The math does not add up and the logic is not sound. That could be the summary of the current AI bubble. If that is true and AI does not deliver on its promise, what could the massive build-up of data centers be used for?
Enter the tokenization of the economy: the process by which legal ownership rights to real-world assets (RWAs), commodities, and currencies are converted into blockchain-based digital tokens. In theory, this allows for more efficient, instantaneous, and secure global trade. In practice, it’s a transformation of the way the economy works by merging finance and digital technologies.
Tokenization enables the financialization of almost anything that exists, turning passive physical realities into highly liquid, speculative instruments. You may tokenize a property title or a future crop, but also part of an artwork or a fraction of a worker’s shift. It enables the speculative economy, which is based on financial instruments and not on trade or production, to continue expanding. This is one reason why financial institutions are backing this transformation.
Tokenization adds a new ledger to the economy. The banking system introduced a structural ledger into every transaction through the use of fiat currency. A transaction was no longer just a deal between two parties; there was always a third party—the banks, which controlled the currency that the state imposed. Because money ceased to be an asset in itself, those who controlled the currency had a ledger of control over the entire economy.
Tokenization proposes that it will reduce the number of hidden ledgers (private banks, clearing houses, or the central bank), but these could be considered all part of one structural ledger: the fiat ledger. Initially, tokenization will add a second structural ledger: the asset token or the cash token. In a transaction between two parties, they will use fiat (the first ledger) in order to trade a token (the second ledger). This is the reason why technology companies are backing this transformation. They will essentially have become an indispensable part of the financial system.
Eventually, when CBDCs or stablecoins substitute for fiat, both ledgers will be merged into one. This will mean that the financial class and the technocratic class have merged into one. For the rest of us, it will mean that the nature of money will have changed. Fiat currency holds a nominal value, even if not an intrinsic one, meaning that a one-dollar bill is tangible and anonymous, independent of the issuer once issued (excluding inflation). CBDCs and stablecoins are not. They are programmable money, which means they can be programmed to behave in accordance with the wishes of those who control them. That is why states are backing this transformation.
This process is ongoing. According to the Boston Consulting Group, in less than a decade, asset tokenization is expected to exceed $16 trillion and represent 10% of global GDP by 2030. The World Economic Forum was more optimistic; it predicted that the 10% threshold would be reached by 2027. Independent analysts, like Patrick Wood, expect this process to be accelerated due to a combination of factors.
In an article titled “An Assessment of the Accelerating Timeline for ‘You Will Own Nothing’,” Wood argues that AI acceleration, the technocratic capture of the legislature, the explosion of data centers, the Pax Silica, the federal wrapper around state law, and the BIS “whip” (meaning that the agreed adoption by a number of banking and financial institutions will force others to do the same) have created the conditions for the continuous acceleration of the tokenization process.
However, the tokenization of the economy does not happen in a vacuum. Every token and every tokenized transaction requires both data storage and computing power. Unlike a decentralized blockchain, such as Bitcoin, ensuring control of the tokenization process and full surveillance capabilities requires that storage and computing capabilities be accessible only by approved entities. That means that the ledger will be distributed through many data centers around the world, but only entities such as central banks, cleared private banks, or tech providers will be able to run the nodes that validate transactions.
When every single transaction in and across multiple countries—whether buying bread or paying for a house—is tokenized and settled almost instantaneously, the amount of information and computing power required reaches industrial data server capacity. But if you add to that the capability to control and monitor those processes through algorithmic intelligence, then what you might need in terms of computing power and data storage might be similar to the data center build-up that we are currently seeing.
I am no expert in either AI or tokenization, but I do know that the computing and storage requirements for each are not the same (with AI relying on GPUs versus tokenization relying on CPUs). However, what I’m trying to argue is not that the current AI infrastructure can be immediately used for tokenization, but that a fully tokenized economy combined with identity-linked transaction data, cross-border interoperability and machine-learning-based monitoring would require infrastructure closer to large-scale AI/data-analytics systems than to a simple payment ledger.
I asked an LLM (Gemini) about this—first, because I was curious to know the answer it would give, and second, because I wanted to know if it technically made sense beyond my coarse understanding. Of course, the response is not conclusive, and some readers might argue the opposite.
