Interview, Fireside Chat
OpenAI CFO Sarah Friar: IPO, AI Rivalries, New Device, and Spending $100B+ on Compute
- Fundraising Milestone: OpenAI completed a $122 billion fundraising round in March, described by CFO Sarah Fryer as the largest private fundraising round in history, significantly exceeding the ~$30 billion Asadi-Ramco IPO.
- IPO Strategy: Fryer stated that an IPO is a funding mechanism and "milestone" rather than a destination, emphasizing that timing relative to competitors (like Anthropic's confidential S1 filing or SpaceX) is secondary to building sustainable, durable companies.
- Market Positioning: OpenAI aims to avoid binary classifications of being solely consumer or enterprise-focused, targeting a 50-50 revenue split between the two sectors while maintaining a "single foundation" model with multiple interfaces.
- User Scale & Growth:
- ChatGPT reaches over 900 million weekly active users.
- Codex (coding interface) grew from near zero in January to over 5 million users in a short period.
- Fastest-growing adoption regions are Africa; fastest-growing languages are Azerbaijani and Kazakh.
- User engagement tiers show a steep usage curve: Free users average 7 questions/day, paid tier users 15/day, and Pro users 11x that volume.
- Compute Economics: Fryer reiterated a metric where one gigawatt of compute capacity is roughly equivalent to $10 billion in annual revenue for OpenAI.
- Infrastructure Investment:
- OpenAI is constructing a 1-gigawatt data center in Saline, Michigan, in partnership with Oracle, expected to be operational by late 2027/early 2028.
- The Michigan project involves 2,500 union jobs and projected $1 billion in tax revenue, with a $45 million investment in local Codex education credits.
- Fryer projects a severe global compute deficit continuing through 2026 and 2027, with supply constraints remaining tight into 2028.
- Capital Allocation & Strategy:
- The company utilizes a "Rubik's Cube" strategy of maximum optionality, diversifying across multiple Cloud Service Providers (CSPs) including Microsoft Azure, Oracle, CoreWeave, AWS, and GCP to shift CapEx to OpEx.
- Chip diversification includes NVIDIA (priority), AMD, Cerebrus, and a custom silicon project with Broadcom.
- Future capital needs are being modeled for 2030–2032, anticipating that while per-gigawatt infrastructure costs may rise, model efficiency gains will drive down the cost per token for customers.
- Product & Market Trends:
- OpenAI is shifting toward "agentic" AI and multimodality, viewing video and voice (Sora, earpiece hardware) as critical future interfaces rather than just text.
- A new consumer hardware substrate is scheduled for unveiling by the end of the year, described as a "paradigm shift" in device interaction.
- Enterprise customers value "intuition" and memory within models (contextual memory of user preferences and company data) over raw data processing, particularly in sectors like life sciences (Thermo Fisher) and finance.
- Monetization & Advertising:
- Fryer expressed interest in a potent advertising platform leveraging user "intent" (from search-like queries) and "memory" (contextual awareness) to surpass current ad models.
- Commitments include keeping an ad-free tier available for users who prefer it and ensuring sponsored content does not bias model outputs.
- While optimizing for revenue might favor API tokens over consumer tokens, the long-term strategy prioritizes mass market accessibility to establish AI as a global utility.
- Competitive Landscape: Fryer noted that while Anthropic has gained traction with developers and enterprise, OpenAI's strategy relies on compounding advantages from a single model serving diverse interfaces, creating network effects through data and personalization.