Token Economics (Tokenomics)

Andrew and Sid break down the hidden costs of AI tokens, why current prices are artificially low, and whether AI tokens could become the next global commodity.

As AI adoption surges and agentic workflows burn through compute, the underlying economics of large language models are reaching a critical inflection point. Join us to explore the rapidly shifting landscape of "tokenomics," the staggering hardware constraints behind the scenes, and what the true market clearing price for AI might actually look like.

To help us unpack this, the hosts dive into the downstream effects of "token maxing," why true economic equilibrium in AI is far off, and how historical technological shifts like electricity can predict our AI future.

• Defining what a token actually is and how text is chunked and processed by specific models.
• The illusion of current token pricing and why heavy subsidization by tech giants obscures the true cost of production.
• Exploring the flawed "token maxing" trend and why organizations are improperly prioritizing raw AI usage over actual return on investment.
• The severe hardware constraints and geopolitical pressures, including skyrocketing GPU and RAM costs, that make running local infrastructure incredibly difficult.
• Analyzing the criteria for money to see if AI tokens can become a true currency, or if they are destined to act as a tradable commodity like oil.
• The "Jevons Paradox" of AI efficiency and why cheaper compute actually leads to massively increased, rather than decreased, usage.
• How the future of work will rely on "cyborging"—combining human talent with AI—to increase productivity, using the surprising resurgence of human travel agents as an example.

This episode is full of economic insights and forward-looking predictions that are sure to change how you think about your next API bill. As we move into a new era of AI, it's the perfect time to explore the fundamentals of the next frontier!

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Transcript

Welcome And The Token Riddle

SPEAKER_01 0:00

Welcome to the AI Fundamentalists, a podcast about the fundamentals of the AI that impacts our lives and businesses. Here are your hosts, Dr. Andrew Clark and Dr. Sid Mengler. Today we'll be exploring the ever-evolving lifeblood of L. Token. Listen in to hear our discussion on token economics. Before we dive in, now since we did it last time on political neutrality, I wanted to start off with a little bit of a riddle and a joke for you, Andrew. So I want to ask, how many tokens does it take AI to change the light bulb? I don't even want to try and answer that one. Well, it's actually smaller than you think. It only takes about five for the actual answer, but then another 500 to provide a brief history of electricity, a disclaimer about standard safety protocols, and a polite concluding paragraph asking if there's anything else it can help you to illuminate. So I think that serves as a good lead into what you guys are going to talk

What A Token Actually Is

SPEAKER_01 1:01

about today.

SPEAKER_02 1:01

As always, to head off, we're going to talk a little briefly about like what is a token, right? So people are saying, like, I'm spending money on tokens, I spent $10,000 on tokens. What are these things people are spending money on? A token is a small representation of data which is fed into one of these large language models we've been talking about this whole time. A token doesn't have to be a whole word, right? So if you put in George Washington, that could be chunked as a single token. That doesn't have to be a few characters. But if you put in a long word, super califragilistic expelidocious, that's going to break it down into subcomponents, and each of those is going to be like a little four-character chunk, which can be used to compose up the larger word. There's usually also a dedicated model that's built just for turning text into singular tokens. So this is not like a small trivial task. This is basically a learn task. And whatever tokens you have, you pass to your model. These are model specific. You can't save money by doing one tokenization and handing off to four different models. Every time you want to go to a new model, you got to calculate those tokens again, and then those get shipped off. And this is part of the cost that you take on as a consumer when you both send models into the model and get tokens back from the model, which get decoded back into text that you can read. We'll have in the show notes an example of how Cloud does this and how you can put in text messages and get out a token count. And you can understand like how a single document that you own could be worth 500 or 5,000. And I think this pretty

Why Token Pricing Feels Unreal

SPEAKER_02 2:26

nicely segues this into token pricing. And where these prices are coming from and to what extent they are real and reflective of the actual cost of producing them. The stance I would take, and I think the stance most people would take is that token prices are currently artificially low. They've been subsidized by the large tech companies and investors to make sure that people have access to them, that they're using them, that they're excited to use them, and that the flow of tokens is going steady.

