Jan 29, 2026 · 1h 18m · news
Anthropic Inference Costs Skyrocket |TikTok Deal Closes |The IPO Market:Wealthfront & EquipmentShare · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20VC, host Harry Stebbings, Rory O'Driscoll, and Jason Lemkin dissect the structural shifts in tech and venture capital, examining high-profile M&As, the crippling infrastructure costs of the AI boom, and the contrasting paths of recent IPOs.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 9.4% of the talking time here. How this is scored →
Harry as informed peer 3.8 Guest teaching 5.5 Guest disagreement 3.0 Harry pushing back 2.3
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:06:58 Rory Rejects 'Fed to the Dogs' Framing
Rory strongly pushes back when Harry suggests subscale IPOs are feeding bad assets to retail investors, stating he does not live by Chamath's rules and calling Harry's phrasing a pejorative statement.
Hardest push from Harry ▶ 1:05:59 Harry Challenges Ethos Subscale ValuationHarry aggressively challenges the logic of Ethos going public at a $1.3B valuation down from $2.7B, repeatedly pressing whether subscale IPOs harm public markets.
Biggest teaching moment ▶ 40:20 Rory Corrects Harry's Pharma TAM MathRory systematically dismantles Harry's $22B TAM argument for Open Evidence by explaining that over half of pharma marketing goes to consumer TV ads and direct doctor ad spend is only $2-3B.
Harry holds his own ▶ 39:50 Harry Builds Comprehensive Pharma Ad TAM CaseHarry demonstrates deep domain preparation by calculating a $4-5B potential revenue run rate for Open Evidence based on specific US pharmaceutical ad spend data.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Capital One Acquires Brex: A "Heroic" $5B Outcome | 2 | 4 | 2 | 1 | Harry introduces the Brex acquisition by Capital One with key deal terms before handing off to Rory and Jason. The guests analyze hubristic financings and late-stage dilution dynamics without host intervention. | |
| The Competitive Aftermath for Ramp | 4 | 5 | 2 | 2 | Harry asks a sharp question about how Brex's exit multiple impacts Ramp's $32B valuation mark. Rory explains how public comps and M&A pricing force investors to re-examine private software marks. | |
| Capital One's Shrewd Strategy & Discover Integration | 0 | 5 | 1 | 0 | Harry remains completely silent throughout this segment while Jason and Rory discuss Capital One's acquisition of Discover. Rory explains the strategic value of closed-loop payment rails. | |
| Geopolitics and Valuations: The TikTok Divestment Deal | 4 | 4 | 2 | 1 | Harry outlines the TikTok divestment terms including ownership structure and algorithm control. Guests compare the transaction structure to Andreessen's early acquisition of Skype from eBay. | |
| Anthropic's Margins and the AI Scale Question | 4 | 5 | 3 | 2 | Harry questions whether Anthropic's cost overruns disprove economies of scale in AI. Rory gently corrects the premise by pointing out gross margins improved dramatically from -94% to 40%. | |
| Building 'Ren' and the Reality of Token Burn | 3 | 4 | 2 | 1 | Harry prompts Jason on his previous prediction regarding 24/7 inference burn. Jason shares practical insights from building an AI agent, detailing how full chat memory expands token burn. | |
| B2B SaaS and the "Final Nail" of Inference Costs | 3 | 6 | 3 | 2 | Harry asks how squeezed SaaS companies can survive rising inference costs. Rory and Jason explain that companies must deliver high-ROI agents to justify premium pricing models that cover compute costs. | |
| TSMC's Capex and the Non-Bubble AI Demand | 4 | 5 | 3 | 3 | Harry asks if TSMC's increased capex eliminates AI bubble concerns, citing Dario Amodei's timeline predictions. Rory clarifies that capex plans show short-term demand visibility rather than long-term immunity from over-investment. | |
