We’re back! Our latest deep dive with Jason Lemkin (SaaStr), Rory O’Driscoll (Scale Venture Partners), and Harry Stebbings (20VC).
And come join the team LIVE at SaaStr AI London, Dec 1-2!! More here.
Bottom Line Up Front
The classic venture playbook is almost … no more.
Traditional SaaS metrics like burn multiples—once the gold standard for evaluating capital efficiency—are being rendered obsolete by AI-native companies growing at unprecedented speeds with radically different unit economics.
Meanwhile, founders with objectively good numbers (triple-triple-double-double growth, solid burn multiples) are getting rejected by VCs focused exclusively on AI breakouts.
The message is stark: if you’re not AI-first, raise capital now at any reasonable price, consolidate where possible, and prepare for a world where even “perfectly good” $15M ARR companies have “zero value to VCs.”
The stakes extend beyond individual companies:
With 6,700+ unicorns and only 15 IPOs year-to-date, massive portfolio consolidation is inevitable.
Tech PE firms face existential questions as AI agents reduce seat-based revenue and products that stayed static for a decade now require constant reinvention.
And towering above it all: OpenAI’s plan to consume more energy than India within 8 years, raising fundamental questions about whether the economics of AI can ever pencil out at scale.
The Burn Multiple Paradox: Why AI Breaks the Rules
The conversation opened with Iconiq’s 73-page State of Software report, which revealed a counterintuitive finding: AI-native companies under $100M ARR have terrible free cash flow margins (-126% versus -56% for non-AI companies), yet their burn multiples are actually better because they’re growing so explosively fast.
Rory O’Driscoll explained the fundamental concept: “It’s basically how many dollars of ARR do you get out of each dollar that you’re spending, right? What is the efficiency you’re creating for each dollar of venture capital you’re lighting on fire?”
The math seems simple: if you’re valued at 10x ARR and you spend $2 to add $1 of ARR, you’ve created $10 of market cap from $2 invested. But as Rory cautioned, this only works when multiple hidden assumptions hold true:
The ARR is actually real (not inflated by aggressive accounting)
Net retention accounts for real churn (fast growth can hide massive customer losses)
Gross margins are sustainable (hyper-growth can mask deteriorating unit economics)
No massive capex requirements (true for most SaaS, decidedly not true for AI infrastructure companies)
Jason Lemkin emphasized the existential risk everyone’s ignoring: “David Sacks coined the Burn Rate Multiple and I think when everything was the same in SaaS and B2B in 2021, it made a lot of sense. All the companies were the same… Then AI breaks it because we’ve never seen growth like this. But the margins are lower.”
The team agreed that while burn multiples remain useful for comparing companies, they’re no longer sufficient as the primary valuation framework. As Rory put it: “We are not in Kansas. There’s a whole bunch of implied assumptions in there which is why even though we love those kind of metrics… we’ve actually come back to saying there’s a real advantage in seeing the GAAP revenue accounting also to make sure all the money is for lack of a better word just showing going up for real, right?”
The Brutal Reality: Good Companies Can’t Get Funded
Harry Stebbings raised what may be the most troubling question for founders: “I have many companies with good burn multiples and they are not getting love. They are not getting attention and they are going, ‘Harry, I don’t get it. I’ve been brought up to understand burn multiples, to understand growth, like what is going on?’”
Rory’s response was unsparing: “There’s only two ways of pricing a deal. You price a deal on hope or you price a deal on the multiples. When you price a deal on hope and growth, you can lean in on anything, right? And you can get prices that quote unquote make no sense because the growth ultimately comes and it all pays off. Once you start valuing things on quote unquote the fundamentals today, right? Then you can value a public company because at 400 million, it’s not nothing. But… a $15 million revenue company that’s perfectly good and has reasonable growth is actually of zero value to a VC because we’re in the upside option game, right?”
This crystallizes the existential crisis for non-AI-native companies: even with objectively strong metrics, if the path to a massive IPO isn’t crystal clear, VCs simply won’t engage. The bar isn’t just high—for certain categories, it functionally doesn’t exist.
Jason was even more direct about the advice gap: “I hear too many folks, they’re like, ‘Oh, you’re triple triple double double or better. You’re you’re golden. Don’t worry, kids.’ And I think that’s terrible advice in 2025. Terrible advice.”
He described portfolio companies at $15M ARR growing 100% with good burn ratios that can’t get funded unless they’re ultra-breakout AI stories. His advice? “If Scale wants to put in money at 250 on that deal… take that deal now.” Even if it means accepting a lower valuation than 2021-era thinking would suggest.
