Jon Chu

guys I just cancelled my Claude plan I don’t know what happened

Big day for Ollama! When we started, open models and the open source AI ecosystem were in their early days with few believers. Our belief in open source has never wavered. With today's fundraising announcement and our 9M+ active builders, we’re ready to scale open models into AI that you can own. All aboard open models! 🧵

VCs: There’s no AI talent outside of SF Anthropic: we need a 16 story building full of people in NYC

You know none of these AI companies actually do FDE right? It's just solutions eng in the old traditional enterprise sense. Nobody knows what Palantir does and nobody knows what FDE is. Shocking

I can’t believe I need to spell this out but people KEEP getting confused so here it is. There’s a few types of “ARR” floating around.. First, most* ARR is reported based on the last months revenue - multiplied by 12. This is called annualized run rate. That’s been standard for a while - even though it has its problems. Types of ARR: 1) Traditional ARR: Annual “recurring” revenue. There’s an actual subscription and you typically pay per seat. Your subscription can be monthly or annual. Annual is typically better since it traditionally has lower churn even though the “per month” cost is lower. Most of modern SaaS historically is sold this way (think Figma, Notion, Workday, Salesforce etc). Gross margins are typically high. Companies trade at 6-20x revenue multiples in the public markets and 20-100x in the private markets. The fact that it’s recurring makes it “sticky” - this is important since it’s different from the bottom two. 2) AI ARR: In the AI world, there has started to be a lot of usage based revenue in addition to seat based pricing. For usage, token usage with the large models or inference costs if you’re hosting your own. Now these fluctuate based on usage and usage one week may not be usage next week. But now the total spend per month is seat based + usage. Now this is where it gets interesting. You can now technically calculate your ARR as seat + highest usage for a given month (or dare I say day) and now annualized run rate is very inflated. Moreover margins on these businesses are very different since gross margins on usage are typically very low since most of that revenue is made by the model or inference companies. Proof is in the pudding here - but most investors are buying the hype and not digging in. Founders are running a tight process to avoid this diligence as well. I have been called a boomer more than once when I have asked 😅 3) GMV or non recurring revenue disguised as “ARR”: This is a new trend where companies who are making non-recurring revenue with very low gross margins are calling themselves ARR. They are being cheeky by calling it annualized revenue run rate by taking the last months revenue and multiplying by 12. But, the main difference is the gross margin profile is very different for these companies since for GMV most of your revenue is passed through in a marketplace. And those businesses should be valued VERY differently compared to traditional SaaS businesses. And it’s non recurring even though the customer is buying today - there’s typically no guarantee they buy tomorrow. Now, in a gold rush everyone is buying more and more so all looks gravy. I have nothing against these businesses - but let’s call a spade a spade is my only ask. Long post, but people really need to understand. Hope this helps clear up any questions! * I say most because I have seen companies repot annual recurring revenue based on a single days revenue multiplied by 365. Not the norm, but definitely happens in case 2.