AI's Affordability Crisis: OpenAI and Anthropic Burn $8–$14 to Make $1
David Cahn's September 2023 'AI’s $200B Question' flagged the gap between AI revenue and infrastructure spend. Nine months later, the gap tripled to $600B. Ed Zitron and DSHR now confirm the scale: AI platforms are burning $8 to $14 for every dollar of revenue, sustaining the drug-dealer pricing model until IPOs force real prices.
Token Subsidies: 40–70x
SemiAnalysis ran tests maxing out OpenAI and Anthropic subscription rate limits with long-horizon coding tasks. Results: a $200/month Anthropic subscription lets you burn $8,000 in tokens; OpenAI's $200 plan burns $14,000. That's a 40x and 70x subsidy, respectively. Under plausible gross margin assumptions (4x token price over cost), using just 25% of the rate limit yields negative 25% gross margin.
OpenAI's 2025 Financials
Zitron published OpenAI's 2025 financials: revenue $13.07B, costs and expenses $34B (including $41.55B non-recurring from conversion to for-profit, net loss $38.5B). Strikingly, $5.73B went to sales and marketing—44% of revenue. Despite this, business adoption has been flat. At year-end, OpenAI had $50B in assets, half in cash.
What This Means for Developers
As AI companies approach IPOs (SpaceX, Anthropic, OpenAI), they must raise prices to generate returns. Early moves: Microsoft shifted GitHub Copilot to token-based billing and tightened rate limits (April 2026 report). If you rely on subsidized tokens, expect cost increases of 40–70x once subsidies end. Budget accordingly, and monitor platform pricing changes.
📖 Read the full source: HN AI Agents
👀 See Also
PrismML's Bonsai 1-bit Qwen models tested: 107 t/s generation on 8GB VRAM
Bonsai models from PrismML are 1-bit quantized versions of Qwen3 8B, 4B, and 1.7B that achieve 107 tokens/second generation and >1114 t/s prompt processing on an RTX 4060 with 8GB VRAM, with significantly reduced memory requirements.
Claude Code v2.1.90 adds /powerup command with gamified feature discovery
Claude Code v2.1.90 introduces a /powerup slash command that provides gamified onboarding with 10 unlockable power-ups, each teaching one feature most users miss. The system includes animated demos in the terminal and detailed documentation with screenshots.
Claude MAX Plan Now Includes 1M Token Context Window at No Extra Cost
The Claude MAX plan has been automatically upgraded to include a 1 million token context window without additional API-based usage charges, with users reporting significantly reduced token usage and elimination of context window management overhead.
Anthropic Paying SpaceX $15B/Year for Compute Through 2029
SpaceX IPO filing reveals Anthropic pays $1.25B/month through May 2029 for compute. The deal supports AI training on Colossus 1 & 2 facilities.