← Back to issue
All-In Podcast

OpenAI CFO Sarah Friar: IPO, AI Rivalries, New Device, and Spending $100B+ on Compute

32m · Transcribed via assemblyai · Watch on YouTube

OpenAI CFO Sarah Friar on All-In — the primary source on the IPO-wave thread this cohort has been trading off second-hand. She confirms the **$122 billion March raise** (*'the largest IPO to date was the Aramco, which was about $30 billion'* — so this private round dwarfs it) and frames an IPO as *'a milestone. It is not a destination… just another way to fundraise.'* On the **OpenAI-vs-Anthropic race** — the week Anthropic confidentially filed its S1 — she refuses the 'third place' framing (*'it does not mean anything yet because you have to run now the gauntlet of the SEC'*) and reframes strategy as a single foundation with many interfaces: **over 900 million weekly ChatGPT users**, *'the noun and the verb'*, Codex *'just hit 5 million over the weekend'* from near-zero in January, revenue *'pretty balanced about 50 50'* consumer/enterprise. The load-bearing disclosure is **compute**: *'compute is a very scarce resource… there's just not enough tokens available'*, and *'in 26 we still won't have enough compute.'* She endorses Chamath's **one-gigawatt ≈ $10 billion of revenue** framing, confirms a **1GW Saline, Michigan data center** inside the Oracle complex (**$1bn in Michigan taxes**, 2,500 union jobs), and — pressed by Sacks on the *'about $50 billion'* all-in cost per gigawatt — lays out the capital-light playbook: ride **multiple CSPs (Oracle, CoreWeave, Microsoft, GCP, AWS)** to *'shift CapEx into OpEx'*, run **multi-chip (Nvidia Vera Rubin, AMD, Cerebras, own chip with Broadcom)*** to stay on the frontier, only now moving to built-to-suit (a SoftBank complex in Texas). Her job, repeated: *'maximum optionality… in a moment where I'm not yet an investment grade type of entity.'* The single most contrarian claim cuts straight at this cohort's central bear case: *'a year ago people talked about the commoditization of the LLMs, and frankly it's gone the opposite'* — because the **harness, memory and context** re-moat the model. On unit economics: the 4→5.4 jump deprecated cost *'something like 97%'* in two years; OpenAI raised 5.5 prices **2x** yet customers still get *'a break of about 20 to 30% cost reduction per token.'* And the future monetisation tell: at least **11% of the search market**, *'very high intent'*, with **memory + context next to intent** as a *'very potent ad platform'* — though *'if I was optimizing only for today, I would give every token to the API… order of magnitude more than to the consumer.'*

Key points

Notable quotes

In the end, an ipo, I say this to the team all the time. It's a milestone. It is not a destination. Do not run your company as if that's some sort of destination. It's just another way to fundraise. We just did, you heard me on the sizzle reel. Raise $122 billion in March, and that was to give ourselves maximum flexibility.

Sarah Friar · 1:12

I think the largest IPO to date was the Aramco, which was about $30 billion. So it is actually incredible that you're going to have potentially three IPOs at a scale that will be bigger even than 2001.

Sarah Friar · 1:51

It does not mean anything yet because you have to run now the gauntlet of the SEC and who knows how long that takes for anyone.

Sarah Friar · 3:13

Over 900 million people use ChatGPT weekly and it's become the noun and the verb. It's how most people experience AI for the first time.

Sarah Friar · 3:57

So first of all, yes, compute is a very scarce resource at the moment. What we see in our business, we're going up that kind of vertical wall of demand right now and there's just not enough tokens available.

Sarah Friar · 8:30

I think you said one gigawatt is roughly equivalent to about $10 billion a year of revenue to OpenAI.

Chamath Palihapitiya · 7:42

So the estimates I've seen is that to stand up 1 gigawatt of AI compute costs about $50 billion in land, power, shell chips, everything, all in around 50 billion. Do you have to front all of that money when you create a new data center?

David Sacks · 22:38

what CSPs do for us in effect is they shift CapEx into OpEx. So you pay as you get the revenue so as you're actually utilizing the data centers.

Sarah Friar · 23:09

a year ago, people talked about the commoditization of the LLMs, and frankly, it's gone the opposite, because as you start building an agentic layer, and we've all started to use this word harness, but the harness is what brings the context.

Sarah Friar · 26:38

I think the deprecation cost was something like 97%. It's like kind of an amazing curve actually. I'm sliding from 4 to 5, 4 it was 97% but that happened in like two years.

Sarah Friar · 16:32

if I was optimizing only for today, I would give every token to the API right every token to the API order of magnitude more than to the consumer.

Sarah Friar · 29:42

Themes

Mentioned

People

Companies

Ideas