The question was simple: can the current build-up of data centers be used to store and process the tokenization of the economy? After providing information on the technical differences, the summary was conclusive:
“The data center boom is laying the physical tracks for a completely digital financial system. While today’s massive capital expenditure is largely driven by AI, the resulting global footprint of high-security, high-bandwidth server farms coul be useful to store, process, and secure the trillions of dollars expected to migrate into tokenized assets over the coming decade.”
Tokenization is a compelling use case. If current LLM models do not lead to AGI simply by multiplying computing power—which seems to be the logic behind the data center build-up—then AI companies will have a difficult time justifying the expenditure under current use cases. If that is the case, then the tokenization process, and its monitoring through costly, newly developed AI models, could use that same infrastructure.
I understand this is a hypothetical situation, but I think it is one that makes sense out of the seemingly senseless, massive data center build-up happening under the AI narrative. I am not necessarily saying that the tokenization and monitoring of the economy was the original intention for the data centers, but perhaps it is one that is considered if LLM models fail to deliver on their promises.
Another possible use case for the data centers, which is related to the programmability of CBDCs and stablecoins, is surveillance. Current modes of surveillance—digital, monetary, communicative, or locational—are compartmentalized. A private technology company might know your online behavior, another your banking history, and another might track your movements. However, if you pull all this information together and deploy algorithmic intelligence—which is what Palantir, among other companies, does—the possibilities grow exponentially. When you pair that with monitored financial behavior, then the full possibilities of programmable money become apparent.
It is possible that, as happened before during the telecom bubble, the use cases that will inherit the current data center infrastructure will be different from those it was allegedly built for. I don’t think it is difficult to see how the current AI bubble could bequeath a technology and infrastructure that will lead to a fundamentally different society, because the nature of what we call money will have changed and the surveillance capabilities of the state will have greatly increased.
The question then might be: could that not have been the purpose all along, and the rest just the necessary narrative to hide it?
Imagine trying to justify spending billions—many of which come from financial institutions and state funding through contracts—and placing a massive strain on energy and water resources to build up data centers at the current capacity, solely with the purpose of tokenizing the economy, implementing CBDCs, and enabling AI monitoring. There would be great resistance. However, if it is done with the promise of building some type of utopian future, then that resistance might be neutralized. At least for those who believe it.
Programmable money is the end of Free Speech, and the right to descent.
I know all this end-times stuff is tiresome, but you can’t deny it rhymes pretty well with the rest of our country’s wacko motivations.
The following is from the Bible Hub;
And this;
Those who warn us about the threat posed by programmable money are saying that one of the unstated reasons for the war with Iran is that those building the program cannot allow any economy of its size to operate outside their control because their surveillance, and thus their control would not be perfect, there would be ‘leaks’ and ‘blind spots’.
And here, all this time, I’ve been worrying about the a**holes using spare data center compute muscle to mine bitcoin.
Catherine Austin Fitts, and Alastair Crooke are doing heroic work warning about the dangers of programmable money, and how it relates to the war in Iran.
somehow I thought this was obvious all along
You are correct.
Tech has been coming for Wall Street for some time and Wall Street, fat on centuries of easy predation, has no idea of what’s about to hit it.
Thank you for a thoughtful analysis, Curro. I think you put your finger on something important when you point out that a rollout of CDBC or stablecoins would place the technocratic class, as you call them, in an essential and crucial position, just when they’re starting to look not only superfluous, but downright malignant. However, even if accomplished, such a plan would face the reality of the chasm between policy and practice, the abstract and the concrete, which finance and tech people see only from the former sides. Here is the American Society for Civil Engineers’ 2025 report card for American energy infrastructure.
Tokenization is possible, but one part of the AI buildup which is already true is it has replaced search largely on account of existing search being (likely deliberately) enshittified and supplemented with AI responses.
https://www.forbes.com/sites/torconstantino/2025/04/14/the-60-problem—how-ai-search-is-draining-your-traffic/
We’ve thus shifted from search engine to answer engine, and organic search traffic is completely disrupted.
Thus, a likely use for the AI data centre is to control the answer to any given question or query. Instead of Search Engine Optimization we now have Answer Engine Optimization controlled by state and corporation, which no one individual can influence.
