The Supply and Demand of AI Tokens | Dylan Patel Interview
Implementation costs are collapsing while ideas remain abundant. The critical skill now isn't execution—it's choosing which ideas justify spending tokens on implementation. Focus on identifying high-value problems where AI can create measurable returns, then deploy the best models available. Your co
45mKey Takeaway
Implementation costs are collapsing while ideas remain abundant. The critical skill now isn't execution—it's choosing which ideas justify spending tokens on implementation. Focus on identifying high-value problems where AI can create measurable returns, then deploy the best models available. Your competitive edge depends on access to frontier models and strategic allocation of token budgets, not coding ability.
Episode Overview
Dylan Patel discusses how his firm's AI spending has skyrocketed from negligible to $7 million annually (25% of salary costs), demonstrating the dramatic shift from execution-constrained to idea-constrained business models. He shares examples of non-technical employees using Claude to accomplish in weeks what previously required teams of hundreds working for years, and explores the implications for token demand, model access inequality, and the coming expansion into robotics.
Key Insights
The Great Inversion: Ideas Are Now Cheap, Execution Is Easy
Historically, execution was extremely difficult and ideas were cheap. AI has completely inverted this relationship. Implementation is now trivially easy but expensive in token costs. The new bottleneck is identifying which ideas justify the capital spend on tokens, since you can implement almost anything but shouldn't implement everything.
Token Access Will Determine Winners and Losers
As models improve, access to frontier AI becomes increasingly restricted to enterprise customers who can pay premium prices. Companies with deeper pockets and better relationships with AI labs will get earlier access to more powerful models, creating a competitive moat based on token access rather than technical capability. This could lead to dramatic concentration of economic power.
Anthropic's Margin Expansion Reveals Insatiable Demand
Anthropic's gross margins jumped from ~30% to 72%+ as revenue grew from $9B to $40B+ ARR—despite not significantly expanding compute capacity. This demonstrates that demand far exceeds supply at current prices, suggesting they could double prices and customers would continue paying. The constraint is compute capacity, not willingness to pay.
Model Release Cadence Is Accelerating Dramatically
The time between major model releases has compressed from 6 months to 2 months because AI makes implementation so easy that labs can test more ideas faster. Anthropic went from L4 engineer capability to L6 engineer capability in just two months with Mythos, representing one of the biggest capability jumps in two years.
Robotics Will Create the Next Token Demand Wave
Current 'software-only singularity' is just a blip. Once AI makes software development trivial, attention shifts to the physical world—where most economic value exists. Breakthroughs in few-shot learning for robotics (6-18 months away) will create entirely new categories of token demand as robots become useful for specific tasks.
Notable Quotes
"What used to matter a lot was execution was very very difficult and ideas were cheap. Now ideas are cheap and plentiful but execution is very easy. So really only the good ideas are the ones that can justify the spend on super cheap implementation."
"Last year we thought we were heavy users of AI. Everyone's using chat GPT. Everyone's using cloud. This year the spend has just skyrocketed. We signed an enterprise contract with Anthropic and it's gone to the point where now, I think when I last talked to you it was 5 million spend rate. It's actually 7 million spend right now."
"We're north of 25% of spend on cloud code as a percentage of salary. And if this trajectory continues, then you know, we'll spend more than 100% by the end of the year."
"He said that was an entire team's job to build that and maintain that now. Rack that up across you know the entire firm it's insane. Malcolm who's an economist at a major bank before, their economist department was like 100 or 200 people. This would have taken the team of 200 economist a year. He's just like completely cracked out on claude. He's like everything has changed."
"If I don't adopt AI, someone else will and they will beat me. It's a bit of an existential like if I don't move faster, someone else will and they will commoditize me."
Action Items
-
1
Secure Enterprise AI Contracts with Per-Token Pricing
Stop using consumer subscription tiers with rate limits. Negotiate enterprise contracts with AI providers (Anthropic, OpenAI) that charge per token consumed rather than fixed subscriptions. This eliminates rate limiting bottlenecks and ensures access to the most capable models as they're released.
-
2
Ruthlessly Prioritize Ideas by Economic Value Per Token
Since implementation is now easy but expensive in tokens, create a framework for evaluating which ideas will generate the highest ROI per token spent. Not every idea should be implemented—focus token budgets on projects with clear, measurable business value that exceeds token costs.
-
3
Build AI Skills Across Your Entire Team, Not Just Engineers
The examples of non-technical people (economists, analysts) accomplishing massive projects show that coding ability is no longer the constraint. Train everyone in your organization to use AI tools effectively, focusing on prompt engineering and understanding what's possible rather than traditional programming.
-
4
Track Token Spend as a Strategic Business Metric
Treat token consumption like you'd treat any other critical resource. Monitor spend rates, identify which use cases generate the most value, and optimize allocation. Understanding your token economics will become as important as understanding your customer acquisition costs.