The latest from the 20VC x SaaStr collaboration with Harry Stebbings, Jason Lemkin, and Rory O’Driscoll
The TL;DR
Brex sold for $5.15 billion—a heroic outcome by any rational standard, yet the weird feeling in everyone’s stomach tells you everything about the era we’re in. Meanwhile, Anthropic’s inference costs came in 23% higher than expected, and that single data point might be the final nail in the coffin for mid-market B2B SaaS companies that thought they’d checked all the boxes.
This week’s conversation covered why “hubristic financing” creates inevitable disappointment even in great exits, how Ramp investors should be rethinking their 32x revenue valuation after seeing a comparable business trade at 7x, and the brutal truth about what happens when you’ve done everything right—gotten profitable, built an agent—but can’t afford the inference to compete.
Oh, and Andreessen just claimed 2/3 of all private AI revenue comes from their portfolio. Let’s unpack what that actually means.
Top Takeaways
1. Brex at $5.15 Billion: Great Exit, Weird Feelings, and the “Hubristic Financing” Tax
Capital One acquired Brex for a stunning $5.15 billion—50% cash, 50% stock. By any objective standard, building something from nothing to $5 billion before you turn 30 is heroic. And an incredible comeback from the challenging times for them around 2023 or so.
But the conversation kept circling back to something Ali Ghodsi from Databricks once said: “I never wanted to raise more than two years ahead of the valuation I was confident I could hit.”
Brex raised at $12 billion in 2021 when growth was 200-300%. The logic at the time? If we maintain this growth rate for two years, we’ll grow into this valuation on a revenue multiple basis. Perfectly rational.
But buried in that assumption is what Rory calls “a whole debt trap”: 2022 hits, growth fades, capital gets scarce, you converge on profitability, growth slows further—and suddenly that legitimate 2021 belief that three more years of 300% growth would justify $12 billion looks insane by 2024.
The result? A $5.15 billion outcome that’s genuinely great for founders, early employees, and most investors—but leaves a strange taste because of the promise made at $12 billion.
“The bad feelings last for a day and the five billion lasts forever,” Rory summarized. “You’ll get over it.”
2. What the Brex Exit Means for Ramp (Not Ideal For Valuations, At Least)
Three things happened for Ramp investors on the same day:
The good news: Brex essentially acknowledged Ramp won, even if they aren’t 100% direct competitors in every segment and with every product today. Started later, doing $1 billion to Brex’s $700 million.
The uncomfortable news: When real money—not a secondary buyer purchasing 2% of the cap table—wrote an actual check for the whole asset, they valued it at 7x revenue. If you multiply Ramp’s billion-dollar run rate by 7, you get $7 billion. Even doubling it for faster growth gets you to $15 billion. Ramp raised at $32 billion.
The structural news: Capital One now owns both Discover (closed for $50 billion) and Brex. Discover has a closed network—meaning Cap One captures all the interchange instead of splitting it with Visa and Mastercard. They can now route Brex spend onto their own rails.
If you’re Ramp, you’re now playing against the A-team with a structural cost advantage.
The mark-to-market conversation is going to be awkward: How do you factor in a comparable transaction at 7x revenue when your last round implied 30x?
3. Anthropic’s Inference Costs Are 23% Higher Than Expected—And That Changes Everything
The headline sounds like a footnote. It’s actually existential for a huge swath of B2B software.
Go back to what Amjad at Replit has been saying: “Everything’s going to get more expensive not less because we’re going to burn ever more tokens building ever better software. We will actually burn an infinite amount of tokens if we can.”
Yes, Anthropic went from -94% gross margins last year to positive 40% this year. That’s massive improvement. But the point isn’t whether Anthropic has a sustainable business model—they clearly will. The point is what this means for every B2B company trying to compete with AI-native startups.
Picture this scenario: You’re a $50-100 million ARR B2B company. You did everything your board asked. Got profitable. Built an agent. It works. Your customers like it.
Now your agent costs $2.50 per interaction. You need 50 million interactions this year. That’s $125 million in inference costs on $50 million in revenue.
Open Evidence has the money. Harvey has the money. Lovable has the money. You don’t.
“I did everything you told me to do,” Jason imagines founders saying. “You told me to get profitable. I did. You told me to build an agent. I did. And now the final nail in the coffin is we just can’t afford the inference. We can’t build a competitive product.”
The way out? Build an agent so good you can charge $10-20K per month because it replaces 20 people with ROI measured in weeks—not a sales pitch claiming it’s that good, but literally that good.
If you can’t do that, you’re racing against competitors who treat inference as their marketing budget because they have no sales team to pay.
4. Open Evidence at $12 Billion: The Hubristic Financing Cycle Continues
Open Evidence raised at $12 billion—a 12x step-up from their $1 billion round with Sequoia at the start of the year. Revenue grew 10x to ~$150 million.
The business is compelling: doctors need decision support for medical questions, the platform has commanding market share among medical professionals, and drug companies will pay handsomely to reach those doctors. The founder sold Kensho to S&P, has a PhD, was AI-native from day one.
There’s nothing not to like here.
But the team kept returning to the same question: Is this the $12 billion Brex round? The one where nothing but greatness happens but you still get caught on the tail end of hubristic fundraising?
Total pharma advertising spend is $22-30 billion—but half of that is TV ads to consumers. Actual doctor-to-doctor advertising is a $2-3 billion market. To get 3x on a $12 billion entry, they need $5 billion in revenue.
