20VC: The Pro-Rata Battle: Valar Atomics Sues Day One Ventures | Meta Returns to the AI Frontier With Muse Wonder | NVIDIA Eyes $30B Reflection Deal & Jev Raises $870M at a $7.5B Valuation

NVIDIA

Across the shows: 28 speakers in other episodes, 22 bullish and 9 bearish calls, past 30 days

• Rory O'Driscoll · Scale Venture PartnersBullish

  Nvidia buying open-model labs aligns its chip interests against closed model companies, so the deal makes sense.

  Rory's logic is that Nvidia is a chip seller whose biggest customers are closed-model labs. Money flowing into the AI pool is better for Nvidia if more of it stays with chips and less goes to model companies. Owning open-weight models therefore gives Nvidia leverage to push back if closed labs get margin-greedy, so the Reflection and Poolside purchases fit its interests.

  “zooming out. You've got a pile of money that's going to go into that collective pool, it's in your interest that more of it stay with you and the chips and less of it stay with the model companies, right?”at 21:02

• Jason Lemkin · SaaStrBullish

  Nvidia's software, hardware and model combination could make it the default open-weight choice within 18 months.

  Jason argues Nvidia is a software company as well as a hardware company, with its engineers and Nemotron models behind Salesforce's models. He says the moat across software, hardware and model is so strong that in 12 to 18 months developers may have no reason to pick an open-weight model other than Nvidia's. He admits this doesn't seem logical today.

  “This is a software company, not just a hardware company. You could imagine that the moat here between software, hardware, model is so powerful that in twelve or eighteen months, why would you want to use any other open weight, open model other than NVIDIA?”at 23:03

TypeSafe

• Jason Lemkin · SaaStrBullish

  TypeSafe's JEV can hold its lead against Cloudflare, OpenAI and Amazon if a focused team keeps building.

  Jason is cautiously bullish because JEV is fast, cheap and accurate enough for around 10% of AI use cases, and a focused team of ex-OpenAI people can keep improving it. He doubts Cloudflare or Amazon will invest as hard in a niche product. He uses JEV himself and says it would be hard to switch away from if it stays ahead on evals. Seven-fifty of funding gives two or three years of runway, and inference costs are not what would sink it. He is unsure whether it will win in a year.

  “I'm cautiously bullish that a focused team rather than four smart kids at Cloudflare can maintain its lead and get to accuracy levels to summarize that were never possible before.”at 29:25

• Rory O'Driscoll · Scale Venture PartnersBullish

  TypeSafe's JEV could win about 20% of frontier inference; at that share it is worth around $50B.

  Rory models frontier AI inference as a $200B market in a year or two. If JEV takes 20% of it, that is $40B of spend. Even if prices fall 80%, that is still $8B, and if JEV gets half of it, the company has about $4B in revenue and is probably worth $50B. He says each step is credible. His main worry is competition, not the business case.

  “If they make it hyper accurate, it changes the game... if it's 20% of your total... 20% of that is 40,000,000,000... if they get half of it, that's a [$4 billion] revenue company, which is probably worth $50,000,000,000.”at 29:42

Isomorphic Labs

• Jason Lemkin · SaaStrBullish

  AI drug discovery potential is massive, and Isomorphic's talent justifies its steep valuation despite the risk.

  Jason changed his mind on AI-driven drug discovery after seeing the medical evidence for GLP-1s. He argues AI could radically speed up drug development, letting an owner of the IP have others do the trials. He says Isomorphic's founder can assemble talent no one else can, so generational talent commands generational prices, and he calls the $40-50B valuation a scary bet with liquidation preference as cover.

  “I think the potential for drug discovery is probably massive.”at 38:37

• Rory O'Driscoll · Scale Venture PartnersBearish

  Isomorphic will find it harder to capture drug value through royalties than investors expect.

  Rory says Isomorphic's value capture differs from foundation models, which replace labour. Here the value comes from discovering drugs, and on a $200B drug a 10% royalty would be the prize. He doubts drug companies will hand over meaningful royalties, since they run phase one through phase three and many discovery partners end up as drug companies themselves. He is glad they are trying but thinks monetisation will be harder than expected.

  “I do believe it'll be harder to capture value than you think, but I'm really glad they're trying.”at 40:23

Flock Group

• Jason Lemkin · SaaStrBearish

  Flock Safety cut too many corners, leaving unsecured cameras and unsupported software exposed.

  Jason cites specific security failures: cameras streaming unsecured across the internet, and dated Android tablets without security patches for over a year to save money. He says that is unacceptable for a security company, where security must be paramount. He worries that cutting corners on hardware and software means other vulnerabilities exist up and down the stack, with stored images unencrypted.