When AI becomes the only way to internet search anything, it only takes an algorithm to control what anyone thinks about any given topic, unlike the previous search engine paradigm which was anarchic.
discipline, my friends.
i have only used the google search ai thing for questions about how to speak to linux mint to overcome some obstacle.
i find it more useful than plowing through a buncha, to me, gobbletygook in the various forii(which are, i assume, very useful if you actually use email,lol…and understand the systems and language to a degree in the first place.)
for everything else…ill skip the first 5 pages of returns, and then plow through as many as 30 more pages to find something akin to what i am after.
i simply wont use ai…and if i am somehow forced to, i’ll abandon the internet entirely.(spent most of my sojourn here on this planet without it , and with things like landlines and payphones and handwritten letters…and physical books(about 3-4 k of those right over there).)
what worries me is the thing from the other day, mentioned by Curro, regarding tokenisation of the title to this property…even without my knowing.
that bothers the hell out of me, and makes me want to clean my guns.
all else…well…my pension runs out in a little under a year and a half…and im already quite used to doing without a lot of things that most people i know reckon as sine qua non,lol.
and i am resilient and resourceful, and have spent 30 years making my side of the place into a swiss army knife of production ability.
if things keep falling apart like this, i expect to have lotsa help, too(built for that, as well)
only thing missing is independent power generation for the well and refrigeration and fans….that’ll give us a 20-30 year buffer to adapt.
and mark of the beast,lol…ill be using that in the feedstore if ai and tokenisation and all the rest ever comes up.
aside from the american penchant for irascibility(yes, much subdued these last 4 decades)…being forced into digital currency just might be the ticket for radicalisation….at least outside the big cities, where folks still have a sense of independence.
According to Ed Zitron only 5 Gigawatts of Data Centers have actually been built, with the others in various stages of planning or construction.
Take a hard look at the circular financing and what the Ramadan War is bringing, some of these centers may end up as empty shells, most will be abandoned before they get that far.
I’ve been through a few cycles, this one is done.
Where’s the effing Money going to come from once the World Economy hits the wall at 200MPH?
FedGov may try to bail out the Industry come Fall, but that won’t be politically feasible if things get anywhere near as bad as they look to.
Watching Oracle go bust will be entertaining…
You are on the right track looking past the dogshit fountains known as LLMs for the real reasons for lots of data centers. Financialization is part of it. Do a good think on how much compute is required to extract every step out of every Walmart video to analyze every “consumer’s” shopping behaviors. Oh, and how much compute is required to analyze all of the voice data from all of the telephones. Whether they are on a call or just listening all day every day.
Must confess … I am not sold on Crypto/BlockChain takeover … despite the obvious largesse of #coinBro political contributions. We’d need both sides of the aisle to be “down with it”, as it were, and I think there are too many ole school “sound finance” people still at the helm (Fed, etc) to basically usurp the governments currency monopoly. Perhaps with a changing of the guard, but hey … I could be wrong, such are these times.
But then yesterday, from an Ed Zitron repost, I saw that Mark ZuckerNerd opened his mouth (via X). Full article here, via CNBC.
So the hedge is … #MOARCloud. And both sides of the aisle are committed to data mining our personal stuff, so the government stands to end up being the ECCOLR: Enterprise Cloud Customer Of Last Resort.
Thanks for this thoughtful piece, Curro. As many know here I work in the field and do work directly with one of the data center… builders (I almost typed “manufacturers” because building a data center is more about heat and power management than simply the building shell). I confess I did not know anything about “tokenization” before reading this piece – for sure none of the organizations I work with are actually pursuing this use case, at least openly in their discussions with me.
I want to add a few items for consideration…
1. The data center builder I work with has been hitting YoY revenue and stock price numbers over the last two years. I will not admit to their name here. They have moved aggressively into the AI data center space in terms of their own modifications/expansions related to the AI – think a centralized power chain/cooling system optimized for AI workloads that can be controlled across the data center – it doesn’t have the same sort of value if it’s hundreds of thousands of unrelated nodes that don’t have the AI workloads controlled from the central unit. These data centers will be used in toto by single organizations. Note this is not ALL data centers currently under construction, just those by this builder/manufacturer.
2. I’m on record here stating that the most durable use case for AI and the build out is mass, real-time surveillance and processing. I maintain this position.
