A single AI coder burns 60 to 100 times more tokens than a casual ChatGPT user.
Anthropic won the AI race, OpenAI gives Codex for free, xAI Grok Build & Claude cancel culture | GMP EP13
One AI coder burns up to 100x more tokens than a casual ChatGPT user — that single "red dot" on the adoption chart is causing the entire compute shortage.
May 16, 2026
1:37:48
Difficulty: Intermediate
Played
God Mode Podcast
Anthropic won the AI race, OpenAI gives Codex for free, xAI Grok Build & Claude cancel culture | GMP EP13
One AI coder burns up to 100x more tokens than a casual ChatGPT user — that single "red dot" on the adoption chart is causing the entire compute shortage.
May 16, 2026
1:37:48
Difficulty: Intermediate
Played
Chapter
No indexed bits in this chapter.
Snapshots ()
Stats
Episode stats
Insight Overview
insights
chapters
Insight distribution
Sub-Categories
Speaker breakdown
Talk Time
Key Quotes ()
This episode
Cast
-
Discussed as the only player spanning all four phases of the AI super cycle, from silicon to robotics to space, including his role on Trump's China trade trip.
-
Indie hacker referenced as the extreme case of AI power usage (50M tokens/minute) and minimal infrastructure spend, with his low-cost stack held up as an aspirational model.
-
Trump's former AI task force adviser cited for publicly calling Anthropic's growth trajectory a path towards an AI monopoly.
-
NVIDIA's CEO mentioned as a passenger on Trump's flight to China alongside Elon Musk, following NVIDIA's record $5.5 trillion market cap milestone.
-
Central to the episode as the company that overtook OpenAI in business adoption and launched the new programmatic usage credit policy that angered third-party builders.
-
Discussed as Anthropic's main enterprise rival, responding to losing business adoption lead by offering free Codex access to new business sign-ups.
-
Elon Musk's AI company discussed for launching Grok Build, voice cloning, and connectors — seen as the fastest-moving but data-quality-constrained competitor in the AI space.
-
Discussed after launching an agent SDK, CLI, and webhooks to position itself as the enterprise AI knowledge layer, with 50% YoY revenue growth cited.
-
Discussed across multiple angles: its AI Book laptop launch, Gemma 4 lightweight model, investment in SpaceX, and talks to build data centres in space.
-
Discussed as the first company to reach a $5.5 trillion market cap and the dominant semiconductor player in Phase 1 of the AI super cycle.
-
Cloud network provider discussed as a cost-saving layer between Vercel and end users for image rendering, enabling Rik to stay on Vercel's free tier.
-
Mentioned as Elon Musk's vehicle for bringing AI into the physical world via the Optimus robot, and as a counter-example for how AI safety headlines are sensationalised.
-
AI video generation tool discussed as now integrated into Codex, raising the question of whether a standalone video editor or an embedded tool will dominate AI video creation.
-
Elon Musk's rocket company mentioned in the context of Google's talks to launch AI data centres in space and as a Google investment portfolio holding.
-
AI-powered IDE praised for its $18/month value, model-agnostic flexibility, and rumoured acquisition by xAI — positioned as the practical choice for vibe coders.
-
Anthropic's AI assistant and coding agent; discussed in the context of token limits, the Max plan cancellation, and new programmatic usage restrictions.
-
A floating voice-activated desktop AI agent tool, compared to a modern working version of Microsoft Clippy, with an open-source first version available on a waitlist.
-
xAI's new agentic CLI tool for coding and workflow automation, launched in early beta for Super Grok Heavy subscribers at approximately $600/month.
-
Open-source framework for programmatic video creation in React, used by Matt for AI-assisted video editing workflows within Cursor.
-
An AI voice chat tool described by Ben as the best voice model available, noted for its highly lifelike and empathetic conversational quality.
This episode
Claims & Sources
2 / 15 cited (13%)
Factual claims made this episode, and whether a source was named.
One user logged 865 million tokens in a single month, costing approximately $2,500 in API costs.
Anthropic overtook OpenAI in US business AI adoption for the first time, per Ramp data.
OpenAI offered Codex free for 30 days to new businesses in response to Anthropic's business adoption milestone.
Global token usage on OpenRouter grew from 7 trillion in January 2026 to 28 trillion tokens by May 2026.
Cursor's $18/month plan includes $180 worth of tokens, plus unlimited but slower Composer usage after the token allowance is exhausted.
Anthropic's new programmatic usage policy cut third-party API rate limits by approximately 40x for some businesses.
Converting a Markdown file to HTML for AI agent use consumed five times as many tokens.
Notion grew its revenue by approximately 50% year over year.
Peter Levels operates AI coding agents that consume approximately 50 million tokens per minute.
NVIDIA became the first company to reach a $5.5 trillion market capitalisation.
xAI's Grok 4.1 Fast was priced at less than one-tenth the cost of Grok 4.7.
xAI sold off part of its Colossus compute cluster to Anthropic because Grok models were not being used as much as expected.
Using Cloudflare as a caching and rendering layer between Vercel and end users cut Rik's Vercel CPU usage by 70%.
The Internet effectively closed down to AI scraping in the last two years, making it harder for new AI companies to access the quality training data that helped OpenAI and Anthropic reach GPT-4 level capability.