  “They've had cameras just streaming unsecured across the Internet... This is unacceptable from a company where security has to be at its core and paramount.”at 1:07:31

• Jason Lemkin · SaaStrBearish

  Flock Safety cut corners on security, and its hardware and software flaws are serious and pervasive.

  Lemkin argues Flock cut too many corners and did not build safe hardware or software, so its vulnerabilities run up and down the stack. He cites specific failures: stored images were unencrypted, cameras sent credentials in clear text, pressing a button on the back of a camera opened a Wi-Fi network any attacker could enter, the police interface required no authentication, and security patches stopped in June 2018 for many cameras. Harry Stebbings agrees these facts are not good, which supports Lemkin's point that the company is now reaping the consequences.

  “Flock cut too many. It just didn't build safe hardware or software. And if you're not investing in the hardware and the software, what about the rest of the team? ... I wonder if they cut too many corners and now they're reaping what it comes from it.”at 1:08:16

Meta Platforms

Across the shows: 23 speakers in other episodes, 24 bullish and 6 bearish calls, past 30 days

• Jason Lemkin · SaaStrBullish

  Meta's Muse Wonder could become a monster AI product, though its headline evals look performative.

  Jason sees a material chance Meta builds a monster AI product because its investment, capabilities and autonomy are very high, and says it could be bigger than the AR/VR goggles bet. He is skeptical of the launch: the model is internal-only with no API, the eval is performative, and it is not cheaper than rivals, so he is not yet impressed by the model itself. He leans on Muse integration to make the product matter.

  “I think there's a material chance Meta has a just a monster AI product on its hands. The amount of investment into it is so high. The capabilities are so high. The autonomy is so high.”at 19:43

Reflection AI

Across the shows: 1 speaker in other episodes, 1 bullish and 0 bearish calls, past 30 days

• Rory O'Driscoll · Scale Venture PartnersBearish

  Reflection's standalone open-weight business is hard to justify, so he would take Nvidia's bid.

  Rory would sell to Nvidia at about $30B. Reflection is an amazing team, but as a standalone business an open-weight model depends heavily on inference monetization, and it is unclear how that works commercially. Excess prices make the acquirer's value very high, and he thinks this moment may not come back. The sale is worth taking because the standalone case is unproven.

  “Oh, I'd hit the bid. Because they're amazing companies, but the standalone business, ... you're relying a lot on the monetization is inference, are you in that? I've always thought open weight is really exciting. But how does it work monetarily?”at 21:57

Thinking Machines Lab

Across the shows: 1 speaker in other episodes, 1 bullish and 0 bearish calls, past 30 days

• Jason Lemkin · SaaStrBearish

  Thinking Machines likely can't get enough compute to stay competitive in the frontier arms race.

  Jason doubts Thinking Machines can be competitive at its roughly $44B round because compute is scarce and everything is being taken. He cites Poolside selling because it could not afford or get enough compute, and says even Anthropic is reaching out to Meta for compute. He adds that memory is hard to get too, so the arms race is harder than software. He hedges that someone closer to the labs should judge.

  “My gut is no, my gut is you can't get enough compute to be competitive. Everything's being taken, right?”at 24:25

Anthropic

Across the shows: 34 speakers in other episodes, 14 bullish and 36 bearish calls, past 30 days

• Jason Lemkin · SaaStrBearish

  Anthropic's inference margins look high, but real gross margins stay weak because of training costs.

  Jason expects cheap Haiku usage to be counted as inference COGS, with training costs assigned to frontier models like Opus. He believes Anthropic's inference margins are high but thinks real-world gross margins remain poor once training is included. He says he will be curious how the company breaks this out in its S-1.

  “I believe that Anthropics inference margins are quite high... I believe their real world gross margins are still pretty crappy because of training costs.”at 49:34

Cloudflare

Across the shows: 1 speaker in other episodes, 1 bullish and 0 bearish calls, past 30 days

• Jason Lemkin · SaaStrBullish

  Cloudflare's 20% layoff was the right call; growth reaccelerated and the stock is up sharply.

  Jason argues Cloudflare cut 20% of staff in a single day not because it was struggling but to speed up. It is an AI beneficiary that was wasting time on work that would not get it there. The quarter beat, growth accelerated further, and the stock is up about 80% this year. He contrasts this with bending spoons' deals, where the cuts are a result of acquisition.