3. I travel a lot between the US and Europe and I have to move money between the countries. I have US and European bank accounts. The US accounts are all atrocious at moving money to Europe (3-5 days for transfers). It’s so bad I had to get a “financial services bank account” (Revolut/Wise) because there was just no way to reliably assume I could access the funds with a US debit or credit card and it was cheaper to use the financial services account to buy and hold euros and other currencies on a virtual card I can link to my phone for tap pay. So it is incredible to me if the US financial services are seriously considering tokenization as an alternate ledger when they aren’t considering bringing the US banking system into closer alignment with other countries for ease of global transactions, but maybe that’s a benefit to the powers that be rather than a drawback.
4. Currently there is a situation with the major cloud providers that most companies lease their compute from: take an example company, let’s say a UK defense contractor, that is required for their own security and compliance reasons to only provision compute within a specific data center and availability zone. These leases are acquired by the companies and they provision to their users based on internal request/need/budget. If the compute isn’t available in the approved zone, they can’t deploy elsewhere. Period. This is being repeated everywhere and I suspect is a major factor in the fundraising for building new data centers (“The customers are saying there isn’t enough capacity, so we need to build more!”)
5. Inference (the actual processing of the LLM response by the LLM before it is returned to the user) is the most time-consuming part of the operation and like anything that relies on a response to wait for some action to complete before returning to the user will provide a better experience the closer it is to the user making the request. The fancy tech phrase for the ideal solution is “edge inference” but this just means pushing more LLM deployments as wide as possible so there are more options for closer responses regardless of where the user is located. The trick here is that the frontier models are only deployed by the model providers, they aren’t self-hosted or deployed privately. So you end up in situations where Anthropic cannot guarantee GDPR compliance under load because their global endpoint that directs all requests behind their proxy to the fastest available LLM may not be in the EU. This is another factor contributing to the “We need to build more data centers!” frenzy.
6. The latest generation of “serious” Chinese open weight models are close to parity with the previous generation of frontier models. That means the next generation is probably going to be capable of full agentic multi-turn actions with an appropriate harness that can manage the tool calling. Once we hit that inflection point – and arguably we’re already there on the model side – the frontier models will begin to see the first real drops in usage because enough big organizations will begin diverting the majority of their bulk load to private deployments of these models which will allow them to have the agent workflows and only pay for the GPUs instead of variable token costs as is going on now with the frontier models. This is yet another driver for the “We need to build more data centers!” frenzy but it is far more muted because the unspoken part is the models are Chinese and nobody wants to be too open about how heavily they are being utilized.
Thanks.
I might be in the minority here but if you think that financial efficiency is worth anything then tokenization can help improve it. Not to say that it cant be abused but the digitization of legacy financial services allows for a similar level of control and abuse where people’s bank accounts can get frozen anyway. For privacy concerns that data is already all there it just makes it more organized.
Tokenization is basically a way for financial data to all be in the same transparent operating system so that it can all interact easier. This means real-time market and transaction data that could be used to allocate focus to what’s needed. An example I have is something that already exists in logging where the cutting machines are tied into real time commodity prices and are fed other data on shipping costs, and other financial inputs, and then the machine determines the optimal cutting path to split the log into the most profitable combination of products. If you consider profit being a metric for how to determine resource allocation then it makes sense from a resource efficiency lens. The problem is that profit isnt a permit metric and often doesn’t reflect externalities and such.
I don’t think tokenization is going away if you look at the scale of adoption and research into it whether it’s on the public or private side. In the US you have really power private interests looking to usurp the public money system with blockchain (Circle and BlackRock). In other places the government is looking at CBDC as a public blockchain based money.
Having a CBDC tied into all economic data of a country, you could run algorithmic monetary policy where interest rates aren’t dependent on a group of economists, but instead have real-time reactions to economic conditions with much more flexible credit tightening or loosening. It could even be multivariable where there are different interest rates for different applications of credit for different sectors to prevent overheating of specific areas of the economy.
