Flux | Desert Mirage
It’s 2012, and I'm stepping off a plane in Phuket with four of my college friends. The air hits you like a warm, wet blanket. We're headed to a place we booked through Groupon, and I'm not kidding, the deal was so insane it felt like a total scam.
Flights from Hong Kong.
Four nights in a three-bedroom pool villa.
Access to a couple of beat-up mopeds.
All for something like $200 or $300 total.
We felt like we were getting away with something, all of this for less than a semester’s worth of textbooks.
Groupon was the golden ticket, the name you associated with a value so good it didn't feel real.
Why didn’t it feel real…because it wasn’t.
Groupon was killer for customers, terrible for businesses.
Groupon’s Stock Chart
And now with Claude Code, I can’t help but thinking history may be repeating
The all-you-can-eat coding tool offers developers access to insane computing power for a relatively low price of $200 a month.
It’s an unbelievable deal, right?
For developers, yes.
For Anthropic, tbd.
Menlo Ventures "2025 Mid-Year LLM Market Update," revealed Anthropic has officially surpassed OpenAI in enterprise LLM usage, capturing 32% of the market.
The engine behind this phenomenal growth?
Code generation.
On the surface, this is a massive victory.
Anthropic identified the killer app, executed flawlessly winning developer mindshare.
But as I read the data, a nagging question emerges, is Anthropic becoming this generation's Groupon?
The Rundown Overview
Groupon's rise was meteoric. It attracted an enormous user base by offering an irresistible deal: 50% off at local businesses.
The usage was off the charts.
The problem?
The model was often unsustainable for the businesses footing the bill.
They were flooded with low-margin, one-time customers, and the economics eventually crumbled.
Subsidized Performance
The Menlo report highlights a key trend: "AI Spend Is Moving from Training to Inference."
Startups now report 74% of their workloads are inference-based.
This is the AI equivalent of "redeeming the coupon."
Every time a developer uses Claude to write, debug, or refactor code, Anthropic bears a significant computational (inference) cost.
While the report notes that "Enterprises Switch Models for Performance, Not Price," this holds true in a market where all major players are engaged in a pricing war and are likely subsidizing access to gain share.
Developers are flocking to Claude because it offers superior performance at a price that does not reflect its true cost.
High Cost Concentration
Anthropic's dominance is overwhelmingly built on code generation.
This is arguably one of the most resource-intensive use cases for an LLM, requiring complex reasoning, long contexts, and iterative interactions.
This is where the Groupon lesson becomes critical.
Groupon's model struggled because it attracted customers who were only interested in the deal, not in becoming loyal, full-price patrons.
Similarly, Anthropic is attracting a massive base of developers who are, by definition, the heaviest and most demanding users of their service.
The critical question the report doesn't answer is: What is the ROI on these high-cost users?
For the business model to be viable long-term, Anthropic needs high-paying, low-effort customers to subsidize these power users.
It's not clear who those customers are yet.
The Road Ahead
Anthropic's achievement is undeniable.
They've built a product that developers clearly love. But phenomenal usage built on unsustainable economics is a house of cards. Just like the local restaurants that partnered with Groupon, Anthropic might be winning a battle for usage while risking the war for profitability and survivability.
As I see it, their future hinges on navigating one of three paths:
The Hard Pivot: They are forced to raise prices to reflect true costs. This would likely mean ceding the developer market to cheaper competitors and the growing ranks of capable open-source models. They would have to find a new, high-margin niche, effectively abandoning the user base that got them to the top.
The Technical Breakthrough: The bull case. Anthropic's researchers achieve a 10x or even 100x improvement in inference efficiency. This is the holy grail that would make their current business model sustainable. It's a bet on radical, non-linear innovation that could pay off spectacularly, but it is far from guaranteed.
The Enterprise SaaS Play: They move up the stack. Instead of just selling the raw model (the "commodity"), they bundle it into high-margin enterprise applications like "Claude for Work" or "Claude for Legal." The high inference cost gets absorbed into a much larger, value-based subscription fee ($50 per seat per month, for example). This is the classic enterprise playbook for monetizing powerful but costly technology.
Until then, its position as market leader, while impressive, feels precarious.
So, what do you think?
Are we watching a repeat of the Groupon playbook in the AI era?