Overview of 20VC
Harry Stebbings, Rory O’Driscoll, and Jason Lemkin spent this episode on the biggest tech and venture stories of the week: OpenAI’s valuation reset, the broad AI trade selloff, DoorDash’s acquisition of Instacart, massive private-market rounds in AI and robotics, and what all of it means for pricing, demand, and liquidity across the tech ecosystem. The main thread was that the AI market is not “cracking” so much as going through a sharp period of price discovery.
AI Market Reset: OpenAI, Anthropic, and the “Wobbliness” in the Trade
OpenAI valuation cut
- OpenAI’s latest round reportedly came in at $1.25T, down from the previously discussed $1.4T.
- The round size was also reduced from $30B to $20B, with only about $11B committed.
- The hosts argued this is not a sign that the AI story is broken; rather, it reflects a comparison between:
- a bigger, faster-growing, more profitable, more liquid security, and
- a smaller, less liquid, still-loss-making private company.
Anthropic and pricing pressure
- The discussion spent a lot of time on Anthropic’s decelerating quarter-over-quarter bookings growth.
- Their view: the absolute numbers are still extraordinary, but pricing pressure is clearly showing up.
- OpenAI and others are seen as cutting prices aggressively, which may force the market to re-rate AI companies.
The bigger question: demand, not supply
- Jason’s key point: if AI demand stays strong, the current wobble will be forgotten in months.
- If demand weakens, then current financing volatility will look like an early warning sign.
- They framed current stress as a supply-side price discovery process:
- What should CoreWeave debt cost?
- What valuation is fair for illiquid AI leaders?
- How much should investors pay for future growth?
AI infrastructure risk
- The group highlighted that CoreWeave debt is yielding around 15%, which they described as a sign of market nervousness.
- They also pointed to Broadcom guaranteeing Anthropic-related debt as another example of how stretched and creative AI financing has become.
Public Market Meltdown: Alphabet, Shopify, and General Nervousness
Alphabet’s sharp drop
- Alphabet lost roughly $500B in market cap in a short period.
- Despite that, Google Cloud is still growing at a very strong rate, and the cloud business is doing extremely well.
- The concerns centered on:
- search cannibalization
- capex intensity
- broader AI-related anxiety
The tone of the market
- The hosts repeatedly described the market as “nervous AF.”
- Even strong companies are getting punished for coming in slightly below elevated expectations.
- Shopify was used as another example of the market’s current intolerance for misses.
DoorDash Buys Instacart: Consolidation and the Agent Threat
Why the deal makes sense
- DoorDash buying Instacart for $17B was framed as a logical consolidation move.
- The deal is a classic example of:
- buying a competitor while it’s cheaper,
- increasing network control,
- and potentially improving economics through scale.
Agents as a real threat
- The hosts made a strong argument that AI agents could be an existential risk to food-delivery platforms.
- The logic:
- Platforms like DoorDash, Uber Eats, and Instacart sit between the user and the merchant.
- Agents can help users bypass the platform by finding the cheapest or best direct option.
- That said, they also noted a counterpoint:
- Food ordering is still a discovery-heavy consumer behavior.
- Many users enjoy browsing menus and images, so the UX of full automation may not displace the app as quickly as feared.
Current adoption is still low
- DoorDash said agents had generated only a few hundred orders out of millions of daily deliveries.
- The hosts treated that as real but early:
- not a non-event,
- but not yet a mass adoption signal either.
Major Private Rounds: Kalshi, Cohere, Physical Intelligence, and More
Kalshi
- Kalshi raised $1B at a much higher valuation after strong momentum.
- The hosts described it as a “money deal”:
- the business is working,
- growth is strong,
- and investors are chasing momentum.
- They also discussed the regulatory tension, especially state-level pressure because the business is increasingly resembling sports betting.
Cohere
- Cohere raised a very large round, with the Canadian government participating.
- The hosts framed this as a sovereign AI story:
- governments want a domestic AI champion,
- but government support does not automatically create a great commercial outcome.
- They were skeptical about the lack of revenue transparency.
Physical Intelligence
- Physical Intelligence was discussed as being worth around $20B despite relatively little revenue.
- The takeaway: AI and robotics valuations are increasingly about the future, not current revenue.
- The team agreed robotics is real and exciting, but commercial deployment will be slower than software AI.
