---
title: "Dan Loeb sold Amazon and set AI's payback at $2-3 trillion of new revenue a year"
description: "The Third Point founder's letter and his fund's latest 13F rank the AI spenders by one rule: who can fund the wait for revenue that does not exist yet. Friday's rout applied the same ranking as the Nasdaq erased its 2026 gains."
publisher: "The Inference"
section: "Economy"
published: 2026-11-21T00:08:58.432Z
modified: 2026-11-21T00:08:58.432Z
canonical: https://theinference.org/article/dan-loeb-sold-amazon-and-set-ai-s-payback-at-2-3-trillion-of-new-revenue-a-year
language: en
keywords: "AI Capex, Hedge Funds, Third Point, Nvidia, Nasdaq"
---

# Dan Loeb sold Amazon and set AI's payback at $2-3 trillion of new revenue a year

> The Third Point founder's letter and his fund's latest 13F rank the AI spenders by one rule: who can fund the wait for revenue that does not exist yet. Friday's rout applied the same ranking as the Nasdaq erased its 2026 gains.

Dan Loeb has turned the AI payback debate into arithmetic, and then into a trade. In his [latest letter to investors](https://finance.yahoo.com/technology/ai/articles/billionaire-dan-loeb-bear-math-215908050.html), the [Third Point](https://theinference.org/markets/companies/third-point) founder wrote that the $5-7 trillion consensus spend by hyperscalers and frontier labs between 2025 and 2030 requires $2-3 trillion of new AI revenue every year, against roughly $6 trillion of total global IT spending today [1].

The [13F his fund filed on November 16](https://www.cnbc.com/2026/11/16/chips-big-tech-and-more-what-hedge-funds-bought-and-sold-in-the-third-quarter.html) shows the conclusion he drew: all 1.75 million [Amazon](https://theinference.org/markets/companies/amazon) shares sold by September 30, and about $1.05 billion of new positions in [Meta](https://theinference.org/markets/companies/meta), [Microsoft](https://theinference.org/markets/companies/microsoft) and [Nvidia](https://theinference.org/markets/companies/nvidia) [1][2]. The Nvidia purchase says the buildout continues. The Amazon sale says the wait for revenue has a cost, and not every balance sheet can carry it.

## The bottleneck is the revenue, not the compute

Loeb's own scale check is stark: the two leading frontier labs bring in well over $100 billion of combined run-rate revenue, a fraction of the annual target [1]. Follow the chain and the constraining link is easy to name. The spending buys capability: chips, buildings, power. The payback requires products people pay for, and nothing in the letter says those exist at anything like the required scale.

*Chart: **AI must find $2-3 trillion a year, a third to a half of all IT spending** Bars chart of annual figures in US dollars: total global IT spending today of $6 trillion, new AI revenue required each year of $2-3 trillion, and the two leading frontier labs' combined run-rate revenue of over $0.1 trillion.*

*Annual figures, US$ trillions, as set out in Third Point's letter to investors: the new AI revenue the consensus spend requires against today's total global IT spending and the two leading frontier labs' combined run-rate revenue. [1]*

The squeeze is already measurable. The Inference's [reinvestment-burden index](https://theinference.org/indexes/hyperscaler-reinvestment-burden), the share of the hyperscalers' cash generation consumed by capital spending, reached 94.8% in the September quarter [3]. Compute is arriving on schedule. Revenue is the unfinished middle, and whoever builds it is doing the work the trillions are betting on.

## The trade ranks the spenders by who can wait

The ranking rule is free cash flow. Microsoft still produced $18.6 billion of it in the September quarter [4]. Amazon's September quarter was a $33.5 billion outflow [5]. Its chief executive, [Andy Jassy](https://theinference.org/markets/people/andy-jassy), expects roughly $220 billion of capital spending this year [1].

[Oracle](https://theinference.org/markets/companies/oracle) has burned cash in seven of the past eight quarters [6]. It sold $20 billion of stock to keep building [1]. [OpenAI](https://theinference.org/markets/companies/openai), by S&P's reading, stands behind about half of Oracle's $664 billion backlog [1].

*Chart: **Oracle's capex has outrun operating cash flow in seven of eight quarters** Column chart of Oracle's operating cash flow against capital spending each quarter from late 2024 to August 2026. Capex is above operating cash in seven of the eight quarters, reaching $28.5 billion against $23.1 billion in the latest.*

*Operating cash flow and capital spending, US$ billions, from SEC filings, at Oracle's fiscal quarter ends (Q4 '24 ended November 2024; Q3 '26 ended August 2026). Source: Oracle filings. [11]*

Meta, now Third Point's largest reported holding at $475 million, is the qualified yes: its own September quarter slipped just below zero, and analysts see the burn deepening into 2027 [1][7][8]. The 13F gives no reason for the Amazon sale; the cash-flow pattern is the visible one [1].