That means either they blow open new budget by shifting pharma rep spend to digital (credible) or they expand into other services to doctors. Both are possible. Neither is guaranteed.
The scary part? “In the moment, it never feels that hubristic,” Harry observed. “I had Pedro and Henrique on in 2021 talking about how they could build a hundred billion dollar business. Twelve billion didn’t seem crazy.”
Thrive is smart. They’ve done the math. Someone else will probably do the $30-40 billion round next year as revenue goes to $400-500 million. But someone’s eventually going to be holding the bag when the music stops.
5. Andreessen Claims 2/3 of Private AI Revenue—Here’s What That Actually Means
The stat sounds astounding: Andreessen-backed companies generate 2/3 of all private AI revenue.
But think about it differently: Add up AI revenue and you get roughly $13 billion from OpenAI, $4 billion from Anthropic, and everything else is in the noise relative to those two. Databricks is third. Andreessen has significant positions in all three.
To a rounding error, the slide really says: “We have money in OpenAI, and OpenAI is 40-50% of total revenue.”
The more interesting question: Is venture finally an asset class? If Andreessen can reliably own 18% of total funding in a given year—which mathematically means ~10% of Series A’s, ~10% of the great deals—and YC is doing the same at the early stage, maybe Gary Tan is right that venture should be 10x bigger.
Rory’s counter: There’s no investing business so good that excess capital won’t ruin it. In retrospect, everyone wishes they’d just gone home in 2021.
6. TSMC Is Telling You Everything You Need to Know About the AI Bubble
Everyone keeps asking when the AI bubble will pop. Here’s the tell to watch.
TSMC—the cynical, been-around-forever foundry that dismissed Sam Altman’s trillion-dollar data center dreams 12 months ago—just reported that “demand for compute is effectively infinite right now” and raised their capex budget.
These are the people who own the problem if they overinvest. If you dig a big hole in Phoenix and put a fab there that no one uses, you’re out $20 billion. They’re leaning in.
The capex is now $600 billion. AI app revenue is maybe $100 billion if you squint. That’s $500 billion in the hole annually. But people with money and conviction are saying the return is there.
At some point, Nvidia puts will be a great buy—every semiconductor cycle for the last 40 years has ended in a massive downswing. But you’re not buying them today.
7. The Challenged Wealthfront IPO vs. Equipment Share’s Perfect Exit
Two IPOs, two very different stories.
Equipment Share: Y Combinator 2015, technology-enabled construction equipment rental. IPO’d at $8 billion, popped 33%, growing 47% at $4 billion in revenue, profitable. “This is what an IPO is supposed to be,” Jason said. “Effortless.”
Wealthfront: Admired software, good deeds, made efficient investing easy. IPO’d at ~$2 billion, immediately traded down 36% to $1.3 billion. That’s barely public. You lose liquidity, lose analyst coverage. “A billion dollars… that’s not that’s even quite … there. You’re barely public.”
The cutoff appears to be ~$3 billion market cap. Below that, you’re in “who cares” land. And once you’re there, the talent question becomes brutal: Do the best engineers really want to join a company in a 30-40% down IPO when they have other options?
The bulls can do it if they have the right leader. But 90% of the team? They’re just blinking at the camera, asking to slip releases, saying Q1 and Q2 look soft but Q4 will be great.
8. SaaS Has an Army (Literally)
Salesforce just won a $5.6 billion Army contract over 10 years.
Some sales rep is getting the mother of all commissions. More importantly: anyone who thinks they’re going to vibe code their way to replacing $500 million Army orders needs to take a walk.
SaaS isn’t dead. The systems of record aren’t going anywhere. But the question remains: Can Salesforce get its mojo back, or is it a utility stock for the next 5 years?
The existential threats are stacking up: seat contractions are real (Workday says seats are “perpetually under pressure”), price increases have become destructive (SaaS prices up 40% in 3-4 years crowding out upsells), and workforces are shrinking.
Those 130% NRR years where Slack grew at 140% without ever raising prices? Those days probably aren’t coming back.
Quotable Moments
Harry Stebbings
“We’re going to forget about Brex in 24 months. We’re going to even forget whether it was two X’s or how to spell it.”
“If you were doing your mark-to-market on Ramp right now, how do you factor in a recent transaction at seven times to your multiple of 30 times?”
“For those that think about the AI bubble—does TSMC raising capex not completely denigrate those risks? When Dario comes out saying we’ll be replacing everyone’s job in under five years?”
Jason Lemkin
“The bad feelings last for a day and the five billion lasts forever. You’ll get over it.”
“If you walked into your board meeting and said ‘Good news guys, inference costs are going down 30% this year because our IT team’s really good at managing costs,’ I would throw my mouse at the monitor.”
“I worry this is the final nail in the coffin. You did everything right—got profitable, built an agent—and now you just can’t afford the inference to compete.”
Rory O’Driscoll
“Things prove up in the end for what they really are, not what you delusionally think they are at one point in time.”
“If you’re not giving value or you’re locked in a war with someone who has infinite capital and is willing to give it away longer than you, then you should figure out how to exit now.”
“At some point, Nvidia puts 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.”
This post is part of the ongoing 20VC x SaaStr collaboration. Watch the full episode for the complete discussion including the TikTok deal analysis and why Andreessen claims 2/3 of private AI revenue.