| NVIDIA's Position and Scenario Planning for Founders | 3 | 6 | 4 | 2 | Harry suggests Nvidia might be underpriced given endless inference demand. Rory dismisses the host's bullish framing by pointing out that every semiconductor cycle over forty years ends in a severe downswing. | |
| Open Evidence's $12B Valuation & the Pharma TAM Debate | 6 | 8 | 4 | 3 | Harry offers a detailed TAM calculation based on $22B in annual pharma ad spend to justify Open Evidence's $12B valuation. Rory directly reframes the math, demonstrating that consumer TV ads make up half that spend and direct doctor advertising is only $2-3B. | |
| Thrive's Bet and Generational Company Valuations | 5 | 5 | 3 | 3 | Harry asks if growth investors are taking on Brex-like risk buying Open Evidence at $12B. Jason explains partner meeting dynamics where growth investors convince themselves to pay top price for generational companies. | |
| a16z's AI Revenue Claim & Venture as a Repeatable Asset Class | 6 | 7 | 4 | 2 | Harry cites a16z report statistics claiming two-thirds of private AI revenue comes from their portfolio. Rory punctures the marketing narrative by showing the metric is heavily skewed by OpenAI and Databricks. | |
| Institutionalizing Venture vs. Excess Capital Risks | 0 | 6 | 2 | 0 | Harry is silent throughout this segment as Rory and Jason debate institutional venture capital. Rory quotes Stan Druckenmiller to emphasize that excess activity and capital can erode investment returns. | |
| Generational Transitions: a16z vs. Khosla Ventures | 4 | 5 | 3 | 3 | Harry asks which firm faces a harder generational transition between Andreessen Horowitz and Khosla Ventures. The guests dodge the direct firm comparison with humor while Rory contrasts benchmark models with institutional platforms. | |
| Equipment Share's Effortless IPO | 5 | 4 | 1 | 1 | Harry introduces Equipment Share's IPO metrics, noting its 47% growth rate at $4B revenue. Jason describes the listing as an example of a frictionless, highly oversubscribed public transition. | |
| Wealthfront and the Perils of Subscale IPOs | 5 | 5 | 3 | 4 | Harry questions whether a company with a 30-40% down IPO like Wealthfront can still attract top-tier talent. Jason explains the difficulty of maintaining organizational momentum in subscale public companies. | |
| Talent Acquisition and the Human Bell Curve | 3 | 7 | 5 | 2 | Rory rejects the premise that non-outlier companies are filled with mediocre talent, explaining that financial outcomes follow power laws while human talent follows a bell curve. Harry expresses confusion at the statistical distinction. | |
| The Ethos IPO and Clearing the 2021 Unicorn Logjam | 6 | 7 | 7 | 7 | Harry aggressively pushes back on Ethos going public at half its peak valuation, asking if subscale IPOs represent feeding bad assets to retail investors. Rory forcefully rejects Chamath's framework and defends clearing the backlog of 2021 unicorns. | |
| Salesforce’s $5.6B Army Contract and the Future of SaaS | 5 | 6 | 4 | 4 | Harry highlights Salesforce's $5.6B Army contract to argue SaaS incumbents will regain dominance using distribution. Jason counterbalances this by detailing structural headwinds like seat contractions and price fatigue. |
Statements from this episode (58)
O'Driscoll: $5B liquidity outlasts the short-term disappointment of a down-round exit
“The bad feelings last for a day, and the five billion lasts forever. Right? So you'll get over it.”
O'Driscoll: Nvidia puts will eventually be a great buy due to cycles
“At some point, NVIDIA ports will be a great buy, because every semiconductor cycle for the last 40 years has ended up in a massive downswing. I ain't buying them today.”
Lemkin: Current market pressures are the final nail for traditional SaaS
“When we talk about is SaaS dead, or what's going on, I worry this is the next final act. I think it's the final nail.”
O'Driscoll: SaaS is not dead and is now strengthened by AI
“SAS is not dead, and now SAS has an army.”
Capital One acquires Brex for $5.15B in cash and stock
“Brex's acquisition by Capital One, 5.15 1,000,000,050 percent cash, 50% shares.”
