Anthropic's $1.4 Trillion IPO Strategy Explained

Three years ago, Sequoia Capital decided it had solved one of venture capital's oldest problems: when to sell. Its answer was an Evergreen fund, a structure that let the firm keep holding stock in its own portfolio companies for years after they went public, on the theory that the real money in venture is made after the IPO, not before. It was, as one of this week's hosts put it, arguably the best idea in venture history. This week, Sequoia killed it.

That reversal is the small, structural hinge of a much bigger week, one in which several things that looked permanent, a fund structure, a founder's marriage, a software pricing model, a market rally, all got repriced at once. Sequoia did not just end its Evergreen fund. It closed its London office, folding European staff back into the US team, and shut down its Scout program, the quiet system of seed checks written by founders and operators on Sequoia's behalf that had been a hidden edge for years (03:54). The firm also merged its early-stage and growth teams into one partnership.

The Evergreen logic was not crazy. Sequoia has historically been linked to something like 25% of the NASDAQ through its portfolio companies (05:44), and a diversified basket of its own winners, held for decades, should beat an index over time. But the fund launched in 2022 and 2023, just as stocks turned down (05:34), and its structure effectively trapped limited partners, the outside investors who fund a venture firm, in positions they could not exit even if they needed the cash. On top of that, individual partners who had already collected their carry, the share of profits a fund manager keeps, had little reason to force anyone else's hand. One host summarized the unwind bluntly:

What the hell happened to the Evergreen idea? This is the best idea in venture ever.

The hosts frame all of it as a reminder that institutions outlast individual arrangements, but arrangements still get rebuilt when they get too complicated to run.

NVIDIA chief executive Jensen Huang, the longest-tenured CEO currently running a major tech company, having led the chipmaker since 1993 (15:52), is now the subject of a board investigation after admitting to a relationship with a person suspected of being a Chinese intelligence contact. The probe itself is not about the relationship. It concerns whether NVIDIA's export compliance and security practices were compromised (14:24). The hosts argue Huang is too important to the US AI industry to lose over this, not because the allegations are trivial, but because his personal conviction that NVIDIA should sell chips to any buyer, including, where legally possible, Chinese customers, long predates any entrapment and makes him hard to replace. The awkward detail is that the investigation is tangled up with Megaspeed, a company NVIDIA has since banned from buying its chips, over the unresolved question of whether it is acceptable to train Chinese models on GPUs parked just across the border.

The week the AI trade got repriced

The NASDAQ 100 lost $2.5 trillion in value in a single 24-hour stretch (21:30), enough to put hedge fund managers into what one host described as a depressive state. The explanation offered is not a crack in AI demand but a mechanical one: leveraged funds get their capital pulled by prime brokers like Citadel at a 5% drawdown and get fired outright at 7.5% (23:14), so a wave of margin calls can move trillions without any real change in the underlying business. As one host put it:

The AI trade doesn't crack on the supply side. It is far better in an IPO to start low and go up than to start high and come back.

That second half of the line is about Anthropic, which is moving toward an IPO with a recommended price range of $70 to $78 a share, above its last private round of $54 a share but deliberately modest (24:25). The plan, the hosts argue, is to generate demand at a conservative price and then raise the range, rather than price high and have to retreat. On a vested-options basis the company is valued near $1 trillion; counting every option ever granted, including unvested ones, the figure is closer to $1.4 trillion (25:12). Anthropic is also expected to carry roughly $200 billion in cash on its balance sheet, reportedly more than any other company in America (27:35), a cushion that lets it tap public markets through converts and secondaries once listed instead of running the private fundraising circuit that OpenAI is still on. OpenAI, for its part, reported 70% annualized revenue growth in the third quarter, with enterprise revenue now a majority of the total (29:43), even as it closed an $11 billion private round at a $1.25 trillion valuation rather than going public.

Both companies are also discovering that agents are expensive to run. A persistent AI agent, the kind that works continuously rather than spinning up for a single task, can cost $20 to $30 a month per user in compute (32:44), far more than the few dollars it costs to run a simple website-building tool. That cost structure is already showing up in how venture investors value AI infrastructure: this week's deal slate included Zipline, the drone delivery company, raising at a $21.5 billion valuation, triple its January price (34:10); Fal, an AI inference company, doubling to $16 billion on more than $800 million of annual recurring revenue; and Revolut, the London-based fintech, trading at $115 billion in a secondary sale, up 53% from its last round (34:20). The hosts split on which is safest. Revolut, with higher margins and a young customer base aging into mortgages and lending, was the preferred pick for durability, while Fal was chosen anyway for its growth, despite gross margins estimated around 6% and a market that could commoditize within a few years.

The same AI boom has a darker side. Fireworks AI, an inference provider, suffered two major security breaches in a 30-day window, both traced to stolen API keys, the credentials that let software call AI models and get billed for it (40:35). Nine leading AI developers disclosed similar swarming bot attacks in October alone, according to security firm Wiz (43:28), with another security firm estimating attackers can run up a million dollars in charges before anyone notices. One host called it the most lucrative criminal trade available right now, more profitable, in his words, than stealing watches off someone's wrist in London (41:40).

Who survives the agent economy

If agents are expensive to run, they are also cheaper than headcount, and that tension is reshaping how software companies are valued. Palantir now generates about $2 million in revenue per employee (50:18), a figure the hosts attribute less to any AI magic and more to a simple truth: when revenue grows faster than a company can hire, margins take care of themselves. Cloudflare cut 20% of its workforce earlier in the year and then saw growth re-accelerate to 39% (51:48), which the hosts read as evidence that agentic spending, not human seat licenses, is now driving its business. Figma, by contrast, fell below a $10 billion valuation for the first time since its aborted 2022 sale to Adobe (53:01), a company that was once offered more than $20 billion. The hosts argue the market no longer believes seat-based software, priced per human user, benefits from the agent shift. As one host put it:

Agents don't care about your UX and their UI, and they don't care about any of this stuff. Humans do, but they're a declining percent.

That same logic explains why Dialpad's $400 million acquisition of the chatbot company Ada (45:04) reads less like an exit and more like a soft landing for a company built on a pre-large-language-model architecture that could not retool fast enough. The hosts argue older, non-AI-native companies simply cannot ship software at the pace their newer competitors can, a gap that is now generational rather than a matter of talent.

The episode closes on California's rejection of a proposed wealth tax, defeated by a wide margin even as the state, and the country, swung heavily toward Democrats in the same election (61:15). The hosts see the vote as practical rather than ideological: voters recognized that a wealth tax risked pushing newly rich tech workers to lower-tax states just as Anthropic prepares to mint extraordinary sums for ordinary employees, with the average Anthropic employee expected to gain about $41 million in the IPO (62:49). It is, in its way, the same question Sequoia faced with its Evergreen fund and its London office: how much do you force people, whether limited partners or taxpayers, to hold onto something simply because it would be good for them. This week, across venture, chips, and tax policy, the answer kept coming back the same way. Let people go.