Tommy Shaughnessy, founding partner at Delphi Ventures and co-founder of Delphi Digital, joins host Laura Shin to discuss the AI business model crisis and its implications for upcoming mega IPOs including SpaceX and OpenAI.

The conversation explores how crypto markets have lost over $600 billion in 30 days due to capital outflows to AI deals, while examining the fundamental sustainability issues in AI company revenue models.

Shaughnessy presents his viral thesis on how enterprise cost concerns and open source alternatives could disrupt the AI flywheel, while also discussing how crypto could potentially solve some of these structural problems through decentralized funding and censorship-resistant systems.

The AI Subscription Subsidy Problem

AI companies heavily subsidize consumer subscriptions - "$200 a month gets you about $8,000 worth of API spend" with subscriptions subsidized by about 40x versus API pricing.

Enterprises transitioning from subsidized subscriptions to API models are hitting spending walls, with companies like Uber and Microsoft cutting their AI usage due to cost concerns.

Open source inference providers like Base10, OpenRouter, Venice, and Nuce API offer "1/10 to 1/100 the cost" while delivering 80-90% of the intelligence quality.

IPO Timing and Market Dynamics

SpaceX IPO expected to perform well initially with consensus valuation "above a $2 trillion value, $2 to $2.5 closing market cap" versus $1.75 trillion going-out price.

AI IPOs will benefit from low floats "sub-10% when historical IPOs have been 10 to 20%" and massive passive capital from index funds seeking exposure.

Public disclosure requirements will expose previously hidden metrics: "what are your payback periods? What are the margins? What is the growth?" creating new scrutiny for AI companies.

OpenAI considering "drastic price cuts" in competition with Anthropic, likely to "eat into their growth and margins" before Anthropic's expected IPO.

Chinese Models and Competitive Threats

Chinese models like "Deepseek, KIMI, Minimax, GLM" are currently open source and "incredible" but may eventually go closed source to capture revenue.

Signs of potential strategy shifts include Alibaba releasing a model open source then "the next day the main dude was fired" and GLM keeping models private before open sourcing.

If Chinese models go closed source without superior quality, it would help US frontier labs by eliminating free substitutes and allowing them to "charge whatever the hell they want."

Censorship and Big Brother Concerns

AI safety restrictions create "Big Brother fearmongering" where labs "dictate what we're able to ask and why" - a concerning parallel to authoritarian control reminiscent of 1984.

Recent viral example showed Fable flagging an innocent question about "mitochondria and whether it's the powerhouse of the cell" due to security features.

Open source AI provides "the ultimate solution to the safety issues" through community auditing, similar to how "Linux is the most secure operating system and beats Windows."

Crypto as Solution for AI Problems

Crypto could address AI's tragedy of commons through "capital formation" via ICOs for major open source AI companies, enabling community ownership of models.

Hermes Agent demonstrates open source success with "3 to 4x the usage of OpenClaw" despite being built by decentralized community rather than centralized lab.

Projects like Venice AI provide "open source access to these models at scale with privacy" while Grass uses "3 million user nodes around the world to get real data for AI labs."

Decentralized training remains challenging despite Google's DLOCO release, with projects like "Prime Intellect, NUSE, BitTensor" showing it's "too cost intensive and too hard."