Idk, food for thought, imo. I don’t think tokenization and blockchain is inherently bad, but there’s potential for it to be implemented oppressively or in a way that is beneficial, or both.
the broom handle i stuck a rubber thing on and use as my walking stick. i can take it anywhere…banks, the local ISD, the airport…and nobody asks a question.
but i can also beat a person to death with it.
i refrain from that…save in defense…because i aint that kind of person.
take a good look at the people pushing this– Karp and Altman, especially.
they are those kinds of people who cant be trusted with a broom handle.
they have said so, repeatedly, and in public.
they lust after a machine god, and will end humanity to get it.
they have, again, said so, repeatedly and in public.
were we a sane country, all those folks would be herded into an old missile silo, tossed a few packages of cookies and soda—perhaps a bag of angry raccoons– and sealed in with concrete.
we simply cannot continue treating such over the top antihumanism as just a quirk.
Why give the raccoons indigestion?
For such a scenario I question if the energy and water resources are there to meet the demands of all those data centers. Yeah, you can build a fleet of nuclear reactors to power them all up. Should only take several years to build them – assuming that they have the time. But water? You cannot print water. Are they going to tow icebergs down from the Arctic to melt for its water? Where is it going to come from? Are they going to prioritize water for those data centers over water for growing actual food with? Same with water used for manufacturing which needs enormous amounts of water. How about drinking water. That’s a biggie. Too many unanswered questions here and people like Altman and Karp would never think about such issues.
Ugh … for those without the HD memory … on this blog before and post the GFC all the feral ideological libertarians econ/tech were blabbing on about how with enough Terabits they could not only control/shape society – in their ideological dicta – but, model and predict the future before it happens. I mean like some feral Zi0nest manner=nothing can stop us. Groan at all the tech mags at the time doing the popular science mag shtick about it to foam the runway for the unwashed …
Ta Curro …
Perhaps I am wrong, but it strikes me that the move to tokenize real-world assets is akin to changing right of ownership to right of use. A bit like the difference between buying a paper book and a Kindle version (and, pointedly, owning an old Kindle that is right on the cusp of Amazon’s recent “turning off”.) If I own one hundred silver coins, I can (if I can find the right people) use them as a means of financial exchange: I can swap my silver for bread. If those coins are tokenized and placed on a ledger, then anyone who (somehow…) gets control over whether or that ledger records a change in ownership basically controls whether that silver is convertible into bread (other than on a black market.)
Excellent essay Curro.
Way back on June 20, 2023 the Bank for International Settlements featured an article entitled “Unified ledger could usher in ‘profound’ economic change.
It stated in part:
“The Bank for International Settlement laid out a vision today (June 20) of a “unified ledger” to sit at the heart of the monetary system, offering the potential for “profound” change in the economy;
The idea would unite public and private money on single ledger, overseen by the central bank. Tokenizing bank deposits and central bank money, as well as other assets, could allow “seamless integration”and programmability on one platform.”
Then in 2025 a special chapter of the BIS’s Annual Economic Report expanded on the idea of a ” unified ledger.”
The head of the BIS at that time Augustin Carstens stated:
“The next-generation monetary and financial system combine the time-tested principles of trust in money underpinned by central banks with the functionality unlocked by tokenization. This system is poised to deliver substantial improvement to current practices and to enable entirely new economic arrangements.
Realizing the full potential of the system requires bold action by central banks, which need to work in partnership with the private sector and other public authorities.”
Mostly for AI, of course.
1) data centers not finished when the bubble bursts will never be finished – with possibly a few exceptions, mostly due to politics.
We are talking about $650 billion in 2026. Who would have that money when the bubble bursts and stocks go down?
2) the AI data centers are not the data centers of old, are not full of blade computers one can re-purpose easily. They are full of extremely specialized stuff, optimized for one goal only.
These are not general-purpose systems. The optimal hardware for AI doesn’t share that important property of the PC.
3) we now see opposition to new AI data center (way overdue), which, while over-hyped, generate at least some output seen as useful.
We would see even more opposition for data centers not even having fig-leaves of general usefulness.
4) re-using the data centers for something you cannot turn off (asset tokens count as that) means you will have to carry the running costs and replacements costs of the data centers eternally.
I’m not sure if the AI data centers are the optimal long-time base for that.
5) if the bubble bursts we’ll see what really is useful.
it may well be that the productivity gain for software development (my area) is large enough to warrant AI for that purpose.
propaganda videos and other disinformation will surely stay, because that kind of stuff never stops.
Everything else… well… your guess is as good as mine.