Rory concurred: “Those companies, Jason, you’re right, is that… you should just get the deal done. Raise at a reasonable price. Continue to grow, but be capital efficient. Don’t get lost in your just your burn multiple. Focus on your cash. You know, if you’re right about your business, you’ll be right in the end.”
The Kingmaker Effect: Why Proximity to Harvey or Abridge Matters
Harry introduced a phenomenon that’s reshaping competitive dynamics: the power of King-Made Startups: “The mimetic—and this sounds obvious given the sheep plate analogies applied to venture—but it’s how concerned investors are by going against a kingmaker whether it’s Harvey or Abridge or any of the king-made companies. We have a couple of companies which are the second or the third and going against the kingmaker in the valley is the most unpopular thing in the world.”
The strategy is working: “Founders listening, when you’re raising, raising to deter others from raising is a really working strategy right now.”
Rory acknowledged the effect but with important caveats: “Do I believe that thinking exists in venture? Yes, I do. I don’t fully share it but I acknowledge that you know you have to factor it into your decision-making and your risk analysis… If your first customers are also VC-backed companies, then you get this dual loop. But look, the reality is… if you’re selling to oil and gas companies, they barely can tell the Sequoias from their KPs.”
The takeaway: the kingmaker effect is real and powerful in Silicon Valley-centric markets, but less dispositive in traditional enterprise segments where buyers don’t track VC pedigree.
Jason noted what makes AI different from previous cycles: “What is a little different in AI is that there are in many cases there’s not an established brand and there’s so much change and so much new budget and so much confusion that so many buyers are under pressure and have a desire to make a purchase… Being number one is so powerful when people know they want to do something—they’ve got to do an LLM for legal.”
He cited the death match between Lovable and Replit as an example: “They’re both at nine figures in revenue. They won’t kill each other, right? But… you want to be that brand that nervous folks don’t know who to buy.”
Valuation Insanity or Justified Vision? The $1 Trillion Question
Harry posed the fundamental question: “Do you think we are near peak madness guys or do you think we’ll look back and laugh at ourselves for having this conversation given the might and the size of the markets that we’re entering?”
Rory’s answer was unequivocal: “One of two things has to be true because that’s actually the interesting insight… If that massive transfer of labor doesn’t happen, then all these valuations are wrong by an order of magnitude. One of two things is going to happen in the next five or seven years, right? Either… you are going to see pretty profound productivity changes. You’re going to see companies like OpenAI hitting two to 300 billion in revenue super quickly. That’s option A. Option B… AI is still going to be wonderful, but you’re going to have a readjustment period that’s going to make your head hurt.”
Jason worried about a different risk—loss ratios: “What I wonder is, are our loss ratios correct? If as venture VC firms, especially with larger funds… as long as we get it, as long as we’re cool, if 80% of our unicorns implode or blow up or make no money, more importantly, just don’t make money for venture… But I’m pretty sure we didn’t model it right in 2020 2021.”
He continued: “It feels almost risk-free again like it did in 2021. It feels risk-free. Valuation and loss ratios both matter.”
Rory distinguished between two types of fund-killing mistakes: “You can imagine people failing because their picking ratio was bad and they just put too much of the money in bad companies, right? Or you can also imagine people failing where all their companies were great, were good companies. But the prices were so bad that they only made a subpar return even though they only had a 30% loss ratio, right? So you both—I mean both things can—I mean the sad thing about venture is you actually have to get both things right to make money.”
On whether this is sustainable, Rory offered a sobering forecast: “I personally think that the rate of adoption forecast over the next four or five that’s implicit in all these data center assumptions will in retrospect prove to be too optimistic.”
Sam Altman’s Energy Crisis: Willing a Trillion Into Existence
The discussion turned to OpenAI’s staggering infrastructure requirements: planning to 125x energy capacity in 8 years, requiring more energy supply than India’s current total capacity.
Jason framed the scale in visceral terms: “Just the little hundred billion that he’s doing with Nvidia… needs more power than all of New York City. So it will only be a couple years where the cities of the future don’t even have humans in them… Our country will be dotted with these Stargates that are larger than New York City with only hundreds of people working in them and the equivalent of millions of digital minds.”
Harry expressed fundamental skepticism about the financing: “I feel very stupid honestly because I look at a trillion dollars required to fund data centers for Open AI alone. I’m like I don’t know where that money comes from… Sovereigns don’t have that. I mean, well, actually, funny enough, they do. I mean, they’d have to put it all in.”