SPEAKER_00 2:57

Yeah, I think this is one of the interesting points that there's not that's going to be all speculation today. There's not a lot of hard data on some of these items, but what really is that market clearing supply and demand for tokens? And we even have, as Sid mentioned, like the thought, because these companies are private at the moment, they both filed for IPO, but they're still private, uh, the big labs, is that they're for every, you know, you're paying $25 a month for access, they're probably it's probably costing them four or five, six, seven times, depending on the user, that amount. And then you're seeing throttling and also like the SLA availability of anthropic is pretty low, which is surprising because like the how much load on the servers. But also, OpenAI has recently started, there's starting to become price wars. OpenAI uh published a few days ago, they might be lowering their prices even more to try and you know get more enterprise market share and things like that. Specifically, as we'll get into some points later, there's some uh like lack of ROI or like the token maxing trend is not really great for anybody of like, cool, we're spending all this money on tokens, what is it actually doing for us? So there's actually looks like there's gonna be a short-term price war, even of how low can you make the tokens while literally burning money to make sure that people are getting addicted to using these things. So a lot of very interesting parts, and that's where today we want to be, we're gonna do more setup of like what are tokens, how these different aspects work. But really, the the big question, and you'll see divergent opinions across the board, is is really the token economics, right? So, like, how does how does this work? Are we in a bubble? Are we not in a bubble? Or is is there value over time of these? Or as we'll dig into farther in this thing, it is it is can tokens actually be a form of of currency or or store value, or what does that even look like? Or is it a long-term something you can like trade on the market, or is it gonna be like, you know, data's the new oil? Is our tokens gonna be like oil you can do futures on and things like that? So lots of way more than we'll be able to unpack today. Hopefully, this is just the first of uh periodic check-ins on token economics. But want to set a little bit of the stage of kind of where we're going here. And the big question mark is the lot is that equilibrium to keeping your macro head of that supply and demand equilibrium. And the only thing I know for fact right now is that it's out of whack. Where exactly the market clearing price is and where's the supply and demand going to stabilize, I think is anybody's guess. And that's where a lot of investors are putting a lot of money on. But uh, I think we can all agree the current price is not aligned with the cost of product production.

SPEAKER_02 5:23

And just to give listeners a sense of scale here, if you haven't been keeping up with the latest API costs and you're just in a subscription, a thousand tokens from OpenAI's GPT-5 is gonna cost you about like a penny. You could scale that all the way up to Anthropic's Opus 4.7, which is gonna be two and a half pennies per thousand token. So there's a good amount of variability in the cost, and that cost is reflective usually of the size of the model. And the larger the model, the larger the infrastructure costs, and that has to get passed on to you. So as you as these models get better and better, if they get even bigger, we could see prices go up even higher. So to think and talk through like what are some of these future costs, as Andrew alluded to, most of us are paying subscriptions where we're paying $25, $50, $100 a month, and we are expecting to receive at least that value and amount of tokens back. But right now, those users are the most heavily subsidized. Uh there are some users that are using 10, 100 times more tokens in dollar amount versus what they're paying into the system, and there are some people that are using effectively none of their token limit and they're just paying the subscription, but the books have not balanced yet. So there is a hope and a dream that there is a normalization moment. So, what do those two outcomes look like? The hope is that

Future Scenarios For Token Costs

SPEAKER_02 6:45

either these models become more efficient in the way that Deep Seek has done it, or harder becomes more efficient. So there is an optimistic projection that we could get token cost to about like five percent of a penny is like the optimistic world that we could end up in, and then tokens will be available for everyone and very cheaply, and they'll be like a public utility, it'll be like your electric bill, no problem. There are also cynical predictions which peg tokens costing about eight cents to 12 cents per thousand tokens. As these models get larger, infrastructure costs don't go down, maybe even have to go up as we replenish our hardware supplies and we see less and less subsidization subsidization, uh, which could create a very alarming cost for many accessors who use these modern energetic flows that absolutely burn through tokens.

SPEAKER_00 7:44

Yeah, there's so many unknowns here, but then also different layers to unpack here. If we're seeing is part of the issue we've had with a lot of like everybody use AI mandates and stuff recently, is then, and this is where you have some of those the value out of whack in an ROI situation as well, is just use tokens. So there have been, and companies are actually already going away from it, but there is a trend for I don't know what was it, like six months, four months, something like that of token maxing where everybody's like, How many tokens can I use? The problem is that's not the right metric for anybody optimizing for. So they're also just wasting money, just making up stuff to have agents do loops and do things just when there's like, what have we actually been doing? I don't know, but I'm gonna I'm gonna look good to my boss if I use tokens. So they've been kind of like this perverse incentives, if you will, of like uh optimizing for the wrong thing. But then also like of the there's a lot of historical analysis I personally like to do more in the past of like Sid mentioned electricity. Electricity historically, I mean, there's initially it's expensive, it gets it gets cheaper, it's really kept price with inflation. You also have like internet connectivity, it's actually decreased a lot uh uh of cost from early on. So like if you look a lot of these technologies, like I think electricity and internet are kind of like some of the better analogies of they get normalized and they and they get they get more the price doesn't just like skyrocket over time. I know that's the big concern with AI, and that's I think a lot of the investment thesis of why so much money is being spent on it right now is that you're gonna have this massive like, okay, we can pull up the prices, but then you have the demand side showing, well, we're actually struggling to see value today. If you're gonna turn up the prices, what does that look like? But then there's also some dynamics of that we're really seeing, and and said you mentioned Deep Seek in as an example, is uh and personally what I've been using in my flow is you don't always need an Opus or a what is it, Fable now? I haven't played with Fable yet, but I need to get play with Fable. You don't need that for everything. So you can in your harnesses figure out what exactly, like where do you need to use that versus I'm a huge fan of the local uh Gemma models from Google that came out recently, where like they're local models, and there's a lot of interest of depending on what's your use case, what's you could still have a large language model and not need to be spending the huge amount, which will help keep this the demand down for some of these systems and the price. You also have some fun open source projects, one of them I like is called Caveman. So it try it also efficiently uses your tokens. So there's so much token efficiency stuff you can do as well. So it's inverse of token maxing is how can we compact? So Caveman compacts all of your conversations with a GENTIC system into short and curt, but then like so as Mike, your your joke at the top of the uh of the program here was that it only takes five tokens to answer how to how to change a light bulb, but it gives you 500. Part of that is the truncation of okay, AI, don't spend a ton of money giving me this whole verbatim's response when all I want is a yes-no answer. So there's lots of optimistics. So this is the more we dig into this, the more like we're just such at the tip of the iceberg of the longer-term evolution here and the amount of different components and optimizations. So, short answer here, I don't think it's as simple as like, oh, the more we make better models, now everyone's gonna go spend more money. There's a lot of nuances here, but as we get into chips and things like that later, there's it so many variables that we're not even gonna try and really to make any predictions today. I think that's a fool's errand at the moment.

SPEAKER_02 10:55

For sure. Predictions could be hard to make,

Token Maxing And Perverse Incentives

SPEAKER_02 10:57

but let's talk about some things that are happening right now. So here are some woes that we've heard in the market. Uber's CTO reported in April that the company had burned through its entire 2026 AI coding budget in just four months. Uh, Microsoft had to shut off cloud code licenses on June 30th after just giving access to their employees earlier this year. GitHub is moving all of its copilot plans to usage-based building, so no more subscriptions for them on June 1st. And they're gonna be linking your AI credits at basically like one cent each, right? So this is gonna be very similar to anyone that used Dolly and was paying for every single image they generated. And even, you know, Sam Albin will come on and publicly admit that token cost is actually a very major concern for a lot of consumers. And we've even seen clients and customers say things like a digital employee might be more expensive than a human one. So here's all the numbers you need. If you spend just 300 to 500 a day and raw API costs, that translates to an engineer salary of 100 to 180k. So if you are fully token maxed, you may as well be paying for an employee at that rate.

