Ep. 17: Coatue: Value The AI Labs Before Training Costs. Also, The Compute Still Gets Paid. (2026)

On 7 October 2026, Coatue Management published its October 2026 Public Markets Update: a thirty-eight-slide argument about where the market is going, and one slide in it that set off the week. That slide gives public investors a framework for valuing the frontier AI labs.

The framework values them before training costs. Take a lab's earnings, remove what it spends to train its models, and what is left is the number to price. The same document says the buildout needs roughly two point two trillion dollars of compute equity, and that it will take new forms of financing to raise it. One deck, published on one day, holding both claims.

Coatue is a technology investment firm of about sixty-seven billion dollars, founded in 1999 by Philippe Laffont, and it runs two books at once: public technology stocks, and private companies before they list. It was also, that week, one of the named backers of an AI data-centre float in Australia.

This episode keeps three things apart. The statements are public and timestamped. The portfolio facts are on the record and cited. The incentive reading — what a valuation framework like this does for a firm holding both books — is analysis, and it is not a claim about what anyone intended.

Statement Timeline

All times show the channel's display timezone (UTC+02:00) first, followed by UTC. Engagement figures are platform snapshots captured on 14 October 2026 at about 10:05, not counters from the moment of posting.

6 October 2026: Reuters, TechCrunch and the Wall Street Journal report that Coatue and Blackstone are leading a round of up to four billion dollars into Lambda, an Nvidia-backed AI cloud company, at a fourteen-and-a-half-billion-dollar pre-money valuation, ahead of a planned IPO. 1
7 October 2026: the deck is published on Coatue's own site, presented as Philippe Laffont's latest views on market volatility, shifting leadership and the transition to the intelligence age, and on the infrastructure investment and financing needed to support AI. 2
7 October 2026, 23:01 (21:01 UTC): Coatue posts the deck to X: "Charts of the day: We are moving from 'waves' to 'ages.'" Captured: 23,804 views, 116 likes, 110 bookmarks, 14 reposts. A second post linking the update carried 158,362 views. 3 4
7 October 2026 (Wednesday): investors begin pulling orders from the Firmus IPO, days after the bookbuild opened on Tuesday, after CDC Data Centres chief executive Greg Boorer says on a podcast that a joint plan to develop 1.6 gigawatts of AI factories is no longer under way. 5
8 October 2026, 07:02 (05:02 UTC): the first teardown. A long-short investor writes that the deck is "either a desperate attempt to save the AI bags through the lab IPOs, or hiring dan ives and cathy wood would actually improve their investment judgment," and lists the capex, Meta and lab-valuation claims he says it makes. Captured: 60,382 views, 298 likes, 420 bookmarks, 19 reposts. 6
8 October 2026, 17:31 (15:31 UTC): the slide on compute financing circulates on its own; that post of it is the most-shared piece of the deck of the week. Captured: 238,601 views, 999 likes, 1,157 bookmarks. 7
8 October 2026, 20:07 (18:07 UTC): Coatue posts the update to its feed with a summary line: the shift toward an intelligence age, the infrastructure powering AI, and more. Captured: 9,127 views, 30 likes. 8
8 October 2026, 20:37 (18:37 UTC): a slide-by-slide critique of all thirty-eight pages lands. It notes that the frontier-lab framework "removes training expenditure that may remain economically necessary," that the roughly two-point-two-trillion-dollar compute-equity figure is a modelled capital requirement rather than an identified shortfall, and that the deck also "describes Coatue's expansion across investment products and asset classes, making its industry analysis relevant to its own commercial positioning." Captured: 4,444 views. 9
9 October 2026, 02:12 (00:12 UTC): the valuation slide gets its own name — "EBTITDA," earnings before training — with the objection that turning training off would flood the market with already-paid-for compute. Captured: 15,191 views, 81 likes. 10
9 October 2026 (Friday): Firmus shelves the five-billion-dollar float on the ASX. It had asked eleven Australian dollars a share, an equity valuation of about thirty billion dollars — nearly triple its valuation from an August round — against roughly thirty billion of debt and a prospectus promising five billion dollars of annual earnings within five years. It has two data centres today, both leased, in Melbourne and Singapore. It will raise privately instead, and consider other markets later. 5 11
10 October 2026, 18:22 (16:22 UTC): the deck's own summary line is quoted back at it — "Winners outperform losers over time" — with the comment "Brilliant analysis from Coatue." Captured: 145,905 views, 2,137 likes, 287 bookmarks, 55 reposts. 12
10 October 2026, 20:49 (18:49 UTC): the same line is posted again by a market commentator, with the deck's exhibits attached. Captured: 37,937 views, 283 likes, 52 bookmarks. 13
12 October 2026, 20:00 (18:00 UTC): the Australian Financial Review reports that Coatue, already a major Firmus backer, had been expected to commit up to one billion US dollars to the float and allocated less than a hundred million instead — a key factor in its collapse. The report rests on three people involved in the discussions who spoke anonymously, and Coatue has not commented publicly. The AFR's Sydney edition carries the story dated 13 October. 14
Same document, not two positions over time: both claims in this episode's title come from the single October deck published on 7 October 2026 — the framework that values the frontier labs before training costs, and the finding that the compute buildout needs trillions of new equity and new forms of financing. Both claims appeared in a single statement, not as positions that evolved over time.

What the episode decodes

The two halves of the deck do different jobs. The valuation framework is what makes a private AI position legible to a public investor: strip the training spend and a frontier lab stops looking like a research project that consumes capital. It starts looking like a software company, which is a business public markets already know how to price.

The financing half explains where the money keeps coming from. Coatue is a crossover investor, so it marks a private book against a public one, and a listing is the event that turns a private mark into a distribution. Anthropic has been reported to be preparing one of the largest listings ever, with OpenAI expected behind it. A framework that makes the labs look cheaper to own is not a neutral piece of research for a firm that holds them.

The deck's own numbers are more interesting than its slogans. It shows an AI winners basket up 405 per cent from late 2022 to 6 October 2026, against 2 per cent for the losers — and it also shows the related long/short series down 46 per cent peak to trough in 49 days. Those are thematic baskets and factor series, not fund returns, but they say something the cover line does not: the right idea and the wrong trade are two different things.

The market's answer arrived inside seventy-two hours. Firmus asked for a thirty-billion-dollar valuation on two leased data centres and a five-year promise, and the buyers walked. Nothing about that refutes the deck's optimism. It does show which audience the deck was written for, and what that audience was willing to pay.

Where the record stops

Coatue's private marks, its stake sizes and its LP communications are not public, and the AFR's account of how much it committed to Firmus rests on unnamed sources with no response from the firm. The decision to value frontier labs before training spend is a choice of framework, not a disclosed accounting standard, and the two-point-two-trillion-dollar figure is a model, not a bill. Whether the labs' economics hold up on that basis is the argument itself, and this episode does not settle it. The incentive reading here is a reading of position and timing, not documented coordination.

What founders and allocators can take from it

For a founder: read a published framework for what it needs, not for what it says. If the buyers of your next round need public AI valuations to hold, your price is riding on their story as much as on your numbers. If you sell compute, watch the financing channel rather than the demand headline — demand was not what failed the Firmus float.

For an allocator: when a firm publishes a valuation framework, ask which of its own assets the framework flatters, then watch where its money moves in the same week. Here that was leading a four-billion-dollar round into an AI cloud company on 6 October, and stepping back from a five-billion-dollar AI infrastructure float by 9 October. The tell is not the words. It is the flow.

References

1
Reuters: Nvidia-backed Lambda targets $4 billion raise ahead of planned IPO
reuters.com
2
Coatue: Public Markets Update Deck and Replay, October 2026
coatue.com
3
Coatue on X: charts of the day, the October deck
x.com
4
Coatue on X: public markets update
x.com
5
Reuters: Nvidia-backed Firmus scraps $5 billion Australia IPO
reuters.com
6
masayoshisson on X: a teardown of the Coatue deck
x.com
7
kevg1412 on X: Coatue on the new forms of financing for compute
x.com
8
Coatue on X: the October 2026 update is out
x.com
9
TheValueist on X: investment implications, assumptions and portfolio risks of the Coatue October deck
x.com
10
Restructuring__ on X: on EBTITDA and the pre-training valuation
x.com
11
Bloomberg: data center darling's $30 billion IPO dream crushed in 48 hours
bloomberg.com
12
UnstructuredCap on X: winners outperform losers over time
x.com
13
PeterJ_Walker on X: the same Coatue line
x.com
14
Australian Financial Review: Firmus flop fuelled by San Francisco investment giant's cold feet
afr.com

Statement Timeline

6 October 2026: Reuters, TechCrunch and the Wall Street Journal report that Coatue and Blackstone are leading a round of up to four billion dollars into Lambda, an Nvidia-backed AI cloud company, at a fourteen-and-a-half-billion-dollar pre-money valuation, ahead of a planned IPO. 1