Jason offered a nuanced view: “I think Sam is just willing as much of this into existence as possible because of the future. And if we come up short, if it’s $400 billion or $600… we don’t have to buy all the GPUs. It would be okay if they had to last 6 years instead of 3 years or whatever the depreciation schedule is.”
Rory distinguished between directional vision and operational reality: “The sheer act of willing something into existence like this has been just amazing… Whatever the prize is for being the best company in AI, OpenAI is going to get that prize, right? And that’s his job. And he’s doing it better than any other CEO of this decade. That’s true. It is also equally true… just because a CEO says we’re going to treble next year doesn’t mean we should buy real estate and hire people as if we’re going to treble.”
His concern: “Five years from now… we said, ‘Oh, I get it. OpenAI is still the best company on the planet for AI.’ Its growth rate has slowed to a shockingly small 50 60%. It’s freaking amazing. It’s doing 30 billion dollars growing at 50%. It’s astonishing. But maybe they don’t need a trillion dollars of capex this week.”
Zuck’s AI Crisis: When Even Loyalists Lose Faith
Harry made a rare public admission about his largest public holding: “I always thought that Zuck earned the right to do the next thing. He earned the right to do the next thing. And I have to say my faith in Meta’s AI strategy has just dwindled and dwindled and dwindled… I’m looking at Alex Wang. I’m looking at the treatment of Yann LeCun. I’m looking at how they structure teams and I’m going this is not wellrun. OpenAI, Anthropic are coming for you.”
Jason was characteristically blunt: “He is not the communicator that Sam Altman is… If I don’t understand where the hell he’s going and Harry doesn’t understand… if we don’t understand he’s not one of the great communicators at the moment… He’s a crappy communicator right because none of us—I’m not saying his AI strategy isn’t S tier but I don’t think any of us understand that not for the life of us where the hell it’s going.”
Rory offered both defense and critique: “You could also say it’s just a harder bet… The bet for the last 15 years was riding this brilliant invention that you had in 2003 and just optimizing it. And that’s hard but it’s a lot easier than now you got to invent a whole new thing a second time.”
He continued with the existential framing: “The big picture comment is if you spend two hours a day on ChatGPT that’s two hours a day that you are not spending on Facebook and we live and die on our attention. So we’re just going to make shit until somehow we get people to come back and play with us.”
The panel agreed that while Zuck was “clear just this week he’d rather burn the 20 billion in operating income and fail than become irrelevant,” the specific strategy remains opaque even to sophisticated investors.
Rory’s prediction: “If there was a Couch bet… that you know some version of this AI strategy will not produce meaningful revenue despite a $20 billion burn and will look more like Meta VR and less like Instagram and WhatsApp… I’d take that bet.”
The Great Consolidation: Fivetran + dbt Labs and What It Means
The panel discussed unicorn Fivetran’s talks to acquire unicorn dbt Labs as a template for the massive consolidation wave that must happen with 6,700 unicorns and only 15 IPOs year-to-date.
Rory framed the math: “We’ve processed 15 out the IPO gate year to date. So 20 for the year. That implies we got 30 years of this stuff to get through. What they’ve done… every time two companies combine… it takes two midsize companies… and makes a bigger… This is part of the job venture is going to have to do to whip their portfolios into shape to be IPOable.”
Jason identified the key enabler: “The fact that Andreessen is the lead or close to it in both deals makes it much simpler on many levels… If Andreessen owns 20% of Fivetran and 20% of dbt and you combine them, it does kind of suck when you own 20% of a portfolio company, combine it with another leader… and now I own 8% after the deal.”
But he emphasized why it still makes sense: “20% of something that’s not going public is not nearly as interesting as 8% of something that is going public, right? And if you believe that it’s not a continuum of value… but rather it’s like electron states, there’s just a gap and then you got to go to the next state.”
Rory stressed the risk: “The fatal mistake that always scares me is not the delusion… The thing that scares me is you go from a decent deal that’s wellrun where you know everything about it to merging with something else and then the combined entity screws it up. That to me is the really shitty outcome where you took your 20% bet and turned it into 8% of a disaster.”
The PE Reckoning: Why Tech Private Equity’s Model Is Under Threat
Harry asked whether tech PE firms are questioning their business model: “When they see the multiples that you can get on the money that’s being moved by your Kushner and your big firms, combined by the increased loss ratio that will happen from an increasingly volatile new AI world. Are you suddenly going, ‘Shit, I’m not getting paid for the risk that I’m taking on the multiple on the upside given the displacement on the downside.’”
