20VC with Harry StebbingsValar Atomics Sues Day One Ventures | Meta's Back to the AI Frontier | NVIDIA's $30B Reflection Deal
20VC with Harry Stebbings
Valar and Day One dispute 1:32
Valar Atomics is in a fight with seed investor Day One Ventures over pro rata rights in a Series B led by Sequoia at a $6 billion valuation. Day One had put about $1 million into seed and $15 million into the A, and the CEO asked major investors to give up their allocation. The discussion centers on whether Day One was wronged, whether the company’s spring-back clause matters, and whether later-stage rounds are now using terms that let big funds take the whole round while others stand aside.
Why rights get waived 2:00
The speakers explain pro rata in plain terms as the right to keep your ownership by taking part in later rounds. They say it is common for a majority of investors to waive those rights for everyone, especially when a big fund wants the full allocation. They also note that spring-back rights can let a skipped investor regain some access if another investor is given a special carve-out, which is why Day One may feel it was treated unfairly.
Letting it go 10:00
The debate turns practical. One side says both the company and the investor handled things badly, and that a VC who feels wronged should raise the issue once and then move on. A personal example comes up about being cut to zero in a later round after years of support, including bridge financing, and how even when the decision feels obviously wrong, pushing too hard can damage the founder relationship without changing the outcome.
PRA pricing debate 14:00
The talk starts with carry rights and whether you should waive them. The point is that the right has value, so if a founder gives it up, the price should be lower. Keith Rabois is cited as the clearest voice on that trade, and no founder in the room is said to be willing to pay half price without the right.
Meta and open models 15:30
Meta’s new model, Muse Wonder, is treated as useful only inside Meta because there is no public API to use. Even so, the group gives Meta credit for getting back near the front of AI work and for having a strong partner in Alex Wang. The model is seen as likely to improve Meta’s own product, and that could matter more than the benchmark score.
Picking the winners 28:00
The discussion says venture only works if you keep investing in breakout companies, even when the price looks high. Jev is described as one of those winners, and the point is that a small team with focus can stay ahead of bigger teams at Cloudflare, Amazon, or OpenRouter. The bet is simple: if the product keeps getting better, you will not switch away from it.
Valuation and acquisitions 31:30
The speakers argue that Jev could become a very large company if it captures a slice of a huge AI inference market. They also note that liquidation preferences matter, but less than the chance of being in a true super winner. The same logic is used to justify today’s high private valuations, while acknowledging that much of the upside depends on continued growth and on big buyers staying active.
Decacorn math 33:31
The conversation turns to how many startups are now valued above $10 billion. It points out that 42 decacorns appeared in 2026 so far, more than the 34 seen in 2021, and far above the long-run pace of about 20 to 30 per decade. That makes the current pace look extreme, even if rising market caps explain some of it.
Science and drug discovery 39:00
The focus then shifts to Isomorphic Labs and similar science-driven AI companies. The claim is that the real value is not just replacing labor, but finding new drugs and taking a royalty on the result, such as a share of a large drug’s revenue. The speakers say this is easier to believe now because GLP1 drugs have proved how large the medical market can be, and because faster discovery could mean better treatments and longer life.
Robotics data demand 44:00
Training data for robotics is growing fast. Companies like Mecca have raised large rounds, and there are several others already above $100 million in annual recurring revenue. The point is that robotics firms and foundation model companies both need data now, and they are willing to pay for it. That has pushed niche providers to very large revenue in a short time.
Model cuts change behavior 50:01
Anthropic’s lower prices for Haiku and Sonnet are described as the first real deflation in AI. Haiku is now cheap enough that you can use it for many jobs without worrying much about cost, and Sonnet is faster while only a little cheaper than before. That matters for business software, where speed and cost both shape whether you can run a workflow at all. The expectation is that lower prices will lead you to use much more of these models, not less.
Layoffs and duty 57:00
The talk opens on whether founders and boards owe anything to the people who work for them when a sale means mass layoffs. The view here is blunt. In a public company, duty runs to shareholders first, and the job in an acquisition period is to maximize value. But there is also a personal refusal to sell if the buyer will quickly cut 80% of the team.
Bending Spoons and scale 59:00
Bending Spoons is treated as the clearest example of a buyer that will slash staff after a deal. That certainty is contrasted with other private buyers that may be less obvious about what they will do. The same lean logic is compared with Cloudflare, which cut 20% in one day and then accelerated again, showing that a hard cut can be a way to move faster rather than a sign of failure.
Bumble and AI dating 1:02:00
Bumble comes up as a private deal at about $1.3 billion, and the business is framed as valuable but hard because of churn and weak lifetime value. The stronger bull case is that AI could improve matching by using far more data than the current app does. Better matches could mean fewer bad dates, higher prices, and a more useful service. The exchange ends by noting that AI avatars might reduce the time you spend meeting people, even if real-world contact still matters.
Software after the panic 1:06:00
The final stretch turns to the so-called SaaS apocalypse. The earlier fear that people would vibe-code their way out of software did not hit the biggest companies hard, but it is still harder for smaller firms. Scale now matters most. A company with hundreds of millions in revenue can still get bought or public, but a company trying to grow from 100 million to 300 million has a much rougher path.
Scale wins the exit 1:08:32
The market is described as a mature, consolidating one where big leaders keep their edge and buy others. The claim is that 500 million in revenue may no longer be enough for a safe IPO, and that software companies may need 1 to 2 billion in ARR to go public with confidence. Recent sales like TechWolf to SAP and Intercom’s AI reset are held up as real wins, but not the kind that change the larger rule: only scale is safe.
Security and market pressures 1:12:02
The discussion turns from software companies to Flock Safety. Its privacy backlash, recent layoffs, and repeated security failures are treated as a serious business problem, not just a communications one. The speakers also note Jensen Huang’s plan to put two billion more into European AI companies, calling it small against Nvidia’s cash flow but useful for spreading risk and cooling anti-American feeling.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
