▶ 0:57 · $POOLSIDE · Poolside acquired by Nvidia for $6B+$1B at $12B pre after failing to raise $2B for compute

episode briefing

NVIDIA Buys Poolside, Invests in Mercor & Perplexity | OpenAI Confirms IPO

2026-08-27 · 18 company · 10 thematic

sentiment

10 bull2 bear6 neu

speakers

Founder of SaaStr, a community and media platform for B2B software founders and executives. He is an early-stage investor and a frequent commentator on SaaS growth and go-to-market strategy.

harry stebbings

Founder of 20VC (The Twenty Minute VC), one of the most popular venture-capital podcasts, and its associated fund.

rory o'driscoll

Rory O'Driscoll is a partner at Scale Venture Partners, a late-stage venture firm. He references Scale's portfolio (Linear, DataBricks, HubSpot, Salesforce, Palantir) and fund dynamics. He previously invested in Anthropic (2021) and tracks AI infrastructure closely.

quote

“they couldn't raise the 2 billion to buy 40,000 GPUs... they had no choice but to fail up for 6 billion plus topping off a billion to 12 billion”

—harry stebbings

now playing · $POOLSIDE

$POOLSIDEneutral· highjason lemkin

15x seed return on Poolside is a 'win' but not fund-returning at $9B exit

A 15x return on a seed investment that couldn't reach frontier model scale is still a 'rich man' outcome, but at $9B exit with dilution the effective entry was ~$600M pre — too hi…

$POOLSIDEneutral· medium· posharry stebbings

Poolside acquired by Nvidia for $6B+$1B at $12B pre after failing to raise $2B for compute

Poolside hit the capital intensity wall for frontier models but still delivered a 15x return to seed investors; illustrates that in hypergrowth AI markets, even 'failed' standalon…

$MERCORbullish· medium· posharry stebbings

Nvidia joins General Catalyst-led $20B round for Mercor at $2.5B ARR

Mercor has scaled to $2.5B ARR in data labeling/training services; Nvidia's participation may be strategic (access to training data flows) or simply deploying ecosystem budget, bu…

$MERCORbullish· mediumrory o'driscoll

Mercor's 80% gross margins warrant AI multiple not services multiple per Croll M&A data

Croll analysis shows M&A multiples don't penalize gross margins above 30%; Mercor's 80% margins and $2.5B ARR justify AI-style valuation despite being a data services business.

$NVDA···bullish· highrory o'driscoll

Nvidia spends entire free cash flow on ecosystem investments to expand TAM

Nvidia is deliberately deploying ~$75B of annual free cash flow into strategic investments (neoclouds, model companies, data providers) to expand the GPU TAM, treating ecosystem s…

$CURSORbullish· mediumrory o'driscoll

Cursor proved negative gross margins can normalize to $60B valuation

Cursor started with heavily subsidized pricing (negative gross margins) but developed own models and capped usage to fix economics; demonstrates that in AI applications, initial m…

$PERPLEXITYbullish· low· posjason lemkin

Nvidia investment in Perplexity props up Jason's portfolio

Perplexity benefits from Nvidia's ecosystem funding; Jason discloses small personal investment.

$OPENAIbearish· highrory o'driscoll

OpenAI must IPO in 2027 at lower valuation than Anthropic due to 'absence of choice'

OpenAI's Q1/Q2 growth slowed to ~18% QoQ (~100% annualized) vs Anthropic's $60B run rate growing faster; with competitors pulling ahead and chip vendors questioning future demand,…

$ANTHROPICbullish· highrory o'driscoll

Anthropic leads platform market at $60B run rate, forcing OpenAI to second place

Anthropic has achieved clear #1 platform position with $60B+ run rate growing faster than OpenAI; enterprises now default to Anthropic for coding, creating oligopoly dynamics wher…

$HUGGING-FACEneutral· mediumrory o'driscoll

Hugging Face at $13B is peak strategic asset for enterprise open-weight counterweight

Hugging Face's model hub is the key strategic asset for any IT vendor (Microsoft, IBM) wanting to counter closed frontier models; current $150M revenue is irrelevant — peak valuat…

$CITADELneutral· lowrory o'driscoll

Ken Griffin's Citadel unwound 80% of Leopold Aschenbrenner's leveraged fund in short-term trade

Citadel's 4:1 leveraged bet on Situational Awareness fund was a classic short-term market-making trade — bought at discount, sold into strength; not a fundamental view on AI but a…

$005930.KS···bullish· mediumjason lemkin

Korean memory stocks (Samsung, SK Hynix) up 56% YTD on AI demand, memory engineers now 'most eligible bachelors'

Kospi (heavily weighted to memory) up 56% YTD despite July drawdown; structural shift in Korea where memory engineering talent is now premium — reflects persistent HBM demand for…

$STRIPEbullish· highrory o'driscoll

Stripe accelerating to 41% growth at scale with 71% billing growth, buying back shares

Stripe's core payments business provides diversified cash flow while AI-driven agentic commerce drives accelerating growth (41% revenue, 71% billing); share count down 3-4 years v…

$DATABRICKSbullish· mediumrory o'driscoll

Databricks IPO at 80% growth will reset public software valuation benchmarks

Databricks going public at ~80% growth alongside Stripe and Anthropic will force a re-rating of all public software comps; only Palantir currently approaches these growth rates am…

$FRACTILEneutral· lowharry stebbings

Fractile raising at $6.5B after Etched at $20B signals AI chip startup frenzy

New AI accelerator startups (Fractile, Etched) commanding multi-billion valuations pre-revenue reflects unbounded founder creativity and capital availability; outcome uncertain bu…

$SUPERHUMANbullish· low· posjason lemkin

Superhuman acquired by Grammarly; Harry and Jason are proud shareholders

Superhuman's email client exit to Grammarly validates premium productivity tool thesis; hosts disclose personal investment.

$ANDURILneutral· mediumjason lemkin

Defense tech will consolidate around 2-3 giants like Anduril due to Pentagon account control requirements

Unlike commercial tech where single products can win, defense requires portfolio breadth to survive Pentagon procurement; Anduril and 2-3 others will achieve critical mass and acq…

$LOCUS-ROBOTICSbearish· medium· posrory o'driscoll

Specific-purpose robotics (15k units deployed) works; humanoid general-purpose is overhyped

Locus Robotics proves focused automation (15,000 warehouse robots) succeeds; humanoid robots chasing human-like dexterity are solving the wrong problem — a Tesla does 100m faster…

←