Sam Altman Brags 100 Token Billion Per Month AI Customer for OpenAI - Tokenmaxing Gone Wild

追加: 2026-06-06

[00:00:00]So coming from Axios, Sam Alman dishes on Open AI's top token user and apparently their top token user uses 100 billion tokens per month. Holy hell. Who do you think that is? What the hell do you think they are doing with all of those tokens? Uh that is a very interesting question. So I will be curious to see what comes out when the when open AAI uh goes for their IPO and they have to put in uh their S1 SEC filings and we get to get get to understand a little bit more about how their business works. I'll I'll admit it right here. Maybe there is something more to their business than I understand.

[00:01:00]What? I might not be perfect. I might not be. I mean, I'm probably right. To be clear here, I am probably right on this whole disaster, but I will say I might be wrong. Remember back in the day? Has anybody followed me long enough to remember when the Firephone was coming out? Amazon was coming out with their own smartphone called the Firephone and I was like, "Wow, they are positioned in the right place in the market to take a massive chunk from Google's duopoly on the smartphone uh world." And uh and yeah, it was it was dead within like 3 months. So, here's the thing. Let let me be crystal clear when I when I put out some proclamations. I I'm just a geek in my basement reading the information that you can get a hold of and after 30 years in the business I am saying what my bets would be but who know who knows maybe maybe there is something going on we will see the S1 filings and we will just be amazed I I think it's going to look a lot like we works S1 filings but look I'll admit maybe I'm wrong and I think this is kind an interesting because like that's one of the big questions with AI is it's just you know who is using AI and how much of these systems is actually being consumed because that's one of my big concerns here right I'm not anti- AI I'm not even anti- open AI to be honest with you it's just the valuation I'm I'm anti open AAI being valued at a trillion that's the issue if you told me open AI was valued at $50 billion Honestly, I'd still think it was a wee bit high, but you know, be like, "Oh, okay. I can get my mind around that." The problem again with with this technology and these businesses and all that is is there's there's a business around it and investors investors want to buy low, sell high. So, when they buy at a trillion dollar valuation, they want to see Open AI go to two trillion and 5 trillion and 10 trillion, right? They want to buy in when OpenAI is worth a trillion dollars and get the returns that the Nvidia investors have gotten when they invested 5 years ago and Nvidia is up by like a,000%.

[00:03:25]And that's a problem when when you when you see the cost of tokens plummeting, when you see the use cases for AI just not being borne out, uh we're going to talk about another video is where Uber companies are literally cutting their AI spends because they are not seeing value there. That's why I look at something like a trillion dollar valuation and it's like that's that's probably not a good idea. But who knows? Who knows? May maybe the maybe the government maybe there there's there's so many people out there that want to burn a 100 billion tokens per month because that's the whole thing right if you look at the models so they got a whole bunch of different models and there's input tokens and output tokens and reasoning token all this [ __ ] going on anyways I don't know it's somewhere between like now like with open AI it's like 40 cents like a buck it's like 40 cents to $5 per million tokens when we talking about a million tokens a million tokens is um uh the three the the entire Lord of the Rings trilogy plus a couple of books That's what we're talking about with a million tokens. And so again, that's that's where I I mean, you've just got to you got to sell a lot got to sell a lot of tokens. So, who knows? May maybe there are companies out there willing to burn hundreds of billions of tokens. And so, kind of like how I get paid, right?

[00:04:39]People ask, you know, Eli, how do you make money on YouTube? Well, it's like, well, when somebody watches a video, they see an ad and I get a fraction of cent for every view. you know, you scale up a fraction of a scent to thousands, tens of thousands, hundreds of thousands of views, you know, that turns into an actual paycheck. So maybe, right, that's the thing, 40, 40 cents, you know, uh, per million tokens, that might not be much, but maybe if you scale that to a 100 billion tokens per month. May May maybe that that comes to a revenue stream that I don't know makes half an ounce of sense. So, OpenAI CEO Sam Alman said the company's top token user is going through 100 billion tokens per month. Uh, as people inside and out of AI labs are spending more and more on tokens, Altman acknowledges that cost concerns have become a huge issue. And this has been a big issue uh ever since the reasoning models came out. That's one of the things like with token consumption you have to be careful about. So, how it was back in the old days with trackt 3.5 turbo, right? When you looked at token usage, you had the input tokens and the output tokens.

[00:05:46]1,000 tokens equals approximately 750 words. So whatever, right? So you you put in your query, which is however long it is, your context window, and then it spits out a result. You add up the tokens for the context window. You add up the tokens for the result. That gives you token utilization. Well, here's the thing. Then they came out with reasoning models. And so reasoning models are better.

[00:06:08]And one of the reasons reasoning models are better is basically when you put in your context window, the model itself or the system itself will go in and it'll kind of like try to figure out what the hell you're actually asking. And so the interesting thing is when it's figuring that out, it's it's burning tokens, right? And then when it tries to come up to an output, it's also figuring out if the output is what it thinks you actually want from what is in your context window. And all of that reasoning starts burning up tokens. So whereas before yeah you might have had 500 tokens for your query and 1,000 tokens on the output right input 500 tokens output 1,000 tokens fine well now it's like you input 500 tokens and then it does all this reasoning that might burn thousands of tokens for the reasoning and then you get your thousand token output and all of a sudden all of a sudden what was 1,500 tokens in cost is now maybe five or 6,000 tokens in cost. Right? Look, at 40 cents per million tokens, who the hell cares? But you start scaling, that starts to get a little bit more expensive. The issue that we have right now is this whole agent thing going on, which yeah, talk about dumb architecture squared. And basically what these agents are doing is you create these AI agents, you give it a specific test set of tasks, and then many times they actually run on a cron job. I was actually kind of surpris and AI. Anyways, essentially cron jobs where every time period the agent is triggered to go out and find information, pull in that information, analyze if that information has anything to do with its instructions. If it has something to do with its instructions, then it figures out what it has to do with instructions. then it creates an output and that gets but that's going like 24/7.

[00:07:59]So as before, you know, every time somebody put in a query, it might cost you 1500 tokens. Now you have AI agents using reasoning models that are just burning tokens all day long, which can get very expensive. It's amazing. Like you you you hear some of these stats from some of these agents. Like the guy who created OpenClaw, literally the guy who created OpenClaw was spending $1.5 million per month on tokens.

[00:08:29]What? One guy.

[00:08:33]So that's why this stuff is getting a wee bit expensive.

[00:08:38]Uh what they're saying, uh quote, "The token leader OpenAI uses about a 100red billion tokens a month. To my embarrassment, that's not the token leader in the world. But we found someone that used even more. Altman said on a live stream about OpenAI's enterprise adoption. The number is staggering when put into context. Over 6 years ago, the top token user at OpenAI blew through about a 100,000 tokens per month. Altman said, uh, while that could sound like a positive for AI Labs, that benefit from token spend, the boom in demand also highlights, quote, the infrastructure challenge ahead of us.

[00:09:10]Elman said, "Compute constraints can prevent AI labs from being able to service all the inbound demand they're receiving, putting a ceiling on potential revenue growth." Though, I will say again, this this whole compute constraint thing, this is a whole bunch of garbage, right? 90% of the users of Open AI services are free users.

[00:09:30]Th this is not everybody using OpenAI is paying for the service and they're hitting compute constraint. 90% of their users are free users. And that that is one thing though again when you start looking at numbers like what does that actually mean? So 90% of the users are free, 10% are actually getting paid. But that actually doesn't tell you what the paid percentage of hardware utilization is. Like it it might be that that 90% of the users actually only use 10% of the hardware and 10% of the users use 90% of the hardware. To be clear, to be that that could theoretically that could be the case. I'm not sure I believe it, but I do find it to be interesting. They they keep pushing this idea of compute constraint. We need more hardware. I would argue I would question that. He said cost concerns are the second most common issue he hears about about from customers behind simplifying AI workflows. Cost never came up before, but is now uh quote all of a sudden a huge issue. Dude, it's not all of a sudden it's it's reasoning models and AI agents. Like two years ago, I'm not sure when reasoning models came out. Anyways, let's just say two years ago, reasoning models weren't really a thing. So again, it was input tokens, output tokens, that's all you had to worry about. Then you start using reasoning models that start jacking the cost up a bit, right?

[00:10:53]Now you got to start thinking about a little more. Uh and then you get AI agents running 24/7 using reasoning models. That's not an all of a sudden thing. It's it's literally how the technology is being utilized thing uh that that that that becomes the issue.

[00:11:08]Shouldn't exactly be a shock to you know anybody paying attention. Quote, "We want you all to be able to use AI and never worry about it being great and affordable." He added, quote, "If there's one thing to get ready for as the next phase over the next year, this is the one I would pick," Altman said, adding uh this will change how companies plan around AI usage and security.

[00:11:29]Autonomous AI that runs without being prompted can get expensive quickly. Yes, if OpenAI can successfully meet enterprise demand for cheaper AI usage, it could allow them to take market share from the their biggest competitor right before both AI labs are expected to go public. And that is an interesting thing to be looking at here, right? So, all these companies are IPOing. So, SpaceX is IPOing June 12th. Apparently, they just set their their stock price at $135 a share. And so their valuation, Lord help us, is $1.75 trillion. They have now put $1.75 trillion in stone for the SpaceX valuation, right? Anthropic, they just uh they just filed to IPO. Who knows when that is actually going to happen. Open AAI still hasn't filed IPO.

[00:12:16]One of my big arguments though with Open AI is at this point it's an also rang.

[00:12:20]Like you think about it for a second, right? You look at Google, you understand the products that Google is putting out and you understand that they are they already have revenue and profit coming in. You look at Anthropic, right?

[00:12:32]Anthropic is highly focused on coding.

[00:12:35]You know what you're buying from Anthropic. When you look at Open AI, they don't really have a brand. They they have a brand name. They have a brand name, but they don't really have a brand. Like what what is Open AI actually for? And I think one of the dangerous things here is if you look at if you listen to Daario, God, God bless Daario. Apparently, he's the only [ __ ] that actually opened up a business book once in his life, he actually talked about cost of resources.

[00:13:04]He talks about resource utilization and all that kind of thing. And he talks about the fact that he he's trying he's trying to find the biggest revenue and profit drivers for AI tools. not simply trying to sell AI to everybody at any price point. Like he's looking at it and trying to figure how what can we actually sell with AI that's going to make us profit. So, one of the curious things here with uh with Sam Alman, right? Sam Alman is is all about bigger is better, right? I read that whole book with Open AI and it is like they believe they they believe in their heart of hearts they believe more hardware, more data equals more AI.

[00:13:44]I disagree with that concept, but that's what they're going for, right? And so, one of the concerns here with Sam Alman, if he's under the more is better mentality. One of the things he might do is uh since all these other companies are are IPOing first, he might go in to try to get market share, like again trying to maybe drop the drop the price on uh on their services or do something else in order to grab market share. But one of the concerning things is what what if he grabs crap market share?

[00:14:13]It's like, look, we won the ghetto.

[00:14:17]It's like, yeah, yeah, great. Yay.

[00:14:21]Right. You know, you have Anthropic over here and Anthropic o only has, you know, 5% of the population of this particular city. A and Sam Alman has 50% of the population of the city. Obviously, Sam Alman's winning. And it's like, well, yeah, no, but but Dar Daario got all the rich bastards.

[00:14:43]Daario found the 5% of the population that can actually pay a crap ton of money for his product. And you know, Open AI went for everybody who is counting out their pennies to pay for tokens.

[00:14:57]Which one of these is a better business model? Right? You think you think about this with, you know, Android versus iPhone, right? Android has a much larger market share than iPhone does. iPhone makes so [ __ ] much more money though, right? That's just that's just how business operates. So, it'll be curious to see as as Sam Alman more fully understands that Open AI is the also ran in this game at this particular point in time. that there does not seem to be any brand strategy for what you buy when you buy open AAI.

[00:15:35]If you're going to buy CH, like what are you buying with Chad GPT? With all of that, it'll be curious to see where they push forward. So, anyways, what do you think about this? What do you think about OpenAI's biggest user uh buying a 100 billion tokens in a month? Do you think this will continue? Again, we talk about that with things like training processes and all that kind of stuff, right? The highest utilization many times is the implementation phase and then once once you're done with the implementation phase, like the the the actual load may drop a lot. Do you think this might just be an implementation phase for a company? I don't I don't know.

[00:16:14]Do you do you do you do you still think Open AI is actually a frontr runner? Put your thoughts down below. If you like these episodes, they are on Spotify and on Apple Podcast. Link is in the doobly-doo. If you want to follow me on LinkedIn, there's a link down there, too.