Increasing Value From Compute
This is the full design paper behind the brief essay “Free-Market Money Redux.” The idea dates to a September 2012 patent filing and a June 2013 Wall Street Journal piece, “Free-Market Money, Courtesy of the Web.” Comments are open; I am looking for the flaws.
Cloud computing has become a fundamental economic resource and could serve as the basis for a new form of money with advantages over both fiat money and existing cryptocurrencies.
A less risky way to finance cloud infrastructure. Unlike dollar debt, a compute-backed liability is closely tied to the compute asset it finances.
Fuel for AI agents. For an agent, the currency is food. Holding it both stores value and hedges its largest operating cost, something no dollar-pegged stablecoin can do.
Backed by a productive asset. A cloud currency is a claim on compute, which produces real economic output every hour it runs, unlike fiat money or cryptocurrencies that are not redeemable into an underlying productive asset.
Designed to keep or increase its value. The currency is linked to a compute benchmark and built to let holders share in advances in cloud technology.
Proven economics. Airline mileage programs, though illiquid and routinely devalued, have been valued at $240 billion globally. The three largest US programs are valued at roughly the entire market capitalization of the airlines that issue them. A transferable, non-expiring claim on compute could be worth far more.
The essence of all business is the trading of one good or service for another. Historically this might mean the trading of the milk from my cow for the wheat in your field.
Over time, clay tokens, representing a claim of a certain amount of a commodity, came into being. These inventions served as an intermediate good, used to make trade easier.
Coins were a further evolution of money. These had intrinsic value because what they were made from had value, e.g., gold, silver, lead and copper. These coins had additional utility because their size and durability made storing and transporting wealth easier. Coins do not spoil or easily burn like grain. Coins do not die like livestock and can be easily moved unlike land.
Money was not exclusively metals. Some societies used shells or beads but metals, particularly gold and silver, came to dominate what was deemed money around the world. The common factors for different types of money were that it be widely accepted, durable, divisible, easily transported, a stable store of value and a unit of account.
The invention of money preceded official government money by at least a thousand years. Governments certainly took advantage of this creation, as it was an efficient way to collect taxes and pay soldiers. Governments also soon discovered they could cheat by mixing cheaper metals into coins, i.e., debasing the currency and using the force of law to require acceptance of those coins.
Productivity in the mining of metals roughly matched productivity in other parts of the economy. Therefore, the supply of precious metals would approximately change proportionately with the supply of other goods. This kept prices of those goods, in terms of a certain amount of metal, roughly stable over time.
Discoveries of new mines did spike the supply of precious metals from time to time, temporarily lowering the value of those metals as a jump in supply hit the market. As the new supply was not continuous, it was digested over time. From about 1800-1900 in the US, the price of other goods denominated in gold hardly changed. Using different endpoints would give different results but it’s clear that over long periods, gold-backed money did not suffer the same kind of systematic one-way depreciation that fiat money has experienced.
Coins produced by private mints (technically tokens since they were not covered by legal tender laws) competed with government mints by not debasing their coins, for instance having more gold bullion in them than similarly valued government issued coins but slightly less than the face value of the coin so they could profit. These private issuers, competing for trust, had strong competitive reasons to produce honest money, unlike monopoly sovereign issuers.
Paper money was a further evolution, representing a claim on some amount of metal, typically gold or silver. This creation was also a market response that allowed for the easier use of money in commerce and the easier transport of larger amounts of wealth. The value of the paper was determined not just by the amount of metal the paper had a claim on but also the perceived credit quality of the issuer. Essentially, whether the holder of this paper could reliably redeem it for the full amount of metal upon demand.
Private banks produced much of the circulating paper money in the US before the Civil War, though the Continental Congress, the Treasury and the federally chartered Banks of the United States also issued paper at various points. Banks profited through seigniorage, issuing bank notes as liabilities that typically paid no interest, while using their assets and shareholders’ capital to make loans or hold earning assets. Some governments severed the connection between paper money they issued and its convertibility into metal by creating legal tender laws that forced citizens to accept this paper money for payment of all debts. The government could then print paper money without any silver or gold backing it: “fiat money.” In 1775, the Continental Congress printed Continentals to finance the Revolution. These ended up becoming nearly worthless. Subsequently the US Constitution granted Congress the power to coin money and in 1792, the US Mint was created to mint coins of gold and silver.
In 1862, the Union issued “Greenbacks” to help finance the Civil War after its reserves of gold and silver coin were depleted. In 1864, Congress prohibited private coinage intended for use as current money, and subsequent taxes in 1865 effectively eliminated state bank notes, creating a federal near monopoly over circulating currency. Although the Constitution restricts states from making anything but gold and silver coin legal tender, modern federal law treats US coins and currency, including Federal Reserve notes, as legal tender. Courts have nevertheless upheld private scrip, notes or tokens when they are local in use or redeemable in goods or services rather than dollars. Legal tender laws also do not generally require private merchants to accept any particular form of payment for goods and services.
A number of companies have issued scrip, redeemable in company products or for use in company stores, but these schemes have always been fairly limited in scope. The most successful of these have been the airline mileage programs that date to 1979, with American’s AAdvantage program launching in 1981. While they created a new source of value for airlines, it is conceivable that these programs would have been even more lucrative if airlines had focused on building the utility and liquidity of these miles rather than focusing on short-term revenue considerations.
While airlines were successful in allowing customers to acquire miles in other ways than through flying or explicitly purchasing them, including through some hotels and certain credit card purchases, they have also made these miles increasingly harder to redeem. Airlines limited the seats and flights available for redemption and, most importantly, continually increased the number of miles needed for a seat. Worse, at least for consumers, miles can expire, so they are definitely not a stable store of value and have never been liquid enough to be considered a true currency.
The invention of Bitcoin has led to thousands of cryptos and will probably lead to many more. The common thread is reliance on blockchain technology that records every transaction with security provided by cryptographic algorithms. Blockchains allow for some degree of anonymity with varying levels of privacy and can be private or public.
Cryptos broadly fall into three categories. One category has as its goal to create something that will function as money and compete against government issued currency, for the most part, by algorithmically limiting its supply. Utility comes from near frictionless transportability and anonymity for holders. The downside to date is volatility that, at least for now, means it is not a stable store of value. Additionally, unlike precious metals that have industrial and ornamental value, there isn’t this type of inherent value in this category of crypto. It is more akin to fiat money minus the discretionary sovereign issuer. The hope of holders is that lasting monetary value exists or will come from scarcity, security, portability and sufficient infrastructure built to facilitate use, certainly possible but the experiment is ongoing.
The second category of cryptos consists of tokens designed to do a task, as with smart contracts. The third category represents title to or a claim on an asset such as a piece of real estate or digital art. In this category are stablecoins, designed to hold a fixed dollar value, backed by reserves of cash and/or short-dated government bonds. The two largest are Tether and USDC, with market caps of approximately $190 billion and $78 billion respectively as of May 2026,1 within a total stablecoin market that has reached roughly $318 billion, an all-time high that exceeds the foreign-exchange reserves of 95 countries, including the United Kingdom and Canada. In 2025, Circle, the manager of USDC, reported $2.64 billion2 in reserve income, essentially all seigniorage, while Tether reported a $10.1 billion profit, roughly $6 billion coming from seigniorage.3
The regulatory environment has also changed. The GENIUS Act, signed in July 2025, created the first federal US framework for payment stablecoins. It requires 100% liquid reserve backing and bars issuers from paying interest or yield to holders, with implementing regulations being developed through 2026. A number of central banks are also considering, piloