According to the article itself, ARR’s definition is clear and strict: annual and recurring. This means that “margin” doesn’t factor in, so I don’t think calling these AI companies’ revenue GMV is all that sensible either.
I do agree the market is eager to conflate run rate and arr, though
> Technology costs do decrease. However, demand always migrates to the newest, most powerful models.
> People do not:
> Choose GPT-3.5 because it is cheap.
> Select a lower-tier Claude model on purpose.
People definitely can, and do, both things for many workloads. Coding is an obvious exception, but oftentimes a cheaper model is good enough. And businesses that had to spend $5/MTok on a frontier model last year can probably get similar performance and spend $0.50/MTok or less today.
Beyond the other comments denouncing it as AI slop, it’s fundamentally wrong. GMV aka marketplace thru-revenue is strictly defined by accounting standards and most tax offices (including the IRS and ATO). It’s not something you just decide to have. One of the core principles of it is that the consumer is aware that the revenue will flow through (eg, if buying something on Etsy/Amazon, that is obvious while others like Uber are more contentious). Using a different model provider is unlikely to pass the test.
This would’ve been a fascinating article if written by an accounting expert but is unfortunately just slop.
This article is AI slop. There’s a valid point in there somewhere. ARR is heavily misused - YC emeritus Michael Seibel harps on this often. There’s also a lot of nonsense in this article though, due to the AI slop.
According to the article itself, ARR’s definition is clear and strict: annual and recurring. This means that “margin” doesn’t factor in, so I don’t think calling these AI companies’ revenue GMV is all that sensible either.
I do agree the market is eager to conflate run rate and arr, though
They called “GMV” - “gross merchandise value” which isn’t inherently “marginal” but still is fairly nonsensical.
AI generated content. Shame they couldn't bother to have a human spend time on this, it's a fair point.
It's also full of head scratchers like this:
> Technology costs do decrease. However, demand always migrates to the newest, most powerful models.
> People do not:
> Choose GPT-3.5 because it is cheap.
> Select a lower-tier Claude model on purpose.
People definitely can, and do, both things for many workloads. Coding is an obvious exception, but oftentimes a cheaper model is good enough. And businesses that had to spend $5/MTok on a frontier model last year can probably get similar performance and spend $0.50/MTok or less today.
Beyond the other comments denouncing it as AI slop, it’s fundamentally wrong. GMV aka marketplace thru-revenue is strictly defined by accounting standards and most tax offices (including the IRS and ATO). It’s not something you just decide to have. One of the core principles of it is that the consumer is aware that the revenue will flow through (eg, if buying something on Etsy/Amazon, that is obvious while others like Uber are more contentious). Using a different model provider is unlikely to pass the test.
This would’ve been a fascinating article if written by an accounting expert but is unfortunately just slop.
This article is AI slop. There’s a valid point in there somewhere. ARR is heavily misused - YC emeritus Michael Seibel harps on this often. There’s also a lot of nonsense in this article though, due to the AI slop.