Video Summary
Peter Steinberger spent $1.3M on tokens in 30 days, showing real-world token scale for experiments.
Prices for lower-end models may fall, but tokens needed for meaningful work (e.g., GPT 5.5) are rising sharply.
Open Router found GPT 5.5 costs 49–92% more than 5.4; some model changes increase effective token length.
Top firms (OpenAI, Anthropic) and well‑funded insiders are locking access, concentrating AI power and jobs.
Learners and indie developers face an advantage gap: those who can pay for tokens iterate far faster.
Not for the models that matter. While lower‑end model prices may stay flat or fall, advanced models (e.g., GPT 5.5) have seen token costs jump—Open Router reported 49–92% higher costs from 5.4 to 5.5 and some tokenizers produce 35% longer outputs, raising effective expense.
His dashboard showed $250k spent in seven days and $1.3M in 30 days, demonstrating that meaningful experimentation at scale can incur very high token bills even in research scenarios.
The video argues that's unlikely without structural change: pay‑to‑iterate advantages mean those with token budgets can reach product‑market fit faster, while many learners lack funds to experiment at that pace.
The piece warns that major players like OpenAI and Anthropic, who control top models and privileged access, may capture most high‑value AI roles and gatekeep advanced capabilities.
"Many believe that tokens are getting cheaper, but this is one of the biggest deceptions in the industry."
Peter Steinberger's recent expenditures reveal a staggering $1.3 million spent on tokens in just 30 days, raising questions about the affordability of AI.
The common assumption that the costs for AI tokens are decreasing is misleading. While prices for lower-end models may appear stable or lower, the tokens necessary for meaningful work are actually rising sharply.
Specifically, the price of tokens from GPT 5.4 to 5.5 has doubled, indicating that the cost of using more advanced models is becoming increasingly prohibitive.
Research from Open Router highlights that the cost difference between versions 5.5 and 5.4 ranges from 49% to 92% higher, challenging the notion that AI will become more accessible in the future.
"In the future, only companies like Anthropic and OpenAI will be the main employers."
Industry leaders such as Andre Karpathy feel the pressure of exclusivity in the AI domain, having recently joined Anthropic to gain access to vital models.
There are growing fears that access to AI tools is becoming reserved for a select few, potentially leaving the broader workforce behind.
Eric Schmidt's recent commencement speech underlines the urgency surrounding AI, but the pushback from the audience suggests a disconnect between the industry's optimism and public sentiment.
"If you're an 18-year-old kid now, what do you learn?"
The emphasis on AI raises critical questions for new learners who may not have extensive financial resources to experiment with token-based systems.
The advantage for those with access to tokens is that they can swiftly iterate towards product-market fit, allowing faster recovery from failures compared to those without such resources.
Companies are increasingly prioritizing speed in deployment, with the aim to fix bugs quickly rather than build more reliable systems from the start.
"AI is becoming a tool primarily for the rich, by the rich."
The stark reality is that AI technologies are unlikely to become cheaper or accessible to everyone, as economic incentives do not support reduced pricing.
While AI tools promise to enhance productivity, the social implications highlight a troubling trend where only the affluent can leverage these advances effectively.
The discussion reveals a consensus that unless changes occur in how AI resources are distributed, inequality will persist and might even widen in the tech landscape.
