The Twenty Minute VCAnthropic Inference Costs Skyrocket |TikTok Deal Closes |The IPO Market:Wealthfront & EquipmentShare
CHAPTERS
- 0:00 – 2:25
Brex sells to Capital One for $5.15B: why it’s still a “heroic” outcome
The hosts unpack Brex’s surprise $5.15B sale to Capital One and push back on social-media narratives that frame it as a disappointment. They argue that building a multi-billion-dollar business from scratch is a major win, while noting the emotional whiplash created by prior peak-era expectations.
- •Deal terms and initial market surprise
- •Why a $5B+ exit is an exceptional founder outcome
- •The gap between outcome reality vs. online “potshot” discourse
- •Capital One’s strategic rationale teased as a later thread
- 2:25 – 9:07
“Hubristic financing”: the hidden tax of raising at peak valuations
Jason and Rory explore why Brex’s 2021 $12B valuation makes the exit feel strange even if outcomes are strong for most stakeholders. They describe how late-stage rounds implicitly promise extreme growth trajectories, creating disappointment when growth normalizes.
- •Why exits below the last round price create a ‘weird feeling’
- •Raising at market price when capital is needed is rational
- •Ali Ghodsi’s ‘raise only what you can grow into’ principle
- •Extrapolation traps: 300% growth assumptions break quickly
- 9:07 – 15:08
What Brex’s sale implies for Ramp (and other spend/finance startups)
The conversation shifts to competitive consequences: operationally, Ramp looks validated—but the deal sets a stark valuation multiple reference point. They also discuss how Capital One’s structural advantages could reshape the competitive landscape for independents.
- •Validation vs. valuation: ‘winning’ doesn’t guarantee public-market multiples
- •Transaction multiple math (e.g., ~7x revenue) vs. private marks
- •Capital One + Discover rails: closed-network interchange advantage
- •Why independents may need to stay private longer
- 15:08 – 16:48
TikTok’s forced divestment: cheap economics, heavy geopolitics
They assess the reported TikTok deal structure where US investors own most of the entity but Chinese owners retain algorithm control. The group emphasizes that this is primarily a geopolitical intervention that produces unusually attractive economics for the selected buyers.
- •Why this deal can’t be analyzed purely economically
- •Purchase price vs. US revenue suggests a strikingly low multiple
- •OPEX/licensing payments back to parent could change real economics
- •‘Directed purchaser’ dynamics and modern oligopolistic capitalism
- 16:48 – 18:39
Why didn’t top-tier VC firms show up in the TikTok deal?
Jason questions the absence (or reduced presence) of firms like Andreessen/Sequoia if the deal is ‘free money,’ comparing it to the famous Skype spinout play. The hosts speculate there may be deal constraints, political complexity, or hidden risks keeping traditional venture out.
- •Skype as precedent: forced divestment created structural opportunity
- •Where are the usual ‘minting money’ firms—why the hesitation?
- •Potential reasons: constraints, optics, governance, hidden risk
- •Investor composition: Oracle, sovereign wealth, and strategic players
- 18:39 – 21:44
Anthropic inference costs surprise: margins improving, but tokens keep exploding
They debate the news that Anthropic’s inference costs came in higher than expected and what that means for AI unit economics. Rory notes gross margins have improved dramatically year-over-year, but Jason argues demand will expand faster than cost declines, keeping pressure high.
- •Inference costs as the defining constraint for AI product companies
- •Gross margin trajectory: from deeply negative to strongly positive
- •Economies of scale exist, but may asymptote below ‘SaaS-like’ margins
- •Token demand expands with capability—cost declines don’t guarantee relief
- 21:44 – 25:39
24/7 inference and ‘memory’: why better products can mean higher costs
Jason connects rising inference to always-on AI assistants and persistent memory, describing a weekend build that ‘remembers everything’ but burns tokens aggressively. The takeaway: the most compelling product experiences often drive usage intensity that overwhelms efficiency gains.
- •Vision: inference running continuously for knowledge work
- •Persistent memory requires massive token throughput
- •Deflation in per-token cost vs. inflation in total consumption
- •Founders must model inference costs rising, not falling
- 25:39 – 32:12
The mid-market SaaS trap: profitable, shipped an agent… now can’t fund inference
Jason raises a boardroom-level crisis for mature B2B SaaS companies: they finally reach profitability and ship an agent, but lack capital to compete with heavily funded AI-native entrants. Rory agrees the escape route is either pricing power from unique value or exiting before a losing war.
