20VC with Harry StebbingsGoogle Loses $500B in 12 Hours | DoorDash Buys Instacart for $17B | Kalshi Raises $1B
20VC with Harry Stebbings
OpenAI valuation reset 1:00
OpenAI’s valuation is cut from 1.4 trillion to 1.25 trillion, and the round is trimmed from 30 billion to 20 billion with only about 11 billion committed. The change is framed as price discovery, not a sign that the AI wave is breaking. The larger, faster-growing, more profitable company is the one people can buy later in public markets, while OpenAI is smaller, still losing money, and illiquid.
AI trade looks wobblier 3:00
The speakers say the whole AI trade has become nervous and uneven. They point to Anthropic’s softer quarter, CoreWeave debt paying 15%, Broadcom guaranteeing Anthropic debt, and thin, wobbly IPOs. If AI demand stays strong, the bond and equity worries fade; if demand weakens, the stress will hit the smaller players first. The market is still sorting out what buyers will pay for future growth.
Cheaper agents 14:00
[14:00] Switching an AI recruiting app to Haiku 55 and Jev cut costs about 70% to 72% for the same B2B workflow, with higher quality. That was new enough to shake the speaker’s view of the market. He says model quality and token pricing have changed so fast that it is hard to predict what these tools and their costs will look like in 12 months, let alone five years.
Private software pressure 16:02
[16:02] That uncertainty leads into Canva, Automatic, and similar software businesses. If agents do the work, some products lose their place. The speaker says many firms may stay private and cut costs, but if they do not adapt, they drift toward quiet obsolescence. He is also surprised that real customers, not just tech users, are already pulling back from products like Canva and Squarespace.
Agent risk and delivery 27:31
DoorDash is used to argue that AI agents may change app businesses only a little at first, but still enough to matter. The speaker says order discovery, pictures, and human choice keep food apps useful, while agents fit better for simple repeat tasks and scheduling. He thinks Uber faces the same kind of pressure, though he still likes Uber as a cheap stock. Their real worry is not one big disruption, but steady small leakage to agents.
Big rounds and cyber risk 32:01
Koshi’s $1 billion round at a $40 billion valuation is treated as a momentum deal, even with Utah suing its founders. Cohere’s Canadian-backed raise is framed differently, as a sovereign AI bet that may matter politically even if the business is not obvious. The talk then turns to Binance, FTX, and whether firms can really cover huge losses, before ending on Artex, an open-source pentesting tool that was quickly turned into a broad attack weapon.
Infrastructure Outage Risks 40:31
A broken wire for a day means money cannot move. The discussion widens that point with the refrigerator hack in Europe, where someone pushed temperatures up to 50 degrees until it was fixed and the frozen peas were saved. The point is that the same kind of breach could hit much more serious systems. Traffic lights, heating plants, and other mission-critical services all sit in the same danger zone.
Choosing AI Deals 41:02
Three companies sit before the committee: Base 10 at about $30 billion, Cognition at $70 billion to $80 billion, and Servil at $3 billion. The case for Cognition is speed and scale in coding agents, but the speaker says Servil offers better upside from a lower price, with a real shot at a tenfold return. The harder question is whether Servil truly expands the market or only takes share from ServiceNow. That decides whether the bet is real growth or just a slice of an old market.
Law work and automation 54:30
The discussion turns on whether a law firm built around AI can hold on to lawyers and still make money. One side doubts that partners will give up profits for equity in a company that may never reach an IPO. The other says some legal work depends on a named rain maker, but much of it is routine. For NDAs and Series B stock purchase agreements, you do not need the famous five-thousand-dollar-an-hour lawyer. Automation should also cut mistakes from junior staff.
Robotics and M&A 59:01
The talk then shifts to robotics and company buying. Physical Intelligence is described as a fast-rising robotics company with little revenue, and the point is made that future value can matter more than current sales. New models are said to show real dexterity, but robot deployment will take far longer than software. Big model companies may buy in to double their market and sell to physical workers as well as knowledge workers. There is also concern that antitrust scrutiny could slow normal mergers, which venture investors need to return capital.
Small firms get absorbed 1:08:00
The speakers say the same story plays out again. A bigger company sees how to use something once it grows large enough, and then it can make that material work inside a $50 billion business. They call it a shame, but likely true, because small firms cannot be assimilated until they are big enough.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
