Unlike internet's one-dimensional, flat-fee growth, AI grows across two exponentials, user penetration and tokens per user, allowing AI labs' economics to work
The next ARR growth engine after coding, the AI infra bubble, the Economics of open vs. closed source model Why is the AI semi buildout longer than the internet buildout?
@fi56622380Fin
Context & Ripple Effects
AI-lab economics have long been constrained by the compute intensity and support burden identified in a 2020 analysis of AI businesses' scaling challenge. By 2024, OpenAI was pursuing a larger recurring-revenue base while ChatGPT Enterprise had more than 300 paying customers, as covered in its business-model push.
The key economic tension sharpened in 2025: token prices fell even as newer reasoning models raised developers' total token consumption and costs. The Aug. 29 analysis frames that usage growth as the mechanism that can support AI-lab revenue rather than treating AI as a conventional flat-fee software market.
First-order effects
AI labs gain a monetization model tied to both adoption and workload intensity, making revenue growth more sensitive to tokens consumed than to seat counts alone.
Developers deploying reasoning models face a countervailing cost pressure: cheaper tokens do not necessarily reduce bills when each task requires materially more of them, as reported in the 2025 account of rising developer costs.
Second-order effects
Developers and enterprise buyers have stronger incentives to measure token use by task and choose models on total workload cost rather than headline price per token.
AI labs must balance usage-led revenue against inference expense; higher consumption improves demand signals while making compute efficiency central to gross-margin discipline.
Third-order effects
If token consumption rises alongside user adoption, AI software economics increasingly resemble metered infrastructure rather than flat-fee SaaS, shifting competitive advantage toward labs that can sustain inference margins at scale.
The open-versus-closed model debate becomes more economically consequential: the viable model is not simply the cheapest per token, but the one that can pair useful workloads with durable unit economics.
The trend: AI monetization is moving toward consumption-led growth, where user penetration and rising inference per user jointly determine the durability of lab economics.
Very good read. Really helped me made more sense of this AI boom amid the amount of noise, both overly optimistic and overly negative floating around the internet and tv media. Very excited to see the next article soon.
A lot of good nuggets in this article. As I've said, when you understand the computational complexity of this workload it shows just how grossly short of compute we really are.
The most important part of this piece for $NVDA is buried in the middle. Codex already generates 64% of combined Codex and ChatGPT output tokens among OpenAI's enterprise customers. Since February, weekly active Codex users have grown 108x in legal, 41x in sales and recruiting,
This is truly a must-read. Please read it, and once you have, bookmark it and revisit it often. Written by one of the smartest people I know, it offers powerful insights into the future ahead. All the more essential if you're an investor.
every author talking about the ai build out loves to feel smart to say eventually there will be a bubble. as @TMTLongShort says, not enough people are emotionally prepared for if this is not a bubble.
Excellent post. AI demand continues to grow exponentially and there are no signs that companies have run out of useful things to do with AI. But enterprise adoption still has visible bottlenecks, most of which are organizational and human, including: • people need time to learn
One of the most important things to know about this AI cycle is that no one knows what they're talking and even very smart and plugged in people are continually reversing their opinions and then re-reversing them three months later.
The average strongly held belief these days in AI has a half life of 6 months at best. Here are just a few the industry has cycled through and probably has no consensus on at the moment: * OSS is too far behind to catch up * The labs can't be profitable at scale * All