Neodrop
Create
FeedDiscoverMy ChannelsAuto-publishJoin Our Discord
FeedDiscoverCreateMy Channels
2026-10-21

Ep. 18: Chamath Palihapitiya: 70% of AI Revenue Is Venture Money. Also, This Market Will Go Absolutely Crazy.

VC Hot Take of the Week
VC Hot Take of the Week@Soren K
Ep. 18: Chamath Palihapitiya: 70% of AI Revenue Is Venture Money. Also, This Market Will Go Absolutely Crazy.

On the fifteenth of October, a venture capitalist with two and a half million followers posted that most of the money moving through artificial intelligence is venture capital wearing a customer's name.

0:00 / 10:08
On the evening of 15 October 2026, Chamath Palihapitiya started posting about how AI companies report revenue. Within hours he had put a number on the customer base: he suspected that seventy to eighty per cent of revenues at AI companies come from other AI companies that raised money from the venture industry. He called it a revenue quality problem rather than a circular one.
A day later, on his own podcast, he described the pricing change he thinks will make the same market go absolutely crazy.
This episode walks that week in order, because the order is the argument. The revenue-quality claim landed first, across four posts. The expansion forecast came second, from a different room. Both belong to the same man, and the gap between them is where the incentives are.
Chamath runs Social Capital, the investment firm he founded, and he is chief executive of 8090, a company that sells AI-built software and runs an enterprise delivery business. His portfolio's single best outcome in the past year was Groq, which he backed about a decade ago; Nvidia agreed in December 2025 to pay roughly twenty billion dollars for assets and a licence. His company raised a hundred and thirty-five million dollars in June 2026, in a round led by Salesforce.
Three kinds of material sit in this guide. The statements are public and timestamped. The portfolio facts are documented and cited. The incentive reading is analysis, and it is not a claim about anyone's intent.

Statement Timeline

All times show the channel's display timezone first, followed by UTC. Engagement figures are platform snapshots captured on 21 October 2026 at about 10:05, not counters from the moment of posting.
  1. 14 October 2026, 21:02 (19:02 UTC): Chamath endorses an immediate listing for Anthropic at a market-clearing price: "it's more critical for Ant to get public and get the $200B they need than to optimize for valuation." He puts the clearing price at roughly a trillion dollars. Captured: 469,108 views. 1
  2. 14 October 2026 (published 16:00 UTC): Ramp's monthly AI Index, written by lead economist Ara Kharazian, reports that the top one per cent of firms account for about 80 per cent of spend on OpenAI and Anthropic, and that per-employee spending among that top one per cent fell 3.1 per cent from its July peak. 2
  3. 15 October 2026, 18:00 (16:00 UTC): the first post. Chamath argues that run rate used to mean contracted revenue and that venture investors let AI companies convert it into last month's usage multiplied forward: "VC investors have enabled the AI industry to bastardize revenue reporting." Captured: 134,511 views, 808 likes, 104 replies, 69 reposts. 3
  4. 16 October 2026, 00:41 (15 October, 22:41 UTC): Social Capital posts the chart behind the claim: AI and machine-learning companies captured roughly 83 per cent of all US venture funding through September. Captured: 140,778 views. 4
  5. 16 October 2026, 00:48 (15 October, 22:48 UTC): Chamath quote-posts it with the sentence that carried: "I suspect 70-80% of revenues at AI companies are from other AI companies who've raised money from the Venture industry. This is not a circular revenue problem but it is a revenue quality problem." Captured: 247,993 views, 1,337 likes, 213 replies, 129 reposts, 57 quotes, 389 bookmarks. 5
  6. 16 October 2026, 11:03 (09:03 UTC): Benzinga turns the thread into a story, framing it as a warning about over-reliance on venture capital, and notes that prediction markets put a 7 per cent chance on an AI crash this year. 6
  7. 17 October 2026, 04:03 (02:03 UTC): the All-In episode publishes, with Ben Horowitz filling in for Friedberg. The running order puts CoreWeave's debt selloff second, at the two-minute mark. 7
  8. 17 October 2026, 04:39 (02:39 UTC): a clip of the same episode circulates: Chamath puts CoreWeave's credit default swaps at about 953 basis points, which he translates as roughly a 16 per cent annual default probability and a coin flip over five years. Captured: 119,159 views. 8
  9. 17 October 2026, 06:47 (04:47 UTC): a second clip, in which he names the boom's Achilles heel. The businesses borrow at a floating rate, so the spread over the risk-free rate is what breaks them, and the sector would need Microsoft, Google, Meta and Nvidia to backstop it in a bad year. Captured: 142,574 views. 9
  10. 17 October 2026, during the same episode (about 55:00): the expansion forecast. Comparing AI to AdWords, he argues the industry will move from buying a million tokens to buying an outcome, and that "when that happens, this market will just go absolutely crazy." 10
  11. 20 October 2026, 15:47 (13:47 UTC): Ramp's economist publishes the second half of the data: within the top one per cent of AI spenders, about 80 per cent are venture-backed, and he warns that a pullback would cascade. Captured: 79,091 views. 11
  12. 20 October 2026, 19:14 (17:14 UTC): Social Capital posts the outcome-pricing clip as its own item, quoting him on the shift that will make the AI market explode. Captured: 84,339 views. 12
  13. 20 October 2026 (Tuesday), New York: Citadel founder Ken Griffin tells Bloomberg's Dani Burger, at Citadel's Future of Global Markets Conference, that the AI run "almost certainly" looks like a bubble, and that corporate America has been uneven at extracting value from its models. 13
  14. 21 October 2026, 02:50 (00:50 UTC): the Griffin quote circulates on X through The Kobeissi Letter. Captured: 391,592 views, 2,360 likes, 254 replies, 174 reposts. 14
  15. 21 October 2026, 06:07 (04:07 UTC): Chamath attaches his own version to it. One per cent of companies, he writes, represent 90 per cent of AI spending; 80 per cent of that comes from unprofitable venture-backed companies; so roughly 70 to 75 cents of every AI dollar is venture money. He adds the line this episode spends its time on: "This hasn't happened yet because most of the breathtaking revenue growth comes from LPs investing in VC funds who then send it to startups." Captured: 205,135 views, 1,789 likes, 154 replies, 167 reposts, 41 quotes, 904 bookmarks. 15
  16. Not one statement: the two claims in this episode's title did not come from a single post. The revenue-quality claim ran on X between 15 and 21 October; the promise that the market goes absolutely crazy came from the podcast episode published on 17 October and was clipped and reposted on 20 October. They are about a day apart, in different formats, with no retraction between them.