O'Driscoll: Building Brex to a $5.15B exit before 30 is heroic
“Going back to the first thing, I think it was a great outcome, and I think, you know, you build something from nothing to a five billion dollar outcome before you're 30. Heroic result. Absolutely be praised.”
Lemkin: The tech industry will forget about Brex within 24 months
“We're gonna forget about Brex in 24 months. We're gonna even forget whether it was two X's or how to spell it.”
O'Driscoll: Fintech startups inevitably revert to financial services revenue multiples
“Things prove up in the end for what they really are, not what you delusionally think they are at one point in time. In, in the end, a financial services company was always going to trade at a financial services multiple adjusted for growth, and that's what hap…”
O'Driscoll: Ramp hit $1B run rate faster than Brex at $700M
“From an operational perspective, this is further validation that quote unquote, I've won. I started later and I'm doing a billion. They started earlier, they're doing seven hundred million.”
O'Driscoll: VC valuations are illiquid price signals set by single buyers
“We live in this crazy land of VC valuations where they're made once a year, when only one person buys, no one can subtly sell. They're very thin markets.”
O'Driscoll: At a $32B valuation, Ramp must succeed via IPO
“You ain't going to get the M&A outcome anymore. You've just got to be the big company and trade in the public markets at a significantly higher multiple than the other financial services companies.”
Lemkin: 80% of B2B public company CEOs are unhappy today
“And all my public company CEOs on they're pretty grouchy today. I would say 80% of the public companies B to B CEOs. Now they may be thrilled when they, when the next generation IPOs, but this isn't the happiest cast of characters.”
O'Driscoll: Brex is worth more to Capital One via Discover rails
“Capital One bought Discover Card. Discover Card has a closed network where they get all the money in the interchange. Right? So that's a really powerful asset for them now with, you know, now that they have Brex, they will probably be directing as much of that…”
O'Driscoll: TikTok US generates $15B-$16B revenue and was bought for ~1x
“I think the company's doing 15 or sixteen billion dollars in the US revenue, And they bought it for like one times revenue plus or minus.”
Lemkin: Andreessen Horowitz tripled its money on Skype in 12-14 months
“And the reason it was a structurally weird deal where they got a good deal. Now they took risks. Skype was a aging platform, but they bought it from eBay, right? Is that eBay needed to divest it? They had enough of it. There was no synergy and Andreessen went …”
Lemkin: Absence of major VCs in TikTok deal signals hidden risks
“There's a reason they're not in that deal. It's free money. Otherwise, right. We're missing something.”
Lemkin: B2B startups cannot cut AI inference costs and stay competitive
“The idea that you can use. Cheap models and cut back on your inference and still be competitive? That's the thing. Still be competitive with the hot Andreessen-funded company? Like, no chance you can be competitive without that inference.”
O'Driscoll: Anthropic gross margins improved from -94% to ~40%
“Remember, last year they had a -94% gross margin. And this year they have a 40% plus or minus gross margin.”
O'Driscoll: Anthropic will eventually become a cash-flow profitable business
“Fundamentally, however, I don't doubt the fact that a profitable, and defined by that as free cash flow, operating income business, will emerge from something like, I mean, Atropic's not going to not have a profitable business model because this clearly is con…”
Lemkin: AI startups must model inference costs rising in 2026
“You need to model in your inference costs are going up this year, not down.”
O'Driscoll: Inference costs consume up to 70% of AI coding app revenue
“If you're a high-priced app, maybe it's 10 or 15%. If you're a coding type app, maybe it's 50, 60, 70% of your revenue, and if you don't manage that correctly, you don't have a business.”
