Overview of 20VC with Harry Stebbings
This episode is a fast-moving tour through the biggest AI, venture, and public-market stories of the week. Harry Stebbings is joined by Rory O’Driscoll, Jason Lemkin, and guest Jack Altman to debate Anthropic’s leaked S-1, Instinct’s massive $1B raise at a $10B valuation, AMD’s reported $8.2B purchase of World Labs, Meta’s talent grab from MongoDB, and what all of this says about venture pricing, talent flow, compute scarcity, and the future of IPOs.
Anthropic’s Draft S-1 Leak: Big Numbers, Little New Information
The group starts with the leak of Anthropic’s draft S-1 and agrees the document was more interesting as a cultural moment than a source of new data.
Main takeaways
- The leak didn’t reveal much that wasn’t already broadly understood about the company’s economics.
- The most notable detail: two customers reportedly accounted for 25% of revenue, implying very large enterprise spend.
- The panel argues the real question is Q3 2025 performance, not stale historical numbers.
- They note that public-market investors may be more patient and structural in how they evaluate frontier AI businesses than private-market observers.
Concerns raised
- Public disclosure forces companies to spell out existential risk, security concerns, and legal liabilities, which could fuel political backlash.
- Harry and Jason both worry that even if an IPO is successful, negative narratives could weigh on trading after the deal.
Instinct’s $1B Series C at a $10B Valuation
Instinct’s huge raise becomes the centerpiece of a broader discussion about consumer AI agents, product usage, and venture sizing.
Why investors are excited
- Jack Altman says Instinct represents a new paradigm: consumer agents that don’t just chat, but perform actions across the internet.
- The team compares it to the early days of ChatGPT and coding agents: once the behavior becomes essential, multiple winners can emerge.
- Benchmarks’ view is that this is a real platform shift, not just another app.
Key debate: usage depth vs. hype
- Jason argues the key question is not whether people like the product, but whether they will use it all day long.
- He contrasts:
- coding agents used for hours a day,
- legal workflows that become deeply embedded,
- and consumer agent apps, where sustained daily use is still unproven.
- The group agrees the product is early, but potentially huge if it becomes a true life-management layer.
Fund sizing and Kelly-bet logic
- The discussion gets into how venture firms should size early-stage bets that have late-stage pricing.
- Jack frames the round as both:
- an early-stage paradigm bet, and
- a growth-style check because the valuation is already massive.
- The broader point: variance is exploding, so funds need enough shots on goal to matter.
AMD Buys World Labs: M&A as the Exit Path for Neolabs
One of the biggest stories in the episode is AMD’s reported acquisition of Fei-Fei Li’s World Labs for $8.2B in stock.
Why it matters
- The panel sees it as validation that Neolabs can become meaningful acquisition targets.
- Harry argues AMD needs assets like this to stay relevant in the AI race against NVIDIA.
- Jack notes that in today’s market, there are roughly 10 companies capable of making $10B acquisitions, which changes the path to liquidity.
Broader implications
- The group believes many infrastructure and world-model companies may not be obvious standalone hyper-growth winners, but they can still be attractive strategic acquisitions.
- The market for AI M&A is becoming a critical outlet for large-capitalization startups.
Meta Poaches MongoDB’s CEO: The Talent War Is Real
The conversation moves to Meta reportedly hiring MongoDB’s CEO to lead its enterprise push, sending MongoDB shares down sharply.
What it says about the market
- The panel sees it as a sign that money, momentum, and the AI zeitgeist are pulling talent toward the hottest AI platforms.
- Jack emphasizes that it’s not just compensation:
- it’s the attention,
- the perceived importance,
- and the chance to work at the center of the industry.
Broader takeaway
- Top talent is becoming more mobile than ever.
- The panel sees this as both:
- a destabilizing force for companies outside the AI core, and
- a positive signal for society, because great talent is flowing to the highest-value opportunities.
AI Infrastructure, Compute, and the End of “Open” as the Default
The conversation turns into a deeper debate about inference, open source, and compute economics.
Key points
- Harry argues open source has likely peaked in market share.
- Jason agrees that frontier labs like OpenAI and Anthropic can always compete on price if they choose to.
- There is also increasing customer anxiety around:
- training on first-party data,
- security,
- and running open-weight models in enterprise settings.
Inference wins
- The panel is bullish on inference companies because:
- AI costs remain unsustainably high,
- and many tasks have become “good enough” problems where brute-force intelligence no longer matters.
- They mention the rise of companies like Modal, Base10, Fireworks, and others as beneficiaries of this trend.
Core framing
- The winner of many AI markets may simply be who owns the compute.
- The battle is increasingly about:
- tokens,
- efficiency,
- and access to infrastructure.
Oura Pulls Its IPO: Public Markets Are Still Hard
The episode also covers Oura pulling its IPO, despite being profitable and well-known.
Why the pull surprised the panel
- The company had major banks involved and appeared close to pricing.
- The panel was surprised because Oura seemed like one of the cleaner consumer IPO stories available.
What it suggests
- Even with strong businesses, IPO execution remains fragile.
- Secondary-heavy deals are especially delicate because insiders care deeply about price.
- The market may be receptive to AI-native IPOs, but there is still a lot of uncertainty around timing and pricing.
Monzo and Nubank: The Strategic Buyout Logic
The panel closes on reports that Nubank may acquire Monzo for $8B–$12B.
Why the deal makes sense to them
- Monzo is a strong company but likely too small to matter as a standalone public-market asset.
- Nubank may be buying time, not just customers:
- a faster path into the UK,
- strategic presence,
- and a broader geographic footprint.
Risks and skepticism
- Harry is skeptical because Nubank appears focused on the U.S. and Brazil, making a UK acquisition feel like a big strategic stretch.
- The team also notes board instability at Monzo and the appeal of simply “ending the pain” via sale.
Venture Capital in 2025: Bigger Funds, Bigger Checks, Bigger Variance
A recurring theme throughout the episode is that venture is being re-shaped by AI.
What changed
- Bessemer raising $5.75B and expanding into larger growth-style investing reflects the new reality.
- NFX’s move to stop raising external capital and invest only GP capital is the opposite end of the spectrum.
The shared conclusion
- Venture is becoming more polarized:
- enormous rounds for frontier AI,
- smaller but still viable seed checks in software,
- and much higher dispersion in outcomes.
- The old venture math still exists, but it’s harder to apply cleanly in a world where:
- seed rounds can be $30M,
- initial AI lab rounds can require hundreds of millions,
- and speed to market is dramatically compressed.
Key Takeaways
- AI is driving extreme valuation dispersion: some companies deserve massive checks, while others are increasingly commoditized.
- Talent is following the center of gravity: money, attention, and strategic importance are pulling people into the hottest AI firms.
- Compute is the new moat: who controls infrastructure may matter as much as model quality.
- Open source may be peaking in enterprise relevance, especially where security and data control matter.
- IPO markets remain unpredictable, even for profitable consumer companies.
- Strategic M&A is becoming a major liquidity path for AI startups that may not want to go public.
Notable Framing from the Episode
- “If this matters, it’ll matter a lot.”
- “We are either investing way too fast or way too slow.”
- “It’s all going to work” — a reference to the breadth of winners in AI infrastructure.
- “The market is sending a signal” — especially around talent and capital flowing into AI.