Venture capitalist Marc Andreessen has outlined a tiered pricing vision for artificial general intelligence (AGI), ranging from consumer-friendly subscriptions to ultra-premium access priced in the billions.
In a post on X, Andreessen proposed four indicative tiers: $20 per month, $200 per day (about $75,000 annually), $1,000 per day (roughly $350,000 annually), and exclusive, frontier-level AI systems around $10 billion.
High-performance autonomous agents, including OpenClaw, can cost between $300 and $1,000 per day to operate at scale, he said in another post.
The development follows a key policy shift by Anthropic, which on April 4 ended support for third-party tools like OpenClaw under its Claude subscription plans. Users must now pay separately via API or usage-based billing, it said.
Andreessen added that the move exposed the cost of running autonomous AI agents. “Frontier models become more expensive to serve. Token demand explodes at exponential rates. Solve for the equilibrium,” he wrote in a post on X.
Replying to the thread, tech billionaire Elon Musk also confirmed that costs for such systems are already reaching about $200 per day.
OpenClaw is an open-source framework that allows users to run AI agents capable of handling tasks such as email, bookings, web browsing, and coding.
However, while the software itself is free, it depends on paid large language models like Claude, OpenAI, and others for intelligence.
Also Read:Why are founders flocking to OpenClaw?
Anthropic said third-party tools like OpenClaw place an “outsized strain” on its systems due to their continuous, high-volume usage.
Hosting OpenClaw itself remains relatively inexpensive. Platforms such as Hostinger offer plans starting around Rs 549-Rs 799 per month (range $5-$9) for server infrastructure. These plans cover compute, storage, and setup, but not AI usage.
The bulk of the expense comes from model usage. Under the new pay-as-you-go system, users are billed per token, i.e., small units of text processed by the model, causing costs to scale rapidly as agents run in loops and perform multi-step tasks.
Also Read: Strategic blunder or genius?: Y Combinator’s Garry Tan reacts to Anthropic’s OpenClaw move
In a post on X, Andreessen proposed four indicative tiers: $20 per month, $200 per day (about $75,000 annually), $1,000 per day (roughly $350,000 annually), and exclusive, frontier-level AI systems around $10 billion.
High-performance autonomous agents, including OpenClaw, can cost between $300 and $1,000 per day to operate at scale, he said in another post.
The development follows a key policy shift by Anthropic, which on April 4 ended support for third-party tools like OpenClaw under its Claude subscription plans. Users must now pay separately via API or usage-based billing, it said.
Andreessen added that the move exposed the cost of running autonomous AI agents. “Frontier models become more expensive to serve. Token demand explodes at exponential rates. Solve for the equilibrium,” he wrote in a post on X.
Replying to the thread, tech billionaire Elon Musk also confirmed that costs for such systems are already reaching about $200 per day.
OpenClaw is an open-source framework that allows users to run AI agents capable of handling tasks such as email, bookings, web browsing, and coding.
However, while the software itself is free, it depends on paid large language models like Claude, OpenAI, and others for intelligence.
Also Read:Why are founders flocking to OpenClaw?
Anthropic said third-party tools like OpenClaw place an “outsized strain” on its systems due to their continuous, high-volume usage.
Hosting OpenClaw itself remains relatively inexpensive. Platforms such as Hostinger offer plans starting around Rs 549-Rs 799 per month (range $5-$9) for server infrastructure. These plans cover compute, storage, and setup, but not AI usage.
The bulk of the expense comes from model usage. Under the new pay-as-you-go system, users are billed per token, i.e., small units of text processed by the model, causing costs to scale rapidly as agents run in loops and perform multi-step tasks.
Also Read: Strategic blunder or genius?: Y Combinator’s Garry Tan reacts to Anthropic’s OpenClaw move