O'Driscoll: AI startups fighting hyper-funded competitors must differentiate or exit
“If your customers are getting value from your AI agent that they can't get anywhere else, and you can make that value clear. Then you can charge enough to pay for your tokens, and yay you. If you're not getting, if you're not giving value, or if you're locked …”
Lemkin: High inference costs will ruin mature mid-tier SaaS companies
“I worry this is the next final act is that you checked out, you did all the right things, right? You're not growing zero percent. Your customers don't hate you. You built an agent and the final nail in the coffin is we just can't afford the inference. We just …”
Lemkin: AI inference costs are replacing traditional sales and marketing budgets
“The other advantage that the new entrants have is that if you have the best agent and you have the kind of market demand we see, right, then you're for you, your inference costs are a marketing cost. The established players don't have that luxury. They're alre…”
Lemkin: Salesforce employees call current AI era most stressful in company history
“And I don't see how even Salesforce is one of the few that can do it with its resources. And even there, it's stressful. If you talk to folks at Salesforce today, they'll tell you this is the most stressful time they've ever worked at Salesforce in history of …”
O'Driscoll: TSMC raising CapEx budget to $50B amid infinite AI demand
“They just did their earnings call, and the comment was, you know, basically, demand for compute is effectively infinite right now, and they're raising their capex. Remember, these are not kind of, these are folks who say, I'm going to spend fifty billion dolla…”
O'Driscoll: AI CapEx spend won't pull back in the next 12-24 months
“So I think it's highly unlikely that you know, this is not going to be the year when people get terrified and say, I'm not going to do it. At some point, I think they will, because I think we probably are over-investing at some level, but right now, people are…”
O'Driscoll: Annual AI CapEx exceeds app revenue by $500 billion
“The capex is now six hundred billion, the app's revenue, you know, squinting is A hundred billion. So you're still, you know, five hundred billion a year in the hole”
Lemkin: No upside for founders in betting AI slows in 24 months
“There's no upside in, in, in, in betting this is gonna slow in the next 24 months. There's literally, at least for 99.9% of us, there's no upside.”
Stebbings: Open Evidence raised at $12B valuation led by Thrive and DST
“Open Evidence raises at twelve billion dollars, labbed by Thriving DST. It's a 12 x valuation step up to where they raised at a billion dollars from Sikora at the start of the year.”
O'Driscoll: Open Evidence has reached $150M in revenue
“And they've escalated to, I believe, a hundred and fifty million in revenue, right?”
O'Driscoll: Direct-to-doctor pharma advertising is a shrinking $2B–$3B market
“Actual direct to doctor advertising is to a three billion dollar marketplace, which is now getting a little bit smaller, right?”
O'Driscoll: Open Evidence must expand TAM significantly to 3x its $12B valuation
“So to get to get three X from twelve billion, you probably have to do some significant time expansion.”
Lemkin: Open Evidence will raise at $30B-$40B valuation next year
“Open Evidence will probably do a round at 30 or 40 next year. I'm actually going to suggest that Thrive is very smart. And they've probably done the math and this is the right insertion point for them. And they believe in it. And someone else is going to do it…”
O'Driscoll: Open Evidence 10xed revenue and valuation over the past year
“They tenxed revenue this year, and they tenxed their valuation plus or minus, and so the revenue multiple is the same, right?”
O'Driscoll: OpenAI and Anthropic generate virtually all private AI revenue
“So, objectively speaking, if you kind of add up all the AI revenue, you know, you're going to get thirteen billion for OpenAI, four billion for Entropic, and everything else is in the noise, right?”
Lemkin: Venture capital is not an asset class because the bottom 75% underperforms
“The classic take in ventures, it's not really an asset class. It's a weird niche of PE. Yes, the top Your quartile, certainly the top decile perform, but the rest is a disaster. So it's not an asset class. If the bottom 75% isn't even worth getting out of bed …”
O'Driscoll: Venture capital has split into two distinct asset classes
“I think venture is actually two asset classes. It's the traditional early stage venture that's existed for 2030 years, and this new late and later stage venture asset class that used to be called small cap growth, and is now it is now privately held.”