- •Break-even SaaS companies face existential AI cost escalation
- •New entrants can treat inference as ‘marketing’ while incumbents can’t
- •Only durable fix: deliver ROI so large customers pay enough for compute
- •Alternative: use proprietary data, open source models, and efficiency—but value must exceed cost
- 32:12 – 39:05
Compute demand signals: TSMC’s capex as the most credible ‘tell’
Rory argues the strongest indicator of real demand isn’t hype from model companies but capex decisions by supply-chain players like TSMC. He notes TSMC increasing capex suggests compute demand is ‘effectively infinite’ near-term, though a semiconductor cycle downturn will eventually come.
- •Discount ‘talking the book’ from AI labs and infrastructure sellers
- •TSMC’s multi-year capex commitments are harder to fake
- •AI pyramid framing: foundries are the base layer signal
- •Bubble timing is unknowable; scenario planning and funding buffers matter
- 39:05 – 43:20
OpenEvidence at $12B: perfect use case, but TAM expansion must follow
They review OpenEvidence’s massive valuation step-up and why the product is compelling: medical decision support, curated sources, compliance, and a clear pharma ad model. Rory cautions that doctor-targeted ad budgets may be smaller than headline pharma spend, implying the company must pull spend from reps or expand product scope.
- •Why the product wins: trusted sources, HIPAA, restricted access, workflow fit
- •Business model: pharma advertising to physicians
- •Market sizing nuance: consumer TV ads vs. direct-to-doctor budget
- •To justify upside, must shift offline spend or expand services/TAM
- 43:20 – 46:42
The next Brex risk: who gets stuck in OpenEvidence-style ‘hubristic’ rounds?
They return to the concept of peak pricing—if OpenEvidence keeps stepping up valuations rapidly, one round may end up being a ‘1x’ despite a great business. The hosts note that in the moment, these deals always feel inevitable because quality is obvious and momentum is powerful.
- •Great companies can still produce mediocre returns at the wrong entry price
- •Valuation climbing in lockstep with revenue can still end badly when tides shift
- •Late-stage partner-meeting psychology: ‘generational company at any price’
- •The round that looks reasonable now may be the one that later feels excessive
- 46:42 – 57:20
a16z’s ‘2/3 of private AI revenue’ claim and whether venture becomes an asset class
Harry brings up a16z’s report and the striking revenue concentration claim; Rory contextualizes it as largely a function of a few massive winners. Jason broadens it into a question: if dominance is repeatable, does venture become a scalable asset class—while Rory warns excess capital can still ruin returns.
- •Revenue concentration: OpenAI/Anthropic/Databricks dominate the numerator
- •Institution-building vs. small partnership models in venture
- •Debate: scale venture AUM vs. returns degradation from too much capital
- •Succession and durability: can iconic, founder-led firms maintain dominance?
- 57:20 – 1:05:58
IPO window reopens? EquipmentShare pops, but Wealthfront shows the subscale trap
They contrast an ‘effortless’ EquipmentShare IPO—large, fast-growing, profitable—with Wealthfront’s weak post-IPO performance and small market cap. The takeaway is that public markets reward scale and profitability, while smaller IPOs risk poor liquidity, limited coverage, and muted talent magnetism.
- •EquipmentShare: tech-enabled physical business with scale and profitability
- •What a ‘clean’ IPO looks like: oversubscribed, trades up, low drama
- •Wealthfront: down post-IPO; challenges of being a ~$1–2B public company
- •Market cap thresholds matter for liquidity, analyst attention, and recruiting
- 1:05:58 – 1:18:04
Down IPOs and clearing the 2021 backlog; plus Salesforce’s $5B Army contract and ‘SaaS isn’t dead’
The group debates whether more companies should accept down IPOs to clear valuation overhang from 2021, arguing that price ultimately clears markets. They close on Salesforce’s massive Army deal as evidence systems-of-record endure, while acknowledging SaaS faces multiple simultaneous pressures (seats, pricing fatigue, AI budget shifts).
- •Down IPOs as pragmatic logjam-clearing for 2021-vintage unicorns
- •‘Price clears all markets’: public buyers will show up at the right price
- •Salesforce’s Army win: systems-of-record aren’t getting replaced by vibe coding
- •SaaS headwinds: seat contraction, price fatigue, AI budget reallocation, and competitive intensity