- They also noted the strategic appeal:
- if foundation model companies want to address both knowledge work and physical labor, robotics becomes a huge TAM expansion.
Dropbox vs. Physical Intelligence
- Dropbox cutting staff was used as a contrast:
- mature software with real revenue can still be de-rated,
- while a much earlier company with little revenue can command a huge valuation.
- The point: revenue multiples alone are no longer enough to explain value.
The Big VC Question: Base10 vs. Cognition vs. Serval
Jason was asked which company he would bring to IC among three very different bets:
- Base10: inference infrastructure
- Cognition: coding agent company, around $70B–$80B
- Serval: next-gen AI application in ITSM / service management, around $3B
Base10
- Attractive because inference can produce immediate revenue velocity.
- But the market is crowded, and commoditization risk is high.
Cognition
- Strong momentum in coding agents.
- The hosts think coding is one of the clearest AI categories with durable demand.
- But the valuation is already very rich, and it may be hard to justify from a pure multiple perspective.
Serval
- Jason’s pick for best upside/multiple.
- His argument: if it becomes a true next-generation ServiceNow, the upside could be enormous.
- The debate centered on whether Serval is truly expanding TAM or just taking share from existing enterprise software.
- Rory’s counterpoint: a lot of these “vertical AI” plays may end up just slicing up existing markets rather than expanding them meaningfully.
Legal Tech and AI: Crosby vs. the Old Guard
Wild, Gotschal, and law firm fragility
- A prestigious old law firm collapsing after partner disputes was used to illustrate how fragile rainmaker-driven professional services firms can be.
Crosby’s bid to acquire assets
- Crosby, the AI law firm backed by Sequoia and others, bid to acquire the assets.
- The hosts liked the audacity and timing.
- Their broader view:
- AI can absolutely take over transactional legal work
- but it is harder to know whether that becomes a $10B+ business
Their key tension
- High-end legal work is still relationship-driven and human-dependent.
- Lower-end legal work is easier to automate.
- So the likely outcome is a smaller but real AI-powered legal segment, not necessarily a total reinvention of the entire legal market.
Cybersecurity Warning: Ardex and the Agentic Threat
- The transcript flagged Ardex, an open-source Chinese pen-testing tool that was reportedly flipped into a real attack weapon.
- The hosts were alarmed by what this implies for the next phase of cyber risk:
- AI can massively scale offensive security tools.
- The danger is no longer theoretical.
- They connected this to broader “Black October” style incidents and warned that critical infrastructure could be next.
M&A, Liquidity, and Why Big Deals Matter for Venture
Salesforce and Cresta
- Salesforce’s purchase of Cresta for $2.7B was cited as proof that big platforms are still buying.
- The hosts think companies are essentially using M&A as a way to allocate “punch cards” across product areas:
- support,
- sales,
- service,
- workflow automation.
Why venture cares
- They stressed that venture liquidity depends on M&A working.
- If DOJ scrutiny or antitrust pressure slows major acquisitions, it becomes harder for VCs to return capital.
- They want big strategic buyers to keep doing regular, clockwork-like deals.
What buyers want
- Large buyers generally want:
- at least $100M in revenue
- enough scale to justify deploying their GTM machine
- Small, bright-and-shiny AI startups can be exciting, but many buyers prefer either:
- very small tuck-ins, or
- scaled assets with obvious revenue.
Key Takeaways
- The AI trade is not necessarily breaking, but it is clearly re-pricing.
- Pricing pressure is becoming a major theme across AI software.
- Public markets are nervous, and even great companies are getting punished for small misses.
- Agents are still early, but they are a real strategic threat to marketplace businesses.
- The best private-market bets are increasingly about future TAM expansion, not present-day revenue.
- Big-tech M&A remains crucial for startup liquidity and venture returns.
- Cybersecurity risk is rising fast as agentic tools get weaponized.
What to Watch Next
- Whether OpenAI and Anthropic can sustain growth while prices keep falling.
- Whether AI demand remains strong enough to justify massive infrastructure spending.
- Whether agents start meaningfully changing behavior in delivery, travel, and commerce apps.
- Whether big acquirers like Salesforce, Microsoft, and others keep the M&A market open.
- Whether robotics and coding agents become the next true AI mega-categories.