's Right (and When They're Wrong)

Stebbings is right on valuation math. Buying into developer tools at $80 billion requires Cognition to become larger than almost every public software company on earth just to deliver a 3x return. At $3 billion, Serval does not need to reinvent capitalism to make venture returns work. It just needs to take 10% to 15% of ServiceNow's lunch.

Lemkin is right on software economics. If your pitch deck claims that an AI agent replaces five IT helpdesk workers and therefore captures all five salaries as software revenue, your model will break. Enterprise procurement teams do not hand labor budgets over to software vendors. They demand software pricing. If an enterprise paid ServiceNow $200,000 annually, they will not pay Serval $800,000 just because the software answers tickets automatically. In vertical software, customers expect deflation, not inflation.

If you sell an agentic product that executes end-to-end work, assume customers will anchor your contract on legacy software seat licenses, not human payroll replacement.

What to Do With This

Audit your pricing model this week. If your growth plan assumes you can charge customers three times the contract size of the legacy software vendor you replace, slash that assumption by 60%. Build a financial forecast that survives on matching incumbent contract sizes on a 1-to-1 basis while winning purely on win rates and churn reduction.