Will Jensen Return to NVIDIA | Kalshi, Mercor, Fractile: Mega-Rounds Continue
20VC with Harry Stebbings
Jensen's exit and return 0:00
The discussion starts with Jensen Huang’s removal from Nvidia and the shock around it. The speakers say a board might have been better off pausing with a leave of absence and an interim chief, instead of making a hard, rushed call. They note the personal allegations, the export-control questions, and the board’s long ties to Huang. One view is that he may not come back unless the board changes. Another is that the board committee on chip imports and exports points to deeper risk.
Anthropic and the IPO 10:01
Attention then shifts to Anthropic’s huge IPO, which is described as the largest ever and badly oversubscribed. The speakers say the Nvidia episode helps Anthropic and hurts the open-weight and China-linked chip strategy. They focus on how little staff a company like Anthropic would need to reach $100 billion in revenue, comparing it with Google and Meta. That gap is used to argue that AI companies may create far fewer jobs than people expect, even as the people who remain can command more money.
AI jobs and layoffs 14:30
The speakers push back on the idea that AI will be disastrous for labor. They say a few strong people can do far more work when they can use compute well, but that old firms will not turn into machines with ten times the output. The change they do see is in front offices and back offices, where automation is already moving fast. They point to the rumored 40% layoffs at ServiceNow and to food delivery as signs of a wider shift.
Anthropic lifts the bar 16:01
Anthropic’s IPO is treated as a major event for venture. It is described as one of the biggest listings ever by dollars raised, but with the striking fact that it happened in about five years, not decades. That speed makes the possible look much larger. Early backers are expected to do extremely well, and the result is said to raise LP and GP appetite for fresh bets. Deal counts may fall while dollars keep rising.
Prediction markets and Mercer 28:32
Prediction markets come up first, then the talk turns to Mercor as a strong business with clear momentum. The sense is that it has pushed into the market well and that its closeness to the labs has helped. Brendan is singled out as impressive. The group jokes that it could put in $200 million, take a small loan position, learn from it, and move on to the next deal.
Jensen and deal risk 29:31
The conversation shifts to Nvidia and Jensen. They say his absence could weaken liquidation preferences because Nvidia, or people around it, might no longer be there to buy companies, backstop them, or rescue teams. Hugging Face is used as the example. If regulatory pressure grows, a deal like that could stall. They think the risk has gone up and hope the breakup fee was good.
Software buying slows 42:32
SMBs are not just shifting to new tools. Many are simply choosing not to buy older software for sales and CRM work. HubSpot, Monday, and similar products now look less essential, even when they are not expensive. The money is going into things that replace human labor right away, like agents that answer calls, reschedule appointments, and handle billing. A small business will sign for that same day. It will often delay the broader software stack.
Fragile systems and security 47:30
Microsoft’s outage, the FAA ground stop, and other failures are treated as signs of a weaker base of systems. These were old-fashioned mistakes, like a router problem or a load issue, not just AI cyber attacks. But the same agent tools that make hacking easier also make overload easier. The worry is that every network, plugin, and mission-critical service will be probed, broken, or stressed until companies have to spend much more on reliability and security.
Vendor breaches and M&A 56:30
Fireworks’ data exposure comes up as another sign that vendors are being judged on the weak edges of their systems. The talk turns to how Cursor, Harvey, and others have openly called out suppliers, and how much harder the tone has become. That leads to OpenAI buying Adessia, a travel-focused assistant from Sonder founder Francis Davidson, and to the question of whether vertical personal assistants are giving way to broader horizontal ones.
Agents reshape workflows 58:30
The discussion shifts to Muse, which is already wired into many workflows for outbound work, competitive review, clips, content editing, web traffic analysis, and deal screening. It can run around the clock, talk to other agents, and even use email and browser tasks to replace tools like marketing automation. The point is that once an agent reliably handles a job, you stop changing it, and teams begin to let agents do the work first.
Agents and labor 1:10:31
The speakers say AI agents can replace people, but someone still has to manage them for long hours. They joke about wanting to be like the best-funded investors, and they compare big venture firms by fund size and returns. The talk then turns to SpaceX, where Starship is praised as a major technical win.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.

