OpenAI plans $2K–$20K/month agent tiers as Sam Altman pitches investors on massive ARR expansion
Mar 5, 2025
Key Points
- OpenAI is launching tiered agent subscriptions ranging from $2,000 to $20,000 monthly for knowledge work, coding, and research, with Sam Altman projecting agents will drive 20–25% of total revenue.
- SoftBank's $3 billion annual commitment to OpenAI agents ($250 million monthly) creates accountability through public disclosure but requires massive deployment scale to justify the spending.
- At $10,000–$20,000 monthly, OpenAI's direct agents must deliver 10–40 times the capability of cheaper alternatives like Devon AI to avoid becoming an expensive wrapper around its own API.
Summary
OpenAI is introducing enterprise agent pricing in three tiers: $2,000/month for knowledge workers, up to $10,000/month for software development agents, and up to $20,000/month for PhD-level research agents. Sam Altman told investors that agents will eventually account for 20–25% of OpenAI's total revenue. The company is already at $4 billion ARR and projecting $11 billion by year-end, according to CFO Kevin Roose.
These prices demand dramatically better outcomes than existing alternatives. Devon AI, built on OpenAI's models, costs $500/month and works effectively in organizations. At $10,000–$20,000/month, OpenAI's direct agents would need to be 10–40 times more capable to justify the premium. Otherwise they become an expensive wrapper around the same API.
SoftBank's $3 billion commitment creates real accountability. Investor Masayoshi Son has committed to spend $3 billion on OpenAI agents this year, roughly $250 million per month or $8 million per day. While Son's capital gives him leverage and genuine interest in AI scale, the growth trajectory from under $100 million in annual agent spending today to $3 billion is steep. OpenAI would need to deploy agents at massive scale and speed while realizing measurable ROI to consume that budget meaningfully.
Altman is using these revenue commitments to pitch new investors on aggressive valuations and growth multiples. SoftBank is using OpenAI's revenue projections to raise debt against. Because SoftBank is public, the spending will eventually appear in financial disclosures, creating accountability that many private venture bets lack.
Proprietary agent products carry real risks. Wrapping agents in a proprietary service risks letting competitors or partners distill, reverse-engineer, or replicate the underlying capability. Keeping the best models on the API while charging at consumer tiers means direct competition with Devon AI, Cursor, and others who can wrap the same models at different price points. This is the familiar software-as-a-service squeeze: distribution and brand matter more than raw capability when the underlying tech is commoditizing.
Altman's fundraising narrative has shifted noticeably. The "AGI is months away" framing that powered earlier raises is losing credibility after GPT-4.5 landed without the inflection-point breakthrough many expected. Agent revenue and large enterprise deals are now the story, a more prosaic but measurable path to expanding ARR. Whether $11 billion in revenue by year-end and $3 billion from SoftBank agents materialize will define whether this pivot is strategic insight or optimistic framing.