  “They've accelerated growth even more. The stock price is up like 80% this year. And it's not bending spoons. He didn't decide to gut Cloudflare, but he made the right call.”at 56:19

Bending Spoons S.p.A.

Across the shows: 1 speaker in other episodes, 1 bullish and 0 bearish calls, past 30 days

• Harry Stebbings · 20VCUncertain

  With 3% organic growth, Bending Spoons must buy large companies to keep up its 80% growth.

  Harry says Bending Spoons has to find large enough revenue-based companies to buy, not small ones, since smaller companies do not move the needle. He cites its roughly 3% organic growth, so reaching 80% growth requires big acquisitions. He adds that churn at the acquired companies forces it to raise prices to the maximum just to keep them running.

  “It's 3% organic growth. So if you wanna grow 80% like they are, you gotta buy big stuff.”at 57:04

Bumble

Across the shows: 1 speaker in other episodes, 0 bullish and 1 bearish calls, past 30 days

• Jason Lemkin · SaaStrBullish

  AI could make Bumble's matches far better, letting it charge more and multiply its value.

  Jason gives the bull case for Bumble as a builder who hasn't used the app. Dating matches are visual and based on little data, so AI using deep analytics and historical data could produce much better matches. Users would pay more for five great dates than twenty bad ones. He also notes others are pursuing the same idea and says it might work. Bumble's take-private at about $500M market cap before the deal is the backdrop.

  “Anchor high and quintuple or 10x the value of this business. That would be the bull case to me.”at 59:42

Intercom

• Jason Lemkin · SaaStrNeutral

  Intercom's AI-native reboot reached nine-figure revenue, far short of billion-dollar scale growing 50%.

  Jason says Intercom rebuilt itself as an AI-native company and reached nine figures in revenue, which is a good achievement and deserves credit. He says it is far from a billion-dollar business growing 50%, so it does not overturn the view that SaaS has lost its scale advantage. He says he is a minor shareholder.

  “Kudos to the Intercom for building an AI native reboot of their company to 9 figures in revenue, but that's very different than building to a billion growing 50%.”at 1:03:42

Venture capital & startups

• Jason Lemkin · SaaStr

  Big funds routinely waive seed investors' pro rata rights in term sheets, a practice that is common.

  Jason says large growth funds like Sequoia often demand that everyone waive pro rata so they can reach their ownership target. Their lawyers insert waivers into term sheets without checking them. He says this happens all the time, and that it is unfair in principle, because the rules are not supposed to favor some investors over others.

  “And I can say as a seed investor, this... happens all the time. The big funds, in fact, they put in their term sheet without even checking it. Their lawyers immediately put in that my pro rata is waived in every single round.”at 6:29

• Rory O'Driscoll · Scale Venture Partners

  A liquidation preference in a hot AI company beats many 2021 unicorns, which will end up worthless.

  Rory argues a liquidation preference in a company like JEV is worth more today than the preferences of a thousand 2021 unicorns. Many of those will end up worthless. He accepts Jason's point that the preference has real value now, even if the downside protection is overstated.

  “I will say today it feels like a liquidation preference in, in JEV is much more valuable than a thousand... twenty twenty one unicorns. Many of those will end up being worthless.”at 30:38

• Rory O'Driscoll · Scale Venture Partners

  Decacorn creation is at a record pace; private valuations rest on option value and liquidation preferences.

  Rory cites 42 new decacorns in 2026 against 34 in 2021, and roughly 20 to 30 per decade over five decades, with only 98 public decacorns ever. He says the production rate is higher than anything seen before. He argues private prices make sense because a 1x preference is the downside and option value is high. Three companies may reach $1T this decade, so the non-zero chance of a huge outcome justifies rich valuations. He does not say outright that this is a bubble.

  “What you can say for certain is the production rate of decacorns is higher compared to anything we've seen before, including five decades of data in the public markets.”at 33:22

• Jason Lemkin · SaaStr

  Venture fund return math needs AI-driven exits, and historical evidence makes that bet suspect.

  Jason uses Menlo's new fund as the example. It must return several billion through exits within a year, which assumes AI spending and recent exits will keep producing winners. He says the recent evidence supports that, but historical VC returns make it suspect.

  “That's the bet. The AI spend and the recent exits of the last X months justify it. Historical evidence makes it suspect.”at 34:45

• Rory O'Driscoll · Scale Venture Partners

  Venture fund math needs several new $25B exits; doable, but the margin for error is thin.

  Rory says he checked the data and that funds need to extrapolate what growth rate is being assumed to make the math work. Only about 130 funds ever exceeded $10B, and only about 70 above $25B. The industry therefore needs about four new $25B-plus outcomes to make the math work. It is doable, but the funds cannot afford to miss.

