Jun 2, 2026 · 32m · allin
OpenAI CFO Sarah Friar: IPO, AI Rivalries, New Device, and Spending $100B+ on Compute
Sarah Friar · 22m spoken Chamath Palihapitiya · 2m spoken Jason Calacanis · 1m spoken David Friedberg · 1m spoken David Sacks · 50s spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In a main stage interview at the Liquidity Summit 2026 hosted by the All-In Podcast, OpenAI CFO Sarah Friar breaks down the company's strategic roadmap, addressing IPO market timing, massive compute capital allocation, enterprise monetization, and upcoming hardware collaborations.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 20.8% of the talking time here. How this is scored →
The hosts as informed peer 6.9 Guest teaching 6.6 Guest disagreement 3.1 The hosts pushing back 5.4
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 3:13 Rejecting Anthropic Superiority Premise
Friar flatly rejects Calacanis's assertion that Anthropic blew past OpenAI, arguing filing an S-1 means nothing and explaining OpenAI's distinct multi-interface model.
Hardest push from the hosts ▶ 3:21 Calacanis Challenges OpenAI's Market LeadershipCalacanis aggressively frames OpenAI as having lost its lead to Anthropic across enterprise, developer, and revenue metrics, demanding to know how they allowed it to happen.
Biggest teaching moment ▶ 30:05 Search Metric Mechanics and Ad Intent AdvantageFriar educates the hosts on how search volume is miscounted, explaining that a single multi-turn ChatGPT conversation matches dozens of Google page refreshes while combining memory with explicit user intent.
The host holds their own ▶ 7:42 Chamath Quotes Friar's Gigawatt Economic RuleChamath demonstrates deep domain recall by holding Friar to her own 18-month-old rule of thumb that 1 gigawatt translates to $10 billion in annual revenue.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| OpenAI IPO Strategy and Market Timing | 5 | 5 | 3 | 4 | Sacks references recent Wall Street Journal reporting on AI IPO timing and compares OpenAI to SpaceX and Anthropic. Friar contextualizes the $120B raise against historic market benchmarks like Saudi Aramco's $30B IPO, framing public listing as a milestone rather than a destination. | |
| Anthropic Rivalry and Multi-Interface Strategy | 6 | 6 | 6 | 7 | Calacanis pushes aggressively with breaking news of Anthropic's S-1 filing, asking if OpenAI was passed in enterprise adoption and revenue. Friar firmly rejects the premise, pointing to ChatGPT's 900 million weekly active users and OpenAI's distinct multi-interface distribution strategy. | |
| Enterprise vs. Consumer Focus and Tiered Monetization | 8 | 7 | 5 | 6 | Calacanis questions whether OpenAI lost focus by pursuing hardware devices and Sora, which Friar dismisses as a false binary before detailing enterprise revenue balance. Chamath demonstrates high expertise by recalling Friar's 18-month-old rule of thumb equating one gigawatt to $10 billion in annual revenue. | |
| Compute Supply Outlook and Multimodal Shift | 6 | 6 | 2 | 3 | Chamath shares personal frustration with rate limits on rival platforms, allowing Friar to detail token availability bottlenecks. She outlines the timeline for severe compute scarcity lasting through 2026 and 2027 while explaining the strategic shift toward multimodal interaction. | |
| Unveiling the Jony Ive Hardware Device | 8 | 7 | 2 | 7 | Calacanis and Friedberg probe into the unreleased Jony Ive hardware device before Friedberg challenges Friar on OpenAI's return on capital model. Friar breaks down unit economics, highlighting a 97% cost deprecation curve alongside long-horizon compute commitments extending into 2032. | |
| Compute Infrastructure Strategy: CSPs, Chips, and Financial Structure | 8 | 7 | 2 | 6 | Sacks challenges Friar with granular estimates of $50B per gigawatt all-in costs to ask how far their $120B raise extends. Friar explains OpenAI's Rubik's Cube financial architecture, leveraging multiple CSPs to shift CapEx to OpEx alongside multi-chip partnerships. | |
| Convergence of the Tech Stack and Enterprise Intuition | 7 | 7 | 2 | 4 | Chamath asks whether vertical integration across chips, cloud, and models will merge the tech stack. Friar explains why staying closest to the end-user preserves profit pools and describes how combining memory with enterprise intuition prevents model commoditization. | |
| Monetization Philosophy and Advertising Strategy | 7 | 8 | 3 | 6 | Calacanis confronts Friar on monetization, referencing Anthropic's Super Bowl ad trolling OpenAI over potential advertisements. Friar clarifies search metric nuances, explaining that a single 50-turn conversation is miscounted against individual Google refreshes, while outlining the value of combining intent with user memory. |
Statements from this episode (20)
Friar: Markets reward long-term durability over being first to IPO
“In the end, the market is a weighing machine, not a popularity machine. No one remembers who went first, Google or Yahoo, Lyft or Uber”
Friar: OpenAI Codex reached 5 million users from near zero in January
“Codex just hit five million over the weekend. We're really proud of that. Coming from almost zero in January.”
