Cost

 

This is where we talk about how sustainable the current business model is.

Compare it to Amazon, Lyft, etc., esp. Amazon:

  • https://www.ben-evans.com/benedictevans/2014/9/4/why-amazon-has-no-profits-and-why-it-works
  • https://www.nytimes.com/2018/11/19/opinion/amazon-bezos-hq2.html (https://archive.ph/c63np)
  • https://www.reddit.com/r/Economics/comments/2fmota/why_amazon_has_no_profits_and_why_it_works/ (lots of good insights)
  • https://smartasset.com/investing/how-a-company-can-succeed-without-making-a-profit
  • https://a16z.com/why-amazon-has-no-profits-and-why-it-works/ (A16Z version)
  • https://news.ycombinator.com/item?id=8274677 (YC discussion)
  • https://business.expertjournals.com/ark:/16759/EJBM_202nonennman2014pp9-13.pdf
  • https://qz.com/1196256/it-took-amazon-amzn-14-years-to-make-as-much-net-profit-as-it-did-in-the-fourth-quarter-of-2017
  • Could be bullshit, but: https://www.youtube.com/watch?v=z8dHzU2qr-Y
  • Amazon unprofitable: https://www.linkedin.com/posts/jonnysitton_amazon-was-unprofitable-from-1994-to-2003-activity-7190985383499739136–OHZ/
  • https://www.dxbnewsnetwork.com/amazon-didnt-turn-a-profit-for-nearly-a-decade-after-its-launch-in-1994 (looks dodgy)
  • https://www.nasdaq.com/articles/why-amazon-has-no-profits-and-why-it-works-2014-09-05
  • https://www.cnbc.com/2018/08/29/no-profits-no-problem-the-economy-has-a-growing-appetite-for-unprofitable-companies.html (Investors have appetite for Unprofitable companies)
  • https://www.cnbc.com/2018/12/18/dotcom-bubble-amazon-stock-lost-more-than-90percent-long-term-investors-still-got-rich.html : At one point, Amazon lost more than 90% of its value. But long-term investors still got rich
  • Video with early history (but errors): https://www.youtube.com/watch?v=LOXRPeadvps
  • BBC: We are famously unprofitable: https://www.bbc.com/culture/article/20240628-a-36-year-old-jeff-bezos-talks-about-losing-money
  • “We are investing” in that BBC interview. The thing is, that only works once you’re cash-flow positive. Until you’re not, your only focus has to be on getting cash-flow positive.
  • https://www.vox.com/recode/2019/8/21/20826405/amazons-profits-revenue-free-cash-flow-explained-charts (there’s a difference between the early years, and when they were cash-flow positive).
  • Atlantic: https://www.theatlantic.com/ideas/archive/2024/02/amazon-profits-antitrust-ftc/677580/
  • Bezos explains why Amazon doesn’t make a profit: https://www.youtube.com/watch?v=Ue9uW1K_RJw (again, difference between cash-flow positive and profitable)
  • ALL THE BELOW ARE GOLD AND FROM THIS VIDEO: https://www.youtube.com/watch?v=Ue9uW1K_RJw
  • BEZOS (video at 3:30): “Those numbers are OUTPUT measures.” Compare that to how COVID numbers were measured – a trailing indicator.
  • At 3:18-4:14 (great quotes) - also talks about good return on capital. Also, in A16Z, look at the gray line where the cost of operating the business (other than hard CapEx) stays flat. No exorbitant salaries.
  • 5:10 - he talks about INPUTS to the business. Free cash flow and Return on Investment. Then work backwards to how you can maximize those. IT’S ALL ABOUT WORKING BACKWARDS TILL YOU GET BACK TO SOMETHING THAT IS CONTROLLABLE (5:20), then he talks about how to control the inputs to Free Cash Flow (lower cost structure). Then he talks about how to reduce costs, where “DEFECTS” come in. What are the defects in AI?
  • “Reducing Defects at the root is one of the best ways to reduce cost structure” - 5:30. JFC, he’s typing it out.
  • He talks about how to control inputs instead of reacting to outputs. KEY!
  • “We don’t meet with our biggest investors, we meet with investors who have the lowest portfolio turns.” – the most loyal instead of the profit-takers. (6:20) Investors vs. traders. If raising money, look for the long term “investors” vs. short-term “traders.”
  • If you’re going to spend time explaining the company, you should do it to people who are long-term investors, rather than traders.
  • “Owners think longer term than renters do.” 8:20.
  • “Nobody ever washed a rental car.” 8:45.
  • Where we spend our time is the most important decision we make in life (7:00)
  • All the Amazon LPs:
    • Customer obsession (build trust)
    • Invent your way into the future (multiple revenue streams)
    • Be patient
    • Operational excellence: don’t fuck up. Hire the right people.
  • What are the most expensive queries? How can you route those to the most inexpensive systems. Cursor does that by hiding it behind an “Auto” button.
  • AWS will hide that by letting you choose the cheapest model. This is like Levi Strauss having multiple suppliers, and choosing which one costs less.
  • Amazon is uniquely positioned (with Bedrock) to choose which model to route your request to. When shit hits the fan, they’ll own the “Router” as the cheapest (and best) place to send a request.
  • Router is where you want to be. It’s sticky and you will NEVER leave it.
  • Individual models DON’T MATTER. They’re commodotized. If one drops out and you switch to another one, you keep working, but the model maker is fucked. The router doesn’t care. Amazon doesn’t care.
  • Microsoft, Google, Oracle – they’re all tied into single models (or worse, their own hyper-expensive one).
  • Amazon doesn’t care. They’re the router. And they’ll make money no matter what.
  • If you’re OpenAI, Anthropic, or anyone else, you want stickiness. You want to make sure customers don’t leave you. MCP and the OpenAI API commodify the model. Customers have zero switching costs.
  • What you want is a sticky service nobody else has. You want multiple revenue streams. You want to invest in innovation.
  • What about Facebook? Multiple streams, but spending money like crazy on building their own people/orgs. That’s dumb. That means they’ll always be chasing creating a commodity and staying relevant.
  • What you want is a unique product that is branded for you. You want multiple streams. Hardware can do it, but it’s expensive (and REALLY REALLY sticky – if done right). Hardware allows you to reduce your operating costs if you can run inference on-device.
  • Training costs are sunk costs. The more you spend, the less you get back. At this point it’s product development and differentiation cost. You might as well spend it on SuperBowl commercials talking about how much better you are.
  • You can use Hallucination as a competitive cudgel AGAINST competitors.
  • If you want to be Top 3 standing, you want to have a super hit (Amazon and Harry Potter). You want to be known as the place to solve a big problem (peace in Middle East, celebrity fixing their life, a touching story). Give people an emotional hook to stay with you.
  • Anything commodotized is a wash. Put it out to help bring in customers from other domains and services. Like WSL for Unix. Or Steam windows compatibility layer.
  • Focus on where you are 100% good. Where you excel. No hallucinations, no cult of mediocirty, no errors. And if errors occur, fiercely tackle them. Become the trusted source.
  • Become the router. If you let Amazon become the router, you’re just a supplier.
  • Jump ahead to personal needs. Uniquitous AI Assistant. In home, when out, when moving about.

