An Apocaloptimist
Tokenomics revised.
Are AI tokens deliberately underpriced?
When you look at the full stack: multi-billion dollar training runs, energy-hungry data centers, specialized chips, safety teams, infrastructure build-outs – the current cost per token feels surprisingly low. Not because the system is cheap, but because pricing today is shaped less by marginal cost and more by strategy.
We are in a land-grab phase. Providers are compressing prices to expand usage, attract developers, lock in ecosystems, build habits around their APIs. Growth first, margin later. It’s an old playbook.
The marginal cost of inference has also genuinely fallen fast. Better hardware, quantization, routing between large and small models. So while the full system is expensive, the cost of producing one more token has been dropping in parallel. That creates an interesting tension between what things cost to build and what they cost to run.
Over the next decade, I think two things happen simultaneously: commodity inference gets cheaper — many tasks migrate to smaller specialized models, local inference improves, everyday reasoning becomes nearly free. At the same time, frontier cognitive capability gets priced around value rather than compute: agent hours, workflow automation, productivity replacement. In that world, tokens stop being the real unit of economics.
We’ve seen this structure before where cloud storage got cheaper per gigabyte, yet total cloud spending exploded. The cheap layer expanded the market. The premium layer captured the margin. AI likely follows the same shape of abundant cheap inference at the base, premium reasoning at the top, with vertical AI systems capturing the largest share of value somewhere in between.
So yes, tokens will probably be cheaper in ten years at the commodity level. Advanced AI being economically cheap is a different question entirely.
The productivity surplus AI creates will be captured somewhere: by the model providers, by the application layer, or by the people and organizations that learn to use it well before everyone else does.
History suggests the middle layer, the one that translates raw infrastructure into specific value, tends to capture the most. That’s probably where most of you reading this are operating.
I suspect we’ll look back at this period as the phase where providers were subsidizing adoption to define the future rails of cognition.
What are you building on those rails?
MZ
The AI-Panic Cycle—And What’s Actually Different Now (44 min)
Excellent interview in the Atlantic on The AI-Panic Cycle with Anil Dash.
“A huge part of the cultural tension around these things is everybody advocating them is like why wouldn’t you love this and everybody whose industry is being….”
The AI Agent Economy Is Here (23 min)
The AI Agent Economy Is — Y Combinator
“For one thing, Claude Code has totally taken over my life.”
These AI Prompts Exposed My Biggest Blind Spots (20 min)
These AI Prompts Exposed My — Daniel Pink
“Most people use AI to write emails or summarize articles.”
How to Build an AI Business That Makes $4,000 a Week (35 min)
How to Build an AI — Peter Yang
“I want you to create a product that you can build entirely on your own that will make money.”
My Multi-Agent Team with OpenClaw (14 min)
This closely matches my experience
“Well, sitting on my desk is a new Mac Mini that I set up just for the purpose of running my team of AI agents using OpenClaw.”
How a visually impaired engineer builds personal software with Claude Code + Wispr Flow (49 min)
A very cool episode of How I AI.
“Right before I started college, I ended up losing most of my central vision due to a rare genetic disorder called Liber’s hereditary optic neuropathy.”
HOW I BECAME AN APOCALOPTIMIST (2min)
THE AI DOC: OR HOW — Focus Features
“If this technology goes wrong, it can go quite wrong.”
Envisioning Research
Apple TV recently announced a Neuromancer series, which feels like a good excuse to share Wintermute — our research hub on AI systems, autonomous agents, and synthetic cognition, named after one of the AIs in the book. Gibson imagined most of this in 1984. We’re now tracking it as emerging infrastructure. Some things worth exploring inside: wafer-scale AI systems, edge neuromorphic processors, and photonic accelerators. Share with anyone who’s read the book — or should.
If Artificial Insights makes sense to you, please help us out by:
📧 Subscribing to the weekly newsletter on Substack.
💬 Joining our WhatsApp group.
📥 Following the weekly newsletter on LinkedIn.
🦄 Sharing the newsletter on your socials.
Artificial Insights is written by Michell Zappa, CEO and founder of Envisioning, a technology research institute.