Friar: Scaling AI models lowers token costs and raises gross margins
“As we, as models get bigger, there's more efficiency that should lower the overall cost to give you a token in the world. That should compound to higher gross margins, ultimately more ways to pay for compute, and then access to compute is one of the really big…”
Friar: OpenAI's revenue is split roughly 50/50 between consumer and enterprise
“Our revenue is getting pretty balanced about 50, 50.”
Friar: OpenAI free users average seven questions daily while Pro users do 11x
“Our free users do about seven turns, seven questions a day. Our first paid tier do double that, about 15. Our real paid tier with the plus, 20 bucks, hopefully you're all on it, or higher, about three X. And pro, about 11 X over a free user.”
Friar: OpenAI will still lack sufficient compute capacity in 2026
“Thank God we did, because in 26, we still won't have enough compute.”
OpenAI breaks ground on a 1-gigawatt data center in Michigan
“Sam right now is in Saline, Michigan. He's gonna be cutting the ribbon in about two hours, so you are getting a sneak preview, but they told me it was okay to say it in the room. That will be, you know, sticking shovels in the ground on a one gigawatt data cen…”
Friar: Would probably not hire finance professionals who cannot use Codex
“It would be like I would never hire a finance person. Didn't know how to use Excel and it pretty much probably wouldn't hire a finance person today. It doesn't know how to use a tool like codex.”
Friar: Additional compute capacity is virtually impossible to buy in 2026
“But the landscape right now, in 26, if you want to buy more compute, good luck to you. Like, tell me, because I don't know where else to find it.”
Friar: AI training mostly occurs in US for national security reasons
“There's training that mostly still all happens here in the United States. For USG reasons, for making sure that a national asset in effect is happening on US soil.”
Sarah Friar: OpenAI will unveil new hardware in 2026 and sell in 2027
“We're changing into a consumer substrate that I cannot tell you what it is, but by the end of this year, we will unveil it. Early next year we'll be able to buy it.”
Friar: OpenAI won't get compute from Michigan data center until late 2027 or 2028
“Like that Michigan data center in Saline, I don't think we will be getting compute out of it until probably end of 27, early 28.”
Friar: Per-gigawatt compute costs are rising due to power and memory expenses
“So we assume right now that computed actually on a per gigawatt is getting more expensive because power is getting more expensive. Memory is getting more expensive and so on.”
Friar: Relying on one chip vendor prevents staying on AI's frontier
“I think if you're only on one chip, there's just inherently a moment where you can't be on the frontier because there's some leapfrogging that happens.”
Friar: OpenAI's next major fall training run will use Nvidia Vera Rubin
“So today, Nvidia remains our absolute priority partner. They have the frontier chip. Our next big trading run in the fall will be done on Vera Rubens.”
Friar: OpenAI uses Cerebras, pipeline AMD chips, and custom Broadcom silicon
“But we also now have chips in the pipeline from AMD. Cerebrus is already online. It's been an incredible low latency chip. Great for devs, for example, that want real time coding. And there's our own chip that we're working on with Broadcom.”
Friar: LLMs are not commoditizing because value lies in agentic layers
“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, the memory, right?”
Friar implies OpenAI will introduce ads, commits to ad-free tier
“We'll always provide A free, a tier, sorry, an ad free tier for people that just don't want ads.”