What the episode decodes

The load-bearing word in the first post is "quality." Chamath separates himself from the people calling AI revenue circular, and the distinction is doing real work. A circular revenue problem means the same dollar is counted twice and the market is a mirror. A quality problem means the dollar is real, the customer is real, and the customer is spending someone else's capital. One is a fraud story. The other is a duration story: the revenue has a clock on it that the seller does not control.
Underneath that sits a customer-concentration claim, and this is where the numbers need care. Ramp's index measures business spend through one company's cards and invoices. It found that the top one per cent of firms account for about 80 per cent of spend on OpenAI and Anthropic. A second post put the venture-backed share of those top spenders at about 80 per cent.
Those two figures multiply to roughly sixty-four cents, not seventy-two. Chamath's version starts from a higher first number — he writes that one per cent of companies represent 90 per cent of AI spending — and takes eighty per cent of that. The gap between the two claims is one rounded figure. It moves the size of the estimate, not its direction, and it is worth remembering the next time the sentence is quoted with the word "roughly" dropped.
The dataset also has a boundary. It captures the software bills that run through cards and invoices. It does not capture what the hyperscalers spend building data centres, which is the largest pool of AI capital and is funded from their own cash flow. His claim is about the revenue that AI companies book, so the boundary is not fatal. It does mean the figure describes a slice of the market, chosen because it is the slice founders quote as annual recurring revenue.
Then comes the other claim, and the two are easier to reconcile than they first look — provided you notice what does the work in each. Outcome pricing is a change to the meter. The buyer still pays the invoice. If seventy per cent of those buyers are funded companies spending LP money, switching them from tokens to outcomes changes what they are charged for, not where the money comes from. A genuine change in the buyer mix would look like enterprises paying out of operating budgets, which is a different thing entirely.
Which is where the company he runs becomes relevant. 8090 sells a software factory and runs an enterprise delivery business for healthcare, insurance, life sciences, manufacturing, financial services and government — customers whose AI spending is a budget line, not a venture cheque. It raised a hundred and thirty-five million dollars in June 2026, led by Salesforce, with WNDR, Craft Ventures, The Production Board and LAUNCH, and angels including Nikesh Arora and Adam D'Angelo. The company built to sell AI to enterprises took its growth capital from venture investors, which is not a contradiction so much as a description of the market as it stands. 1617
His own book has the same shape. In his April 2026 letter, Social Capital's best outcome of the prior year was Groq, backed about ten years earlier, when Nvidia agreed to license it for roughly twenty billion dollars. That is cash from a public company's balance sheet, not from another funded startup — the kind of revenue he is now calling the good kind. 1819
Put those together and the definition of good revenue is not neutral. It flatters the asset he already holds, describes the customer his company already sells to, and points at the financing channel he says the industry must move into. The forecast of an explosion is not about the revenue he just called fragile. It is about the revenue base he is positioned in, once other people's budgets replace venture cheques.
The LP line pulls the same way. Naming LP money as the engine of the growth is a criticism that costs a general partner nothing to make in public — he is not selling AI revenue to anyone. It buys the thing a man who runs money needs most during a nervous quarter: the appearance of being the one who saw it first.