Jason identified the core problem: “These products didn’t change from about 2008 until 2023. They’re the same products… It wasn’t just that we had high NRR which was the spreadsheet glue for the PE model. It was the fact… that the products can’t be static and AI is the accelerant there… The rate of change is unprecedented in business software. Unprecedented.”
He gave the example of his first venture investment, Pipedrive: “It took him four years to launch a mobile app… Could you imagine today waiting four years to launch your AI co-pilot? Like you’re dead in the water.”
Rory expanded on the implications: “There was 15 years where we made the same product and you didn’t have to think that much about product direction at the macro level… You look at Salesforce 2002 and 2022, it’s the same thing, right? And that’s now changed… On the PE side, you’re exactly right. You look at that and you go, I might get away with next 10 years making the same thing… And so maybe I’m entering into some kind of tech risk that I’ve never before internalized. A new tech risk.”
Jason added the seat model crisis: “Now that we’re running 12 AI agents, you know, we only need two seats of Salesforce. Because you don’t have the people. We just don’t… It makes the PE model worse and our jobs harder. It just literally… makes it even tougher for PE to buy these seat models.”
Key Takeaways
Burn multiples are useful but insufficient: AI companies break traditional SaaS assumptions around gross margins, churn visibility, and capex requirements. VCs must look at GAAP revenue and absolute cash positions, not just efficiency ratios.
Non-AI-native companies face a binary funding environment: Even with triple-triple-double-double growth and good burn multiples, companies perceived as “legacy” struggle to raise. Founders should take reasonable deals now rather than optimizing for price.
The kingmaker effect is real but not dispositive: In Valley-centric markets, being the #2 or #3 behind a heavily-funded leader significantly impacts fundraising. But in traditional enterprise segments, customer buying decisions care less about VC pedigree.
Massive consolidation is inevitable: With 6,700 unicorns and 15 IPOs year-to-date, the math demands hundreds of mergers. Firms like Andreessen that own both sides of deals have structural advantages in making these combinations happen.
Tech PE’s classic SaaS funding model faces existential threats: The shift from static products (unchanged for 10+ years) to rapidly evolving AI-powered solutions introduces unprecedented tech risk. Meanwhile, AI agents reducing seat-based revenue attacks the core cash flow assumptions.
OpenAI’s ambitions may be both directionally correct and operationally overstated: Sam Altman is “willing a trillion into existence,” but even if the vision is right, the actual capital deployed and timeline may be far more modest than the rhetoric suggests.
Meta’s AI strategy lacks clarity even to sophisticated investors: Despite a $20B annual commitment, investors including long-term bulls struggle to understand the strategic endgame beyond “avoid irrelevance.”
Loss ratios matter as much as entry valuations: Even if VCs pick the right companies, paying prices that require 100x returns to generate acceptable fund returns means most funds will fail even with good picking.
Quotable Moments
On burn multiples in the AI era:
“A $15 million revenue company that’s perfectly good and has reasonable growth is actually of zero value to a VC cuz we’re in the upside option game.” — Rory O’Driscoll
On outdated advice:
“I hear too many folks, they’re like, ‘Oh, you’re triple triple double double or better. You’re golden. Don’t worry, kids.’ And I think that’s terrible advice in 2025. Terrible advice.” — Jason Lemkin
On the kingmaker phenomenon:
“Going against the kingmaker in the valley is the most unpopular thing in the world. Founders listening, when you’re raising, raising to deter others from raising is a really working strategy right now.” — Harry Stebbings
On OpenAI’s prize:
“Whatever the prize is for being the best company in AI, OpenAI is going to get that prize.” — Rory O’Driscoll
On Sam Altman’s energy vision:
“Our country will be dotted with these Stargates that are larger than New York City with only hundreds of people working in them and the equivalent of billions of digital minds.” — Jason Lemkin
On financing the impossible:
“I feel very stupid honestly because I look at a trillion dollars required to fund data centers for Open AI alone. I’m like I don’t know where that money comes from.” — Harry Stebbings
On Meta’s communication problem:
“He’s a crappy communicator right because none of us—I’m not saying his AI strategy isn’t S tier but I don’t think any of us understand that not for the life of us where the hell it’s going.” — Jason Lemkin
On tech risk in PE:
“Maybe I’m entering into some kind of tech risk that I’ve never before internalized. A new tech risk.” — Rory O’Driscoll
On consolidation math:
“20% of something that’s not going public is not nearly as interesting as 8% of something that is going public.” — Jason Lemkin
