If you like this post, please share it with one friend and help us grow our audience.
What are the frontier labs doing? Every other week, one of the big 6 frontier LLMs leap-frogs over last week’s winner. They ship a bigger, faster, smoother model that is designed to get deeper into the enterprise stack.
It’s an amazing race to witness as an outsider and a potentially terrible business to run as an executive. Every leap costs billions of dollars that ultimately get matched and surpassed.
Collectively, these big 6 frontier labs have invested hundreds of billions of dollars in model training runs, GPUs, data centers, and AI engineering over just the last few years. Zoom out further and you will see trillions of dollars worth of AI investments over this decade.
They are massive. They are impressive. They are everywhere. But what kind of business are they?
This week we hosted an event that pondered this question. I want to tell you about it, but let me get something out of the way first.
Provocation of the Week
The frontier labs are pricing themselves like ConEd.
This week’s provocation is inspired by an ON_Discourse Group Chat we hosted this afternoon. It featured a long-time agency executive and founder who recently published a provocative article about the business model of AI. It started with a direct question:
How can OpenAI deliver a return on the most expensive fundraise in human history?
The numbers are epic: this ten-year old company has raised $186B so far. In order to cover that hole, it would need a return of $1.2T. Can it do that at $20/month? $200?
To answer that question, our special guest asked another one first. What are these businesses for? At first, it feels logical to compare them to SAP, Oracle, or any other SaaS enterprise. But the business models do not align at all.
SAP and Oracle make money at scale. They sign ten-year enterprise contracts and lock in relationships forever. They are paid to keep the lights of the enterprise running.
The frontier labs use money to scale. And that scale is being sold to us as a superpower. Don’t take my word for it, this week’s Google I/O featured Demis Hassabis declaring “AGI is on the horizon” in between product demos for agentic shopping carts and a brand new superfluous video model.
Our guest sees this contradictory business model through the agency lens. Agencies have always lived in the gap between billing and outcomes: a flat fee for the campaign that generates a billion in revenue. The frontier labs are running the same mismatch in value but at a planetary scale.
Except they might be on the wrong side of it. Or, to put it another way…
The labs are spending like they are curing cancer and billing you like you plugged in a toaster.
What’s Next
If you have thoughts, feedback, or a perspective worth sharing, reach out: chmiel@ondiscourse.com. You might see your reaction in next week’s edition.
ON_Member Events
ON_Discourse Group Chat: Follow My Flow featuring Matt Alldian - Life at 400M Tokens
Tuesday, May 26, 11am-12pm EST
Zoom
This is your backstage pass into someone’s real AI workflow, a candid, step-by-step reveal of how their stack actually works in practice. This week Matt Alldian from Cardboard Toast will deliver a 6 month update from his first presentation. Since presenting in January, Alldian has seen an exponential expansion of his AI stack - nearly all of it self-generated. He went from 108 projects to over a 1000. From 83 code repos to 172. From 63M monthly tokens in January to over 400M in May.
He attributes this growth to a reusable set of patterns he has built up over time. Instead of building each new project from scratch, Alldian composes each new project out of these patterns.
What does that mean for him and his work? And can you replicate his system?
Join us and find out.
ON_Discourse Speakeasy: Monthly Cocktails for AI Leaders
Wednesday, May 27, 5pm-8pm EST
NYC
Join us for cocktails and excellent conversation with ON_Discourse members, including leaders, founders, builders, and operators.