← Back to issue
All-In Podcast

Thomas Laffont: The $4T AI IPO Wave Is Coming… and We've Never Seen Anything Like It

32m · Transcribed via assemblyai · Watch on YouTube

Thomas Laffont (co-founder, **Coatue — ~$55B AUM**, raising another **$1B for AI**) delivers his annual All-In deck, and it is the **supply-side spine of this week's IPO-wave thread**. The throughline: the private cohort is no longer a balance-of-payments problem — it's about to flood the public market with paper. Laffont's new index, the *'Magnificent Ain't'* (SpaceX, Stripe, Anthropic, Databricks, Revolut, ByteDance, Anduril), already *'represents almost $4 trillion of value and has really crushed the traditional kind of mag 7.'* The unlock is liquidity: **SpaceX in 'the next few weeks', Anthropic's confidential S1 'just today', OpenAI publicly committed** — and *'if you add up the totality of just those three companies… it's basically going to be more than the 10 years combined.'* That reframes the ecosystem from cash-consuming to cash-returning. The growth is the justification: **OpenAI + Anthropic ARR is already 'bigger than Google Cloud and Azure', and Laffont models it 'bigger than AWS and potentially bigger than all of Microsoft by 2028.'** On valuation he refuses the bubble label — *'these are not fake companies'*, **profitable** (Anthropic had a *'profitable month'*), trading at *'the lowest multiple of earnings of the S&P 500'* — but concedes the **power-law / K-shape** risk: no new centicorn in years would be *'a warning sign.'* The most counterintuitive datapoint: centicorns ($100B+) have a **31% chance of a 10x**, versus ~8% for unicorns — value concentrating, *'the cost of not being in a winner are higher than ever.'* He sizes the **AI revenue ecosystem at ~$140B today → ~$300B this year → double in 2027** (consumer subs + ~$150B of AI-enabled ads + enterprise/Claude-Code/Codex), the cleanest answer this cohort has given to *'where's the ROI?'*. Two tradeable tells: he echoes the **memory-shortage** thesis (*'no TSMC for memory'* → ASIC-vs-DRAM multiple gap, *'memory per user could quintuple'*), and floats a future **OpenAI-vs-Anthropic price war** once these balance-sheets weaponise. Verdict on the commoditisation bear case: *'pretty thoroughly disproven.'* The public market, he says, is *'the great test equalizer.'*

Key points

Notable quotes

It represents almost $4 trillion of value and has really crushed the traditional kind of mag 7. Almost every single one of these names has outperformed that index.

Thomas Laffont · 0:14

We know Anthropic, just today the headlines hit that they've submitted confidentially for their S1. If you add up the totality of just those three companies, you can see that it's basically going to be more than the 10 years combined

Thomas Laffont · 0:14

Now even bigger than Google Cloud and Azure

Thomas Laffont · 0:14

by the end of the year it could be bigger than AWS and potentially bigger than all of Microsoft by 2028

Thomas Laffont · 0:14

if you're a center corn, 100 billion or more the odds. And by the way, we're putting in public and private companies, you now have a 31% chance of having had a 10x.

Thomas Laffont · 0:14

The winners are compounding faster than ever, which means the cost of not being in a winner are higher than ever.

Thomas Laffont · 0:14

We believe that it's about 140 billion today. It'll be about 300 billion this year and it'll double in 2027.

Thomas Laffont · 0:14

So the one pushback I would have just on the valuation argument is these are not fake companies.

Thomas Laffont · 22:15

you're making venture investments in trillion dollar companies and giving them 50 times revenue, 100 times revenue evaluation

All-In host · 21:18

I do think that the narrative of oh, these models are commodities and these companies are going to get. I think that's been pretty thoroughly disproven now. Right.

Thomas Laffont · 24:29

Could we see a price war between OpenAI and Anthropic as a question? Right. If these companies have so much capital, is one of them ever going to pull a price lever to try and compete with the other rationally?

Thomas Laffont · 31:09

if I want to design a chip like OpenAI, I can go to TSMC. And I know it's hard but at least I have TSMC to help me if I want to make memory. Well, there is no tsmc, right? So what should the memory multiples be versus ASIC chips as an example?

Thomas Laffont · 26:47

Themes

Mentioned

People

Companies

Ideas