7 October 2026: the deck is published on Coatue's own site, presented as Philippe Laffont's latest views on market volatility, shifting leadership and the transition to the intelligence age, and on the infrastructure investment and financing needed to support AI. 2

7 October 2026, 23:01 (21:01 UTC): Coatue posts the deck to X: "Charts of the day: We are moving from 'waves' to 'ages.'" Captured: 23,804 views, 116 likes, 110 bookmarks, 14 reposts. A second post linking the update carried 158,362 views. 3 4

7 October 2026 (Wednesday): investors begin pulling orders from the Firmus IPO, days after the bookbuild opened on Tuesday, after CDC Data Centres chief executive Greg Boorer says on a podcast that a joint plan to develop 1.6 gigawatts of AI factories is no longer under way. 5

8 October 2026, 07:02 (05:02 UTC): the first teardown. A long-short investor writes that the deck is "either a desperate attempt to save the AI bags through the lab IPOs, or hiring dan ives and cathy wood would actually improve their investment judgment," and lists the capex, Meta and lab-valuation claims he says it makes. Captured: 60,382 views, 298 likes, 420 bookmarks, 19 reposts. 6

8 October 2026, 17:31 (15:31 UTC): the slide on compute financing circulates on its own; that post of it is the most-shared piece of the deck of the week. Captured: 238,601 views, 999 likes, 1,157 bookmarks. 7

8 October 2026, 20:07 (18:07 UTC): Coatue posts the update to its feed with a summary line: the shift toward an intelligence age, the infrastructure powering AI, and more. Captured: 9,127 views, 30 likes. 8

8 October 2026, 20:37 (18:37 UTC): a slide-by-slide critique of all thirty-eight pages lands. It notes that the frontier-lab framework "removes training expenditure that may remain economically necessary," that the roughly two-point-two-trillion-dollar compute-equity figure is a modelled capital requirement rather than an identified shortfall, and that the deck also "describes Coatue's expansion across investment products and asset classes, making its industry analysis relevant to its own commercial positioning." Captured: 4,444 views. 9

9 October 2026, 02:12 (00:12 UTC): the valuation slide gets its own name — "EBTITDA," earnings before training — with the objection that turning training off would flood the market with already-paid-for compute. Captured: 15,191 views, 81 likes. 10

9 October 2026 (Friday): Firmus shelves the five-billion-dollar float on the ASX. It had asked eleven Australian dollars a share, an equity valuation of about thirty billion dollars — nearly triple its valuation from an August round — against roughly thirty billion of debt and a prospectus promising five billion dollars of annual earnings within five years. It has two data centres today, both leased, in Melbourne and Singapore. It will raise privately instead, and consider other markets later. 5 11

10 October 2026, 18:22 (16:22 UTC): the deck's own summary line is quoted back at it — "Winners outperform losers over time" — with the comment "Brilliant analysis from Coatue." Captured: 145,905 views, 2,137 likes, 287 bookmarks, 55 reposts. 12

10 October 2026, 20:49 (18:49 UTC): the same line is posted again by a market commentator, with the deck's exhibits attached. Captured: 37,937 views, 283 likes, 52 bookmarks. 13

12 October 2026, 20:00 (18:00 UTC): the Australian Financial Review reports that Coatue, already a major Firmus backer, had been expected to commit up to one billion US dollars to the float and allocated less than a hundred million instead — a key factor in its collapse. The report rests on three people involved in the discussions who spoke anonymously, and Coatue has not commented publicly. The AFR's Sydney edition carries the story dated 13 October. 14

Same document, not two positions over time: both claims in this episode's title come from the single October deck published on 7 October 2026 — the framework that values the frontier labs before training costs, and the finding that the compute buildout needs trillions of new equity and new forms of financing. Both claims appeared in a single statement, not as positions that evolved over time.

What the episode decodes

Where the record stops

What founders and allocators can take from it

Ep. 17: Coatue: Value The AI Labs Before Training Costs. Also, The Compute Still Gets Paid.

Statement Timeline

What the episode decodes

Where the record stops

What founders and allocators can take from it

References

Related content

Ep. 17: Coatue: Value The AI Labs Before Training Costs. Also, The Compute Still Gets Paid.

Statement Timeline

What the episode decodes

Where the record stops

What founders and allocators can take from it

References

Related content