SPEAKER_00 12:10

I think that's like the key data point here. And I'm hearing a lot of folks looking back, like look doing that analysis now of like what are the jobs of like agentix systems, even despite you use the most advanced model or despite what people say, it is very much like having a very junior employee that you're having to like teach through everything. Now you can get a lot of value when you have a lot of you have a whole fleet of junior employees doing things and like checking different aspects of stuff. It can be very productive, but to the point of like, at what point does it be like you could have a really, really great talented engineer who's now also learning and then will eventually move up to be a senior engineer that can do other other stuff. Man, I'd prefer to invest in that, in that person, than to just have like a uh be paying anthropic for a eh, you know, output on some of this. So it's that it's gonna be very interesting how that calculus comes over time. And it is to to Sid, your point of like with Altman is now mentioning it, that's probably one of the reasons they're reducing the price or wanting to start that price war to keep everybody wanting to use AI AI, but that puts the the the investors' considerations on like, well, what's the what where does this balance longer term and the amount of like the valuations that's in AI is assuming you're gonna have these like a massive profits, obscene amount of profits to justify the valuations we have currently. And that's looking less and less likely. And then we have to figure out to be able to even them function today at their current size, how in the world at the current token price, we have to get so much more efficient with our chips and things like that, because they also they all have a shelf life. So with all these companies investing a lot, you have like three what three to five years said for a server, like not that long before you have to then replace them. So if you're now like you're so you're you're they are so below what the actual cost is to then keep it so you're not actually just paying an employee.

SPEAKER_02 13:56

And I think hardware is actually the big limiting factor here. So let's say that you're saying, like, okay, fine, there's all these big providers out there, but it's no problem. I'll just spin up my own hardware and do it on my side. That's surely feasible, right? And I guess I want to highlight to listeners that like I think that's a great solution if you already have hardware, if you already have a server sitting at home, if you already have a couple laptops, if your company already has access to some data frames that are maybe sitting around underutilized, by all means. I think a local model is a really great way to save some money, and then you're only on the hook for the electricity. So you're not in charge of any of the costs if you get that more directly. However, if you're coming in brand new and you think that you're gonna build a server for this, I just want to highlight a couple numbers here. RAM prices are up 220, flash storage 472%. So sorry, for anyone that's trying to build a a network attack, a network attached storage server, and a GPU is up 100 to 300%. These are crazy increases in prices, which are actually extremely limiting. And this is actually you know a direct consequence of you know the Samsung's and the SK high indexes of the world having to shift all their production over to enterprise level RAM and storage and GPU production, and there's just not a lot left for consumers to it's wild.

SPEAKER_00 15:17

Yeah, so that that's that then we'll get into some of the geopolitical and things like that later. But I actually was looking at and everybody had the same idea. I tried to go so strangely, long term long time ago, Apple used to be very expensive for hardware. They're actually one of the cheapest places now to buy computers, but uh their Mac Mini and their Mac Studio, the two like boxes, because you can be using some of these like agents and things. You don't need to have a you can even communicate with like Hermes via Telegram or all these different things like that. Everybody had the same idea. You cannot buy a Mac Mini or a Mac Studio. They're sold out for like three months straight. Uh and they have and our Apple has the in for a lot of these factories. Like you can't, to Sid's point, you can't actually buy this stuff right now.

SPEAKER_02 15:59

Yeah, so I mean, you know, there's a really big price squeeze. I mean, we're we're in such a uh price squeeze

Real Budget Woes In Companies

SPEAKER_02 16:05

or even seeing some like truly absurd ways of trying to resolve this. Uh, some of you may have heard of underwater server farms and data farms. Uh, there's the next level now. It's the space data center. People have been talking about this, saying, like, oh, well, if we just put the servers in space, we won't have to pay for any of the cooling. I think that any of our listeners can see how this is like obviously outrageous. But, you know, just to give you two very strong reasons why you can't do this. A, your server is gonna get bombarded by cosmic waves and UV radiation, which is extremely bad for data integrity. So expect a lot of cost just getting spent in fixing bits or losing hard drives. And also, it's not cheap to put something in space. If we're really floating, putting something in space as a way to save money and escape regulation. We're clearly in a space where people are so concerned about money that they'll say pretty much anything. So, why has this become such a problem and why is this like getting sort of hand? We've been talking about this before with this idea of these like token maxed agentic flows, but I guess we can be more specific into what is this whole slop canon world? Why are people spending so much on tokens and why are people using so many tokens? And I think this is partially for quite a few reasons. I think maybe what's more tr most transparent to workers is that workers are seeing a push from their employers to use as many tokens as possible, and organizations are even using token count as a metric for your productivity, uh, which is I think very funny because I think we've at least most of us have learned the lesson that metrics like lines of code produced by an engineer is not a good metric. If we just look to Goodhart's law, when a metric becomes a target, it ceases to be a good metric. So people that learned, oh, I just have to write a lot of redundant code with huge comments without any refactoring, and I can be really productive. Okay, I'll do that. People have learned the same thing for tokens. Tokens are super easy to game. You say, How did I square the light bulb? And please tell me everything about every light bulb that's ever been created, and make sure it's foolproof and also just do it for me. Whatever it takes to get tokens, people will learn how to do that. And that, unlike code cost, is real cost that the organized organization has to actually take on them. Uh, I think very finally, as an example, Amazon used to have a token leaderboard and they would measure how many tokens every employee was using and then rank them. Uh they have they've shut this down since then. And the most obvious reason is because it was costing them way too much money, and they had to publicly come out and say, don't use AI, just to use AI. So maybe we're cresting over into this kind of like walking back this idea that like tokens should be spent freely and tokens are a metric of productivity. I think we're still cresting. I think there's organizations that are still on the other side of this that are saying, like, how many tokens can you use?