  • Hard press on privacy. Build trust.

  • Google just wants to collect your data. Hit them on trust.
  • Amazon just wants to sell you more goods. Hit them on ads.
  • Microsoft just wants to sell more windows and office. Hit them on irrelevance. They couldn’t get a phone or Cortana up and running.
  • Create partnerships. Build an ecosystem. Nobody can do this alone.
  • Differentiate between requests: easy ones go to a small LLM. Hard ones go to a bigger one. Others go to a secure function (calculator or secure data lookup).
  • MCP is built by little kids. It’s shit. Start from scratch and build in attestation (trust), caching (cost), payment (revenue), security (more trust – keep out the assholes), automatic updates (concern-free), hybrid (functionality where models can’t do the job).
  • Include humans. At all touch-points. They’re not your enemy. They make you better. Any story that says AI is replacing humans, quash it. If you let that get traction you’re a fucking idiot and deserve to be pilloried and strung up at the marketplace. Don’t create fear. Create trust.

CHECK THE SHAREHOLDER AMAZON LETTERS

Don’t care about the stock market. Find investors that are aligned with the opportunity: long-term being in the top 2. Survival of the fittest. Surviving a crash: investing in new initiative.

Video of a stock-car race. Which of them runs out of gas first.

© 2025, Ramin Firoozye. All rights reserved.