Roughly a hundred interesting startups used to be in play at any one time. David Cahn says the conversation in venture capital has narrowed to about ten companies.
Most of the money chasing those ten is not venture money at all. Cahn's argument is that the entire pool of global capital has crowded into one slice of the market — late-stage, pre-listing artificial-intelligence companies — and that slice is about to meet the public market for the first time.
"As soon as one of these companies gets out, then I think you actually have a better valuation framework for the rest of the market."
Cahn is a partner at Sequoia Capital, was an early investor in Hugging Face, and wrote the essay that framed the data-center build-out as three separate industries rather than one.
I listened to the full segment so you can skip it. 17 minutes of audio, 10 minutes of reading.
Here are the 8 takeaways that matter.
👤 Guest: David Cahn, Partner at Sequoia Capital, an early investor in Hugging Face who writes on the economics of the AI build-out
🎙️ Hosts: John Coogan and Jordi Hays, who present TBPN live on X and YouTube every weekday
📰 Published: 9 September 2026 on YouTube (TBPN)
🔴 YouTube | 🟣 Apple Podcasts | 🔗 Show notes | ⏱️ 17 min | ✅ Time saved: 7 min
Key Takeaways
Venture used to track a hundred companies at once and now tracks about ten
The capital crowding into them is global capital, not venture capital
There is no public comparison for any large private AI company yet
The first big listing sets the framework everyone else gets valued against
The neoclouds' real innovation was financing, not technology
Anyone can start one; there are already 15 and more coming
Energy will not follow the neocloud pattern because it needs patents, not just capital
Form Energy spent nine years building batteries before signing a project with Google
Cahn will not back a hardware seed round if he doubts the founder can raise a billion dollars
The partnership's test on a founder is whether they would go and work for that person
Power is already a late problem — the shortage runs for about the next two years
Nuclear is a 2040s problem, which is why an early-stage company can still start on it now
1. The Top 10 Own the Talk
Cahn opened on how the job has changed rather than on any single company.
His starting point is that attention has collapsed onto a handful of names. "When you think about doing this job five, six years ago, you had a hundred interesting companies that everybody was looking at and you're trying to find the next one and that's still happening and yet so much of the conversation today is about the top 10 companies."
The hosts pushed on what is actually driving the change — the concentration of capital into the largest model companies, or something else
The hosts' own theory was liquidity, not size. They argued that Hugging Face and OpenRouter created exits worth more than $10 billion apiece, which changes the arithmetic for an investor writing a billion-dollar-valuation check into an early round, because ten-times returns now exist at that scale
2. Hugging Face's Option Value
Cahn said the pattern is old, not new. Sequoia was an early investor in Hugging Face, and the thesis was community and bottom-up adoption carrying option value that a spreadsheet cannot price at the time
He expects more of it. Investors ask a hundred questions about what a community company becomes, and the pull of the market answers them later
His framing of the bar: it is hard to be relevant and build something important, and both Hugging Face and OpenRouter cleared it
3. No Comps Until an IPO
The comparison he reached for was 2021. CrowdStrike and Datadog were public, so an investor buying private versions of the same business model knew roughly how the thing would trade once it was liquid. Databricks sat in that frame
Nothing plays that role for AI. There is no listed company whose numbers a private AI valuation can be checked against
One listing fixes that for everyone. "As soon as one of these companies gets out, then I think you actually have a better valuation framework for the rest of the market."
He called the transition healthy rather than dangerous. Being able to look at the liquid and the illiquid side by side gives the private market clarity it does not have now
The hosts added a caution about the shape of it: lockups stagger the sale, many limited partners will take stock and hold, and registered investment advisers will hold too. Optionality on liquidity is not the same as cash on a balance sheet ready for the next generation of startups
4. SpaceX Unlock Was a Dud
Cahn treats SpaceX as the bellwether for the whole sequence. It has held up through liquidity unlocks that the press had expected to break it
The event nobody was worried about afterwards. "I remember the day of the big unlock somehow traded up like 7% if I remember correctly." He put that down to a short squeeze
The point he drew from it is that this will not be a single day. Demand for these companies is real and they are growing quickly, so the transition plays out over months rather than in one wave
5. Server, Steel and Power
His second focus is the full stack underneath the models. "Server, steel, and power are kind of the three pillars."
Servers are the part everyone watches, which is the chips. The hosts made that point and Cahn moved past it
Steel means the industrial side, and his example was Elon Musk building his own gas turbines. The constraint he named is that the components that go into a data center cannot be made faster
Power is the one with a hard clock on it. Cahn said there is no more power available for roughly the next two years, and that capital projects of this size have to be planned two, three or four years ahead — which means committing to a view of a world nobody can describe yet
He added that investors are now well educated on the compute market and familiar with every part of the stack, and expects innovation across all of it
6. Grid Needs IP, Not Capital
The hosts asked whether the opportunity in power looks more like the neocloud market — many good outcomes, no dominant winner — or like the rocket-launch market, where one company took everything.
Cahn's answer was the rocket case, and his reason was what the neoclouds actually invented. "I think on the Neocloud side, the primary innovation was financing." Buy chips from Nvidia, finance them with Blackstone, repeat
That is why the field filled up. "Getting access to power is hard. It's not easy, but there's 15 neoclouds. There's going to be more."