SPEAKER_00 19:09

Yeah, and as we talked about on this podcast before, like it's uh when we had a couple podcasts back when we talked about history and and things like that, like so much of history just repeats itself on these areas. Same with like used to be the lines of code engineered rights, you said your point. And like we're having that again, or like the choosing the wrong metrics. Like it's notorious in businesses, always choosing a metric, and it's just easy to have a nice little easy metric. But uh, I was reading like an interview recently from Elon Musk, think what you will about him. He's very productive at getting things done. Was that the one of the odd ways to like maximize, like we've talked about maximizing utility, like that's really the measure of success is maximizing your utility to the world, or like what are you actually getting done, like uh that that makes somebody else's life better or makes the makes the world better is like a measure. That's very hard to do in like a business, but it's when we oversimplify is it becomes a problem in finding you know the wrong targets and people optimize for a metric if it's tied to performance versus more holistic measurement. But also it's it's not gonna dive into full of it here, but if you go back to like Bitcoin mining and things, we have some of these same type of like now, unless you have this huge like underwater server farm type thing, it's not efficient for anybody to be doing that. But there was a whole nother back uh back in that day, hardware was when everybody was kind of going wild about mining Bitcoin, hardware was was out. So a lot of this stuff goes in cycles, and there's I think there's some too in depth

Hardware Limits And Going Local

SPEAKER_00 20:33

for today's conversation, but there's so much of this to know what's the hype and what's the long-term signal and trying to figure out what's there of like I don't think AI is going anywhere, but I think like Google is well positioned since they make their own chips, they're just building inherent into the products. I think Apple's gonna be in a good spot, but it's like AI just becoming kind of part of your flow versus these contrived, let's use it everywhere just because it's AI versus knowing what it where it works and where it doesn't.

SPEAKER_02 20:55

And I think to build a little bit more on the historical context of what's happening here, this is actually very analogous to electricity. We were under the impression that once electricity became really efficient and really cheap, the amount of use would normalize and people would just get to enjoy the cost savings and live a cheaper life. But that's not what happens. There's a thing called Jevons paradox, which kind of highlights this idea that efficiency gains can actually result in increased use and not less human work. So what happened when tokens got cheaper and more accessible? People thought, what if I had three agents do the same thing and then they voted? That's just increasing you know usage super linearly, and that's not efficiency savings. This is showing that you know people have a hunger to use this technology, and any efficiency gains will be used to uh increase and improve their usage of it.

SPEAKER_01 21:45

Yeah, no, I completely agree. And some of the things that you guys have already touched on, they it just reminds me of economics one on one, supply and demand. Andrew, like you talked and said what you talked about, we're seeing these crazy moon uh pun intended moon tracks. Shot ideas to increase supply. Supply is going to come up and then the uh efficiency at which people use AI is going to go down. I think as we go forward, AI tokens may seem like a thought of yesteryear as how they capture or how companies capture and capitalize on use of AI will change. It'll be folded into software. There's a long track record of obscuring what raw prices are. So I think it'll be a really interesting evolution. And I would guess that there's probably going to be an oversupply of infrastructure to be able to prepare that stuff as everyone does data center wars and all that loveliness. And we'll see. Oh my gosh, geez, no one wants it anymore. I'll give you absolutely rock bottom places, and you know what comes with that? We're going to find different ways to weave it into your cost model for any software that you're buying, using, etc. So very exciting times.

SPEAKER_00 23:04

Yeah, this raised the question as well, and I've been hearing this a lot and thinking about it personally, is like for a lot of SaaS platforms, do they really need Fable? Or can you use a Gemma, right? So, like, depending on what is it? Is it like glorified semantic search? Well, just self-host a model, right? So, like, that'll be interesting to see how that plays out as well. Of uh, I really don't see a moat in a lot of these systems longer term and why like Cloud Code, and I guess now Codex, I haven't played with it yet, is even better than Cloud Code. It's the harness, which is the the the tools, the skills, the things around it, more than the model themselves.

SPEAKER_02 23:39

So we alluded to this before, but I mean this is a question for our economist here is with all this craziness around tokens, with how they seem to have almost inherent value, do we see a future where this AI tokens can evolve into a kind of money or a commodity? Oof.

SPEAKER_00 23:55

Well, this is this is the big question for sure. Um, but I think it it's good to there's really three attributes or or principles of what constitutes money. So I think that's where we really need to start this analysis. And I mean this will just be very high-level back to the napkin today, but the three components are store of value, uh, unit of account, and medium in exchange. So, what do those mean? So, store of value means that the value stays relatively stable over time. So uh I don't think we've met that criteria yet. Uh, back in the day, one of my uh PhD research papers was actually doing this for uh crypto tokens, for like could a central bank actually use uh use tokens, public tokens, uh in in their reserves. And one of the things we found was that store of value didn't work well because it's not a volatility. And I think there's a lot of uncertainty of like if I buy a token now, is it going to be deflationary, meaning the value it will become more important over time? But then that's there's economic issues with that potentially. Will it become inflationary, which is uh and uh become worth less over time? So there's a lot of things to unpack there. I think it's way too early to tell, but uh

Slop Canon Metrics And Jevons Paradox

SPEAKER_00 25:05

the other part is it's not really unified right now. Like a token, as I mentioned, the token is is really around four or five characters, is what a token is, but there's different token how do you get it into that system? And then like the models at themselves, they'll have separate tokenizers and things. So it's not as it's not really unified and there's a lot of variability. So I don't think we quite hit store of value yet, but I could see longer term that could be something that's more commoditized and more unified of how we of tokens. Uh, unit of account means that you can you can reliably know that uh so store of value is that the value stays pretty consistent stable over time. Unit of account is something that I can go to Sid and say, hey, I want to I want to buy this from you from five tokens. And it's something we can be doing kind of uh that do accounting in and have reliability. Medium of exchange is the ability that he would he'd be willing to accept my tokens for his. So it's they're interchangeable. There's that interchangeable aspect of them. So the really the underpinning of those three items is really trust. And this is where like the Federal Reserve is probably the best central bank uh the world has, and with its dual axioms of kind of like uh in the equilibrium here of full employment or as close as possible and and stable inflation, of keeping an inflation target, like don't let inflation go too high, don't let uh unemployment go too high, and you use open market operations as one of their levers of increase the money supply, decrease the money supply. But the money supply, everybody will will interchange in dollars, even globally. And then you also denominate other assets in in dollars. And even though, like, we all like, oh my goodness, inflation, I hit over 4% this last reading in the consumer price index because of how high like oil prices are. We're talking about, oh my goodness, it's 4% year over year, or 4.7, I think. Tokens, as we're mentioning right now, like at the top of this podcast, they're all over the map. Tokens are you have no idea right now. So I don't think right now that they're they're they're not standardized enough. They haven't been around long enough to or earn like the store of value. And uh an anthropic token and an open AI token aren't really something that's interchangeable. So I don't think we hit any of the values of money yet.

SPEAKER_02 27:08

And that's a really good summary. I think that it kind of highlights this idea that like people have this idea that like it could be a speculative commodity, but it has a lot of inherent issues, the same way that like Bitcoin, while it was touted to be our next currency, actually didn't meet a lot of these marks, and so it hasn't really seen a lot of real life usage as a medium of exchange. There are niche uses, but I don't think that we've seen wide adoption. Do we have any concerns that these issues that we're seeing with token cost and token pricing and all this fluctuation is coming from more than just hardware? Are there like other political aspects that might be affecting these prices and their volatility?