O'Driscoll: Excess capital will erode venture capital returns
“I don't think in either case they benefit from excess capital because I do believe that, you know, Martin Biggs, I've said it before, there's no investing business so good that excess capital won't ruin it, right? And I do think that excess capital will make t…”
O'Driscoll: Nearly all 2020–2021 venture investments will yield 1x or less
“Because if you think everything in 20, 21 was either priced wrong and makes a one X or early and just totally wrong and makes, you know, less than one X, let's just say, right? Other than a few companies that were the early precursors of AI”
O'Driscoll: Small VC firms like Benchmark face higher generational transition risk
“It's harder to be someone like Benchmark, small and brilliant, and manage generational transition, which is why it's awesome that they do it, because the asset is the brains of four or five individual people.”
O'Driscoll: Andreessen Horowitz will survive generational handoffs by institutionalizing
“The beauty of what Andreessen are clearly trying to do, and why I think they'll be able to manage it, is they're basically trying to transcend the individual by just being an institution. Their fundamental bet has been that venture capital is going to go the s…”
Stebbings: EquipmentShare popped 33% at IPO to an $8B market cap
“Equipment share IPO, Popped 33%. Eight billion dollar market cap. Growing 47% at four billion in revenue.”
Lemkin: Wealthfront's IPO is deeply broken and trading down 30-40%
“It's a deeply broken IPO. You know, it's trading down what? 30 or 40% from its IPO and it's subscale. Like it's worse. It's subscale, right?”
O'Driscoll: Wealthfront will compound out long term, but lacks short-term liquidity
“I think they will compound out. I actually like the company and have a mental note here to go check on it and see the valuation and maybe buy some, but you, Jason is right. It, it's not going to be a liquidity event in the short term because there's, it's not …”
O'Driscoll: $3B market cap is the effective cutoff for tech IPO liquidity
“The objective fact is, Three billion plus or minus appears to be the point at which you're, you know, it's easy to go public and it gets a lot easier the more you go up from there, right? Maybe three is a cutoff, right? And when you do something at two and the…”
Lemkin: Failed founders are currently the hottest tech recruiting category
“They may be failed founders themselves, which is like the hottest recruiting category in tech right now, right?”
Lemkin: Companies of any scale only need two or three key leaders
“You only need two or three leaders, a company of any scale. The best ones will find two or three.”
Lemkin: 80% of employees at top companies do not add value
“The reality is even at the best, 80% of folks are not contributing significant value mathematically, but.”
O'Driscoll: Venture cost of capital should be 30% versus 11% public
“The venture cost of capital should be around 30%, and the public cost of capital should be around 11.”
O'Driscoll: 2021 unicorn winners will be those accepting down-rounds
“We will look back and go, the winners were the ones who got a crazy valuation in 20, 21, and then were able to get out from under that valuation via a down round, a down IPO, or a down M&A. The bad ones are the guys who are still sitting there looking at their…”
Lemkin: High-profile tech unicorns are aggressively seeking M&A exits
“And we did an MNA review a little while ago, and I was shocked who's on the block. I mean, everyone's for sale. And I was shocked that folks worth less, I mean, worth more, much more are willing to be acquired by someone with a fraction of their revenue.”
Stebbings: US Army awarded Salesforce a $5.6B contract over 10 years
“I do think a hat tip deserves to be given to our friend of the show, Mr. Mark Benioff, who army just awarded sales. 5.6 billion. Billion dollar contract over 10 years.”
O'Driscoll: AI vibe coding will not replace enterprise systems of record
“I think deals like this put to rest the, anyone who thinks that they're going to vibe code their way to a product that can replace a five hundred million dollar army orders.”
Lemkin: Shopify grew over 40% while keeping headcount flat for 3 years
“Shopify has held headcount flat for three years and grown 40 something percent in that time.”
Lemkin: SaaS product prices increased 40% over the last 3-4 years
“SAS price products are up 40% the last three to four years.”
Lemkin: Clever pricing models cannot force customers to spend more
“Tweaking pricing models doesn't change how much folks want to spend for a product, right? That's a fallacy. That's what consultants do. They're pricing consultants. It's great, but if no one wants to pay more than 20 grand a year for your product, you can't fo…”