  “And so you got to get four new ones. It's doable, but you can't miss a lot here.”at 35:47

Robotics & autonomous vehicles

• Rory O'Driscoll · Scale Venture Partners

  Robotics data spend will scale only after humanoid deployments happen; so far deployments are lagging.

  Rory argues robotics data companies can only reach a billion or two once deployments happen, and that timing is still unproven. He cites a logistics company's post summarising 42 humanoid proofs of concept with zero deployed, and a BMW-Figure relationship that talked big without humanoids on the floor. He thinks the lack of deployment will filter back to spending on robotics data until the deployment unlock happens.

  “I saw a GXO, which is a logistics company... the summary of the post was 42 POCs of humanoids, zero deployed.”at 42:23

• Jason Lemkin · SaaStr

  Demand for high-quality robotics training data is off the charts, even with skeptical robotics bets.

  Jason says interest in robotics is at an all-time high and every robotics company needs training data. He cites a YC founder who did $30M in 60 days on a niche data-labelling product and turned away about $270M of demand because the work needs many specialised humans across countries. He would rather back the data infrastructure than the robots, though he admits he is not sure which robotics companies will win.

  “the demand for high quality training data, even if it's just paying people to wear sensors and do work, it's just off the charts.”at 43:08

AI pricing & unit economics

• Jason Lemkin · SaaStr

  Haiku's 90% price cut makes it the first deflationary AI product, with speed gains and cheaper usage.

  Jason argues that LLMs have not been deflationary for his portfolio companies, because cheaper tokens were offset by much higher consumption. Haiku 5.5 is nearly as good as Sonnet for many uses at a tenth of the price and can reason on demand, so teams stop rationing tokens. Sonnet 5.5 is about 10% cheaper and twice as fast, and speed is also deflationary. He expects Jevons-style usage growth, possibly 10x, so B2B workflows get more AI for less money.

  “They're really, from my perspective, the first thing that's actually been deflationary with LMs and AI. Nothing else has been deflationary...”at 46:10

• Rory O'Driscoll · Scale Venture Partners

  Anthropic's price cuts bet on highly elastic AI demand; spend rises only if usage jumps enough.

  Rory frames the cuts as a price-elasticity bet. Cutting prices on an inelastic product leads to high fixed costs, low demand expansion and lower prices, which is bad for the seller. Because AI providers are cutting prices, they must believe demand is very elastic, so total spend goes up as usage rises. He says the next thing to watch is how the price cuts play out.

  “So the bet here, as you see these price cuts, is in fact that this stuff is very elastic in terms of demand.”at 49:06

Software & SaaS

• Jason Lemkin · SaaStr

  The SaaS selloff recovered only for companies above $1B revenue; smaller SaaS still suffers.

  Jason argues the SaaSpocalypse is not over. Public software indexes rebounded about 22-28% this year, but only companies with massive scale above roughly $1B in revenue recovered. Companies without scale face a worse SaaSpocalypse, with no one wanting to buy software. He says a $500M IPO bar is gone, and companies now need $1-2B of ARR to go public safely.

  “I think it only passed for those with massive scale. And I think everyone without massive scale is living a worse saspocalypse”at 1:01:43

• Rory O'Driscoll · Scale Venture Partners

  Mature SaaS consolidation means buyers take only a few mid-sized companies each year, leaving most stranded.

  Rory says the industry is maturing and consolidating, so market leaders with scale and high market caps acquire the rest. With perhaps 200-300 similar mid-sized SaaS companies, big tech can only buy about 10 a year because acquisitions distract. Most companies without a path to scale will be left as one of hundreds of mid-sized firms compounding slowly. He says they need to go AI-native and drive toward an outcome, expansion is what could save them, and an exit could be a dividend.

  “Big tech can only buy so many at a time because it's a distraction. So they get hoovered up at the rate of 10 a year.”at 1:05:05

Open-source AI models

• Jason Lemkin · SaaStr

  Enterprises won't adopt China-based open-weight models, and they'll be niche within a year.

  Jason says no enterprise wants China-based open-weight models, and cheap Haiku pricing makes open source even less attractive. He expects enterprises to prefer domestic solutions from vetted vendors, so these models could be niche within twelve months. He cites Salesforce ripping out DeepSeek-based tooling before a deal closed as evidence, and he admits the trend could still reverse.

  “No one wants to use at least China based open weights models. No one in the enterprise wants to use them.”at 50:47