SPEAKER_00 27:47

One very interesting elephant in the room, I think, is the geopolitical aspect as well. Where are the majority of the world's chips made? Sid, you visited there last year.

SPEAKER_02 27:58

It's Taiwan, it's TSMC in Taipei, yeah.

SPEAKER_00 28:02

Yeah, so there's a lot of uh there's and we're not gonna get into any politics or any of that kind of stuff today, but for anybody that knows what I'm talking about, there's a lot of interest around that country right now from multiple other countries. And that's there's a major component for that, is because there's a lot of chip manufacturing there. There is some uh US is looking at building some domestic capability uh and things like that as well. But that there's there is some ramifications for the in the broader like who controls AI or is an AI an important thing longer term, but also like the literally access to it based off the short-term lived chips and things is connected. There's some geopolitical concerns. And I said, you mentioned uh back there as well. So there's also money and then there's commodities. So I I think tokens probably have a better future as being a like a commodity and/or at least something that you can maybe be we you have the hedges, you have data centers, and we right now have um, I believe XAI or sorry, uh yeah, I think XAI uh are they're renting one of their data centers actually to anthropic, correct? Sid? So like there's a little bit of like the data centers could be something that is a little bit of a commodity potentially. So like in maybe tokens are still a little bit because tokens are model specific, although I do believe longer term those could be abstracted a little bit more. The the compute access could be something more commoditized. So one example is oil, is definitely like a I would say is a good analogy as a commodity. It's not money, it's hard to, it's not it unit of account, medium of exchange, it moves a lot, but it's also you can't really walk around with a barrel of oil and go to McDonald's and get a sandwich, right? So like it's not something you can really use as that medium of exchange too well. So it's more of a commodity and not money. I think tokens have more of a chance of being commodity than money based on those areas. And like Delta has actually has their own refinery, which helps them do hedging of future prices. So there's certain airlines that do better than others if they pre-purchase or they hedge some of their um with futures of oil prices so they can kind of normalize so they don't get really in trouble. I forget there's one airline that doesn't. Majority of them hedge, though. There could be that kind of aspect longer term of hedging data center usage for these companies of like you don't know what's going to happen with Taiwan and some other places. So maybe there's ability to, maybe it's the data center more than the token, but we could if we can abstract tokens into like an aggregated amount, could be something that's more of a commodity that could be having, I could even see a world of derivatives. So a derivative is a financial instrument on top of something. So this really you have oil, but you trade the future value of purchasing or selling that with a put or a call like future. So I could see something like that happening with tokens, but not as money in the near term.

SPEAKER_02 30:44

That's right. And I think that highlights this idea that not only do tokens experience effects from the market, the market is getting a direct effect from the token world. You know, most of these large AI companies and research labs are US-based, at least currently. Our Nvidia's are here, our Amazons are here, our metas are here, our Googles are here. They're all happening here at home in the US for us. And we're we're seeing this interesting world where AI growth is now actually driving and carrying the market. So a lot of this growth is almost cyclical. As AI growth powers the market, powers AI, and we're seeing this like huge rush forward, which you know people are now describing as quintessentially bubble-like. But this kind of geopolitical influence isn't just an impact on tokens. Because tokens have gotten so big and so important, we're actually now seeing the entire US market, which would, I guess, otherwise be considered in a recession, being carried by this AI growth and holding up our markets.

SPEAKER_00 31:45

Yeah. And that's always the is the music gonna stop ever or or or not? And I don't, and that's I think what the market is trying to figure out right now, because you did also have record profits across the board, earnings for a lot of for

Can Tokens Become Money

SPEAKER_00 31:56

uh companies. Um yeah, it's very, very interesting, but this is where some of the position we are seeing companies like Google, I think, doing a great job of it's just getting embedded everywhere. And that's I think the big thing longer term is how much is AI the separate thing versus it's just like just like a computer. Or as Steve Jobs uh has a famous quote of like bicycle for the mind, when he had seen the most efficient way for a human to travel is actually by bicycle. The least amount of energy used is by bicycle. So computers were essentially a bicycle for the mind. It feels like our bike got upgraded a little bit, but it also becomes like computers are everywhere and and they're just critical to the economy. It's less like it's a you're a computer person, everybody works on a computer, right? So that it's gonna be interesting to see how all this plays out.