*Chart: **Microsoft and Meta still fund the build; Amazon's cash gap hit $33.5 billion** Line chart of quarterly free cash flow for Microsoft, Meta and Amazon from Q4 2024 to Q3 2026. Microsoft stays positive throughout at $6 billion to $26 billion; Meta slides from about $14 billion to slightly negative; Amazon, charted from Q4 2025, falls to minus $33.5 billion.*

*Quarterly free cash flow, US$ billions, from SEC filings, at each company's fiscal quarter end. Amazon is charted from the December 2025 quarter because of a break in its earlier series. Sources: Microsoft, Meta and Amazon filings. [12][13][14]*

## Nvidia is the tell that the orders keep coming

Owning the chip seller into the spenders is a bet the orders keep arriving, and they are. Nvidia's October quarter brought about $115 billion of revenue [1]. Management guided the next quarter near $130 billion [1]. But the seller has joined the financing chain: days sales outstanding rose to 64 as Nvidia extended payment terms for investment-grade customers, one more balance sheet carrying the wait [8].

Friday forced the same test on everyone at once. [Anthropic](https://theinference.org/markets/companies/anthropic), [the largest IPO in history at $81 billion raised](https://theinference.org/article/anthropic-ipo-nears-500-billion-in-orders-before-nasdaq-debut), gave back much of its first-day pop [9]. The Nasdaq 100 fell [5.7% in The Inference's index](https://theinference.org/indexes/qqq), and the Composite erased its 2026 gains [10][9].

Nvidia dropped 10.7%, a fall compounded by the scandal that ended [Jensen Huang](https://theinference.org/markets/people/jensen-huang)'s run as chief executive [9][2], and markets price a December rate rise above 95%, which lifts the very hurdle rate the bear math rests on [9].

## The dates that will settle it

Oracle reports December 10, Amazon guides 2027 capital spending in early February, and Third Point's next 13F lands in mid-February [1]. None of that changes the direction: the buildout will not stop because the payback runs late, and the cost of the wait will keep moving to the balance sheets least able to carry it.

The larger reading is an opportunity. A $2-3 trillion annual revenue gap is a demand signal, the largest in software history [1]. The capability is bought and paid for; the revenue still has to be built, and whoever builds it earns the payback everyone else is waiting for.

## Takeaway

Loeb's letter prices the AI buildout at $2-3 trillion of new revenue every year, and his fund's 13F shows the trade that follows: keep the spenders whose cash flow can carry the wait, and sell the one whose cannot.

## Sources

1. [Yahoo Finance, Billionaire Dan Loeb's "Bear Math": AI Needs $2-3 Trillion of New Yearly Revenue. He Sold Amazon, 20 November 2026](https://finance.yahoo.com/technology/ai/articles/billionaire-dan-loeb-bear-math-215908050.html)
2. [CNBC, Chips, Big Tech and more: What hedge funds bought and sold in the third quarter, 16 November 2026](https://www.cnbc.com/2026/11/16/chips-big-tech-and-more-what-hedge-funds-bought-and-sold-in-the-third-quarter.html)
3. [The Inference, Hyperscaler reinvestment burden index, as of 30 September 2026](https://theinference.org/indexes/hyperscaler-reinvestment-burden)
4. [Microsoft, quarterly figures from its SEC filings, retrieved 20 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000789019)
5. [Amazon, quarterly figures from its SEC filings, retrieved 20 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001018724)
6. [Oracle, quarterly figures from its SEC filings, retrieved 20 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001341439)
7. [Meta, quarterly figures from its SEC filings, retrieved 20 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001326801)
8. [Yahoo Finance, Inside Dan Loeb's Portfolio: Every Move He Made in the Third Quarter, 19 November 2026](https://finance.yahoo.com/markets/stocks/articles/inside-dan-loeb-portfolio-every-154521181.html)
9. [Yahoo Finance, From Nvidia To Amazon, Tech Giants Face Stunning Reversal As Nasdaq Erases 2026 Gains, 20 November 2026](https://finance.yahoo.com/markets/stocks/articles/nvidia-amazon-tech-giants-face-213447664.html)
10. [The Inference, Nasdaq 100 (QQQ) index, 20 November 2026](https://theinference.org/indexes/qqq)
11. [Oracle, quarterly figures from its SEC filings, retrieved 21 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001341439)
12. [Meta, quarterly figures from its SEC filings, retrieved 21 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001326801)
13. [Microsoft, quarterly figures from its SEC filings, retrieved 21 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000789019)
14. [Amazon, quarterly figures from its SEC filings, retrieved 21 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001018724)


---

Dan Loeb sold Amazon and set AI's payback at $2-3 trillion of new revenue a year — The Inference. Canonical: https://theinference.org/article/dan-loeb-sold-amazon-and-set-ai-s-payback-at-2-3-trillion-of-new-revenue-a-year