Where the record stops

Whether the claim is true is not settled here, and cannot be settled with public data. Revenue quality is not a reported metric. Nobody files it, no auditor signs it, and the seventy to eighty per cent remains an estimate introduced with the words "I suspect."
The portfolio side is thin in a specific way. Social Capital does not publish its holdings, its marks or its stake sizes. The Groq outcome is visible because Nvidia had to describe a twenty-billion-dollar transaction. What else sits in the book, and what has been sold, is not public.
The two claims also do not sit at the same level of confidence. The revenue-quality thread is stated as an inference from other people's data. The explosion forecast is stated as a market call with no data attached at all.
And the clearing event has no date. He says the industry needs one before reliable spending shows up in public-market earnings, and does not say when it comes or who pays for it.

What founders and allocators can take from it

For a founder: audit the payer mix before the growth rate. If most of your revenue is other funded companies' budgets, your growth is a function of their next round, and that is a customer question that a better valuation argument will not fix. If your customers pay from operating budgets, the pressure you should watch is price, because Ramp's data shows blended token prices falling while volumes rise — a vendor charging by the token has to run faster every quarter to stay level.
For an allocator: when a well-known investor publishes a quality standard, ask which of his own companies already meet it, then watch which customers his next round comes from. A standard that describes the portfolio you already own is not worthless — it is a map of where that investor expects the money to go. The tell is not the diagnosis. It is the customer list.

References

  1. 1
    Chamath Palihapitiya on X: how to get Anthropic public

    x.com

  2. 2
    Ramp Economics Lab: October 2026 Ramp AI Index, Cracks in the AI Thesis Part 3

    ramp.com

  3. 3
    Chamath Palihapitiya on X: VC investors have enabled the AI industry to bastardize revenue reporting

    x.com

  4. 4
    Social Capital on X: AI companies captured ~83% of all US venture funding

    x.com

  5. 5
    Chamath Palihapitiya on X: 70-80% of revenues at AI companies

    x.com

  6. 6
    Benzinga: Chamath Palihapitiya flags revenue quality problem in AI

    benzinga.com

  7. 7
    All-In Podcast on X: the 17 October 2026 episode

    x.com

  8. 8
    Fireside Alpha on X: Chamath on CoreWeave's default odds

    x.com

  9. 9
    All-In Podcast on X: Chamath explains the AI boom's Achilles heel

    x.com

  10. 10
    All-In Podcast: full transcript of the 17 October 2026 episode

    podcasts.happyscribe.com

  11. 11
    Ara Kharazian on X: the top 1% of companies drive 80% of AI spend

    x.com

  12. 12
    Social Capital on X: the pricing shift Chamath thinks will make the AI market explode

    x.com

  13. 13
    Bloomberg: Ken Griffin says AI almost certainly in a bubble right now

    bloomberg.com

  14. 14
    The Kobeissi Letter on X: Ken Griffin says AI is almost certainly a bubble

    x.com

  15. 15
    Chamath Palihapitiya on X: $0.70-75 of every AI dollar

    x.com

  16. 16
    8090: why we raised our Series A

    8090.ai

  17. 17
    Business Wire: 8090 raises $135M Series A to accelerate rollout of Software Factory

    businesswire.com

  18. 18
    Chamath Palihapitiya: 2025 annual letter

    chamath.substack.com

  19. 19
    CNBC: Nvidia buying AI chip startup Groq for about $20 billion

    cnbc.com

VC Hot Take of the Week
VC Hot Take of the Week@Soren K

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.

Related content