SPEAKER_02 32:41

Absolutely. And I think that brings us to just about the end of our conversation. So I'd love to hear any concluding thoughts you guys have in the world of tokenomics and token usage and how we have to live in this world now.

SPEAKER_00 32:55

Yeah, I mean, I like the trend that I'm not I'm not surprised whatsoever. We're going away from token maxing. We're actually, I think, on a longer term go back to the realization that human talent really matters. We've kicked humans to the curb a little bit recently of like, eh, I can just replace everybody with AI as a like uh to to afford the tokens. But to Sid's point, now people are realizing back, like, hey, you actually have to, you have to do the math. Sometimes it's worth it, sometimes it's not. I I think generally we're gonna have an increase in productivity with AI. I think it will eventually have a big bump as people know how to use it. But I think where we're gonna really land is cyborging it, is where the most important thing is. Not that the big labs have the incentive, which is actually short-termism, if you think long term, AI will replace all humans, well, then who's gonna the economy's gonna crash because no one's gonna be able to buy the goods and services, right? But your longer term, I think we're gonna see like computers. It's like when computers came around, they didn't replace humans. Actually, to the other point earlier that Mike had, we actually made more work, but we all got productivity went up, quality of life went up, quality of living went up uh across the board. And you have this humans can be much more productive. So it's like we're abstracting another layer away, and what jobs are will shift. That's the creative destruction, if you will, in the economic parlays. But the value of like what makes humans human, the unique things that AI does not, you know, as we've talked about, does not think or reason the same way, or is not creative like a human is. So that curativity, maybe our jobs will shift, but we'll just I think longer term we're gonna get to that equilibrium of everybody gets more productive, economic growth increases, everybody's using AI like it's a computer, not like a computer, but it's that native thing. But the value for human experts and human

Geopolitics Chips And The Bigger Market

SPEAKER_00 34:36

expertise stays. And I think that's the narrative that's been kind of missing is like, hey, we're gonna get rid of all the humans, replace them with AI. I think that's short-termism, and I'm trying to raise capital to for my AI lab type thing, versus where we're gonna end up longer term, is really the the equilibrium in my mind is that the level of productivity and actual economic growth and quality of life grows up across the board, and humans do what humans do best, and AI can help humans do more and do the things that humans don't need to be doing, and there's a good symbiotic relationship, kind of like we saw with computers, with automobiles, with electricity. I think longer-term AI is gonna become a commut gasoline, if you will, for even back having something to don't want to get in political there, but like in any case, you have these things that just increase quality of life and they just become everywhere. I think we're gonna get that with aid.

SPEAKER_02 35:26

That's absolutely right. And and then with the world moving towards these agentic workflows, I'll remind all the audience that these agentic workflows are still like in their nascents, they're still early, they're still young. We have a lot to learn about what makes them more efficient, what perceiv what increases our perception of them being human and competent, and then maybe actually showing actual increased competency and ability and usefulness as agents, which I think is very important and very relevant. And our we hope that in the future episode on agentic engineering we can revisit some of the stuff we talked about with our travel planner as an agent and see how does AI slide into that now.

SPEAKER_00 36:06

For sure. Excited for that one. And last part I'll I'll leave with on that point is there's a recent article in the Wall Street Journal. Guess what industry, ironically, is having a huge renaissance. So many travel agents. People, despite like travel agent is one of the classic examples, like, oh, I can use AI to book my flights. Well, what's the growth industry right now? People actually want more human travel agents to do like the last minute and like the humanness about it. So that's also a very interesting data point that it reinforces that, and those

Human Value Cyborg Workflows Closing

SPEAKER_00 36:36

humans are probably using AI to help do some of the stuff, but it's you still want that it just like elevates, it raises the floor for everybody versus it being like, hey, we my bot can book my flights. I don't need a human. No, you actually need the parts of humans more. So I thought that was an interesting data point, just one data point, but the fact that one of the jobs you would think is one of the ones most exposed to AI is actually one that is a growth industry at the moment.

SPEAKER_01 37:02

Thanks for tuning in to another episode of the AI Fundamentalists. Make sure to smash that follow button on iTunes, Spotify, or your favorite streaming service so you don't miss out on new episodes. Until next time, keep thinking, questioning, and learning about AI.