What Crypto's Funding Cycle Predicts About AI
The word "token" anchors both crypto and AI because is the same financial function dressed up in a new investment cycle. Crypto tokens converted narrative into a priceable, tradable unit. An AI's LLM tokens convert narrative into a metered commodity, sold below cost, whose usage numbers stand in for value the way "total value locked" once did. Both are instruments that make speculative capital flows look like revenue.
I co-founded CryptoKitties in 2017 (the first mainstream non-fungible token, or NFT, and the origin of the ERC-721 standard) and I was on the founding team at Dapper Labs when it waas created in 2018 (now boasting a $60B+ valuation). I've seen firsthand:
- how capability narratives get calibrated to fundraising needs,
- how circular capital manufactures the appearance of demand, and
- how insiders read the difference between a real business and a well-funded story.
The AI funding cycle is re-running crypto's structural playbook at two orders of magnitude larger scale.
We're going to map the overlap of crypto, AI, and venture capital
Over the next 6-9 months, I plan to I map crypto's financial mechanics onto AI's, mechanic by mechanic, to give the public a working framework for reading this industry's claims:
Circular capital as revenue. Chip makers invest in AI labs that spend the money on their chips; cloud providers invest in labs that route the funding back as compute commitments; equity-for-purchase deals stack across the industry. Crypto's precedent was protocols investing in one another with their own tokens and exchanges collateralized by tokens they issued. The mechanic is identical: recursive capital that makes demand look organic. Crypto showed how it unwinds.
Subsidized usage as growth proof. Liquidity mining and token incentives bootstrapped crypto's supply side; below-cost inference pricing bootstraps AI's demand side. Both defer the organic-unit-economics question indefinitely, and both render "usage" unfalsifiable as evidence of value.
Capability claims as collateral. When the capital structure is recursive, hype inflated capability narratives are load-bearing financial instruments, and the next funding round is priced against them. Crypto's whitepapers and roadmaps served the identical function. I intend to show the public how the sausage gets made, and how to spot claims calibrated to a raise rather than to reality.
Unpriced inputs. Crypto externalized its costs onto retail participants; AI externalizes them onto creators. Training data taken without licensing is a massive unpriced input now converting into settlements and contingent liabilities across the industry. An honest AI income statement prices its inputs; the sector's economics currently work only because it has not had to.
Why these questions are worth answering
The venture capital dynamics, the circular deal-making, the pricing strategies, and the economics of AI companies spending at therir current scale defies conventional business logic. That spending only defies business logic if you have never seen the logic before. I've experienced it firsthan. Crypto's 2017–2022 cycle is an example to learn from, particularly the part where the recursion unwinds and the comparison becomes predictive. This will tell us which claims to discount, which numbers are manufactured, and where the losses land when subsidy ends. Historically, they land on the public.
Alignment with CHT's framing
CHT's core method is connecting day-to-day harms to the systemic incentives driving them. This project names the incentive machine itself: a capital structure that rewards narrative over capability, punishes candor, and prices safety as drag. It also serves CHT's stated interest in what would shift incentives from the inside — because the crypto cycle offers concrete evidence about which interventions (disclosure norms, pricing transparency, input accounting) actually change insider behavior and which are theater.
What's next for this project
In roughly 6-9 months, my research will be complete and shareable in the form of:
- A comprehensive report: The Token Economy Ran Twice — one chapter per mechanic (crypto precedent → AI analog → what it predicts), written for a general audience.
- Op-eds: I intend to produce a public literary piece called "How to Read an AI Deal Announcement"; a piece timed to the next major circular-financing announcement; and a piece examining the unpriced-inputs argument aimed at a business outlet.
- Podcast appearance(s)
I've set a stretch goal for this project that makes its data and thesis more visible, engaging, and educational.
Stretch goal: The AI Balance Sheet
What it is: An interactive tool that surfaces the true cost of AI.
How it works: The user picks a frontier AI company. The tool surfaces four estimated figures side by side (each sourced and ranged rather than asserted as precise):
- Estimated compute cost to train the model, drawn from published third-party compute-cost research
- Estimated cost of the training corpus if licensed at the rates established by recent public data-licensing deals and settlements, extrapolated transparently
- Capital raised to date and what share is circular: vendor financing, compute-for-equity, and similar structures already disclosed in reporting
- A labor-exposure indicator for the sector most disrupted by that model's primary use case, sourced from published labor-economics research
Why I want to create it: This data is readily available, but difficult to digest at scale, and even more difficult to understand when it comes to "invisible" or diffused costs. This stretch goal makes the data and its thesis accessible and educational. It also provides a recurring point of engagement for anything attempting to advocate or understand the true cost of AI (provided data sources are evergreen).