TL;DR
Rik, Ben, and guest Matt break down the week's sharpest AI moves: Anthropic overtook OpenAI in business adoption (per Ramp data), OpenAI fired back with 30 days of free Codex for new businesses, and xAI launched Grok Build for developers. Rik cancelled his $100 Claude Max plan live on air, splitting the budget across Cursor, Codex, Claude, and xAI. The key takeaway: a single AI coder burns 60–100x more tokens than a casual ChatGPT user, making that "one red dot" on the AI adoption chart far more impactful than headcount suggests.
#Anthropic vs OpenAI
#AI business adoption
#token consumption
#Cursor IDE
#Claude Max plan
#Grok Build
#AI super cycle
#HTML for AI agents
#Notion agent SDK
#voice cloning
#Peter Levels stack
#vibe coding
#agentic goals
#compute shortage
#enterprise AI lock-in
#Anthropic
#OpenAI
#Cursor
#Claude
#Codex
#xAI
#token economics
#AI adoption
#enterprise AI
#HTML vs Markdown
#Notion agents
#Peter Levels
#agentic AI
2 minute taster
One Red Dot is Melting the Internet
The viral 'AI adoption' chart misses the most important variable: token consumption. A single AI coder burns 60–100x more tokens than a casual ChatGPT user, meaning the one red dot representing 3.2 million people is actually equivalent to 50–100 dots in compute terms. That tiny slice of developers is the reason compute is scarce.
Rik Cancels His $100 Claude Plan Live on Air
Rik cancelled his $100/month Claude Max plan in real time, reallocating the spend to Cursor ($20), Codex ($20), Claude ($20), and xAI ($20). His logic: you don't need frontier models for most tasks, and diversifying across providers eliminates the frustration of hitting weekly token limits mid-week.
Anthropic's API Crackdown: Building on Someone Else's Subsidy
Anthropic's new programmatic usage policy slashed third-party API rate limits by up to 40x for some businesses — even those paying $40k/month. Businesses that built products on Claude's subsidised pricing are now forced to rethink their architecture. Rik's blunt verdict: you built a business on their generosity; don't be surprised the generosity has limits.
Look closer
Rik, Ben, and guest Matt discuss the week's biggest AI news: Anthropic beating OpenAI in business adoption per Ramp data, OpenAI's free Codex response, xAI's Grok Build launch, the Claude Max plan cancellation controversy, token economics across AI tools, HTML vs Markdown for AI workflows, Notion's agent platform opening, and the Trump-Elon-Jensen China trip.
Chapter list
- Vibe coding
- A style of AI-assisted software development where builders describe what they want in natural language and let AI write the code, requiring little to no manual coding knowledge.
- Tokens
- The basic units of text that AI language models process and generate; pricing and usage limits for AI APIs are measured in tokens rather than words or characters.
- Agentic AI
- AI systems that can autonomously plan and execute multi-step tasks without requiring a human to approve each action, often running continuously in the background.
- Ramp data
- Usage and spend analytics from Ramp, a corporate card and spend management platform, which tracks which AI tools US businesses are paying for.
- Claude Code
- Anthropic's terminal-based AI coding agent that can autonomously read, write, and execute code within a developer's project environment.
- Codex
- OpenAI's cloud-based agentic coding tool that can run tasks in sandboxed environments, allowing developers to queue multiple coding jobs simultaneously.
- PRD
- Product Requirements Document — a specification that outlines what a product or feature should do, often used by AI agents as a planning artefact before writing code.
- VPS
- Virtual Private Server — a rented server hosted in the cloud that gives developers root access to a Linux environment for running applications and agents.
- Remotion
- An open-source framework for programmatically creating videos using React and JavaScript, allowing developers to render video content through code.
- OpenRouter
- A unified API platform that routes AI model requests to multiple providers (OpenAI, Anthropic, xAI, etc.), tracking aggregate token usage across the ecosystem.
- Super Grok
- xAI's premium subscription tier for Grok, gating access to advanced features like Grok Build early beta and higher usage limits.
- Slash goals (/goals)
- A Claude feature that lets users set an overarching objective for an AI agent session, instructing it to continue working autonomously until the goal is achieved.
- Reinforcement learning
- A machine learning technique where an AI model learns by receiving rewards for correct outputs and penalties for incorrect ones, used to improve model quality over time.
- Colossus
- xAI's large-scale GPU cluster used to train and run Grok models, reportedly sold off in part to Anthropic due to lower-than-expected Grok model utilisation.
- Open source
- Software whose source code is publicly available for anyone to read, modify, and redistribute, often used as a go-to-market strategy before monetising premium versions.
- AGI (Artificial General Intelligence)
- A hypothetical AI system capable of performing any intellectual task a human can; in this episode used loosely to mean a decisive capability lead over all competitors.
- Subsidization
- In AI business context, when a provider prices its API or subscriptions below cost to attract users and developers, effectively subsidising growth with investor capital.
- Walled garden
- A closed ecosystem where a platform controls access to services and restricts interoperability with outside tools, used here to describe Anthropic's and Notion's data lock-in strategies.
- Supabase
- An open-source Firebase alternative providing a managed PostgreSQL database, authentication, and storage — popular with indie hackers and vibe coders.
- Hetzner
- A German cloud and dedicated server provider popular with indie hackers for its low cost and bare-metal VPS offerings compared to AWS or Google Cloud.