Hardware works on a different clock. Citing a partner's hardware manifesto, Cahn described a company storing what he called potential energy: "And these hardware companies spend five, six, seven, eight years building this potential energy." At some point it converts to kinetic energy and the share price moves. SpaceX, on his count, took 24 years to bake
Form Energy is his worked example. Sequoia announced the investment last week. The company builds grid-scale batteries, has been at it nine years, and signed a large project with Google earlier this year. The founding team includes the person who started Tesla's energy business and a battery scientist Cahn put among the top three alive
The logic of the bet is a chain of simple questions. There will be a lot of solar and wind; that means a lot of batteries on the grid; a company that spent nine years building the intellectual property to put them there is the one that scales into it
His conclusion is three, four or five companies with deep intellectual property rather than fifteen with good financing. Sequoia's nuclear investment is the other piece he named
7. Can This Founder Raise $1B?
In hardware, fundability comes before quality. Cahn told the hosts what he says to seed-stage hardware founders: "even if I think your IP is amazing I can't invest if I don't think you can raise a billion dollars because you just won't get there." The alternative outcome is being consolidated into somebody else
What he looks for is selling ability, and he thinks it is partly innate. He named two El Segundo founders whose ability to sell the vision was, in his words, on another level before they had raised $10 million
The category has to allow the pitch. A founder cannot sell a trillion-dollar vision in a market that does not contain one
The partnership's decision test is a single question. "And the framework we often use is would you go work for this person?" His reasoning: taking a job puts all of an employee's equity into one company, so the willingness of others to do it prices the founder
The second test is whether the story holds together. Form Energy's pitch, as he retold it, is that iron is cheap and air is cheap, so build an iron-air battery, and batteries are needed because the sun goes down
Clarity is a proxy for recruiting. "If you can articulate to an investor a very clear articulation of why this vision makes sense, why your plan makes sense, then you're going to be able to articulate that to an employee." A partner with a physics doctorate handles the science underneath
The relationships are long. Cahn said he had known the Form Energy chief executive for six years before investing, and one host summarized it back: the decision is not made in a single pitch
8. Synchronize the Timelines
One host put a long question to Cahn: software is getting faster while the physical world is not, gross domestic product growth has not arrived, and the more exciting cycle is machines building machines rather than another software feature shipped in two weeks instead of twenty.
Cahn's first answer was about matching clocks. "I think it's so important to synchronize the timeline of your investment with the timeline." The company's timeline and the market's have to arrive together
His example is unforgiving on power. "So for example if you're trying to solve AI power and you're starting today it's kind of late you need we need more power like literally in a year." A founder solving that should have started years ago
Nuclear is the opposite case. If the inflection is in the 2040s, an early-stage company chasing it is on schedule
He linked the shift to hardware back to moats. Digital timelines move so fast that it is hard to build a durable advantage, and years of physical work is where investors now think one can be built. The end state he named is "the machine that makes the machine" — the re-industrialization of the country
The hosts' own framing was that this cycle will be the biggest one. Generating an image on a screen is a trick; telling a system to make a thing in the world and having it appear is the version that reaches people who do not work in software
Cahn noted that science fiction tends to run ahead of reality, citing Isaac Asimov imagining the most valuable company in the world as the one that makes the robots, and said that is part of why some investors hold Tesla
Bonus Insights
Cahn's opening line to the hosts was that they had covered a lot of ground before he arrived, from Mark Zuckerberg's castle to the probability of human extinction. The hosts asked, half seriously, whether the extinction number should be factored into Series A financial models
He described his own habit of joking with his brother about market commentary: a stock moves 2% and a headline explains why, when 2% moves happen every day and nothing happened
The hosts noted that a Sequoia partner had already been on the show discussing the firm's nuclear investment
Cahn's bottom line is that the private AI market is being priced without a reference point, and the first large listing will supply one for everything else — while the durable money in the build-out sits in companies that spent years building intellectual property in power and hardware rather than in the ones that assembled capital quickly.
Products, Companies & Tools Mentioned
Sequoia Capital (Cahn's firm — he described partner meetings spent on how the AI market converts and what the firm underwrites)
Hugging Face and OpenRouter (The two community-led companies Cahn used as proof that bottom-up adoption carries option value investors cannot price early)
SpaceX (His bellwether for how private AI names will trade — held up through liquidity unlocks and, on his recollection, traded up about 7% on the day of the big one)
CrowdStrike, Datadog and Databricks (The 2021 comparison: public companies whose trading told you what the private version was worth, which AI has no equivalent of)
Nvidia and Blackstone (The chip supplier and the financier behind what Cahn called the neoclouds' real innovation)
Form Energy (Sequoia's grid-scale battery investment, announced last week — nine years of work, iron-air chemistry, and a large project signed with Google this year)
Google (Named as the counterparty on Form Energy's big project)
Tesla (Where Form Energy's chief executive started the energy business, and the stock Cahn says some investors hold for the robot thesis)
Books & Resources Mentioned
Steve Jobs Would Have Loved AI, and Been Hated for It (The show's own notes for the day, which list the running order and the headlines behind the news block)
Isaac Asimov's I, Robot (Cahn's example of science fiction preempting reality — the most valuable company in the world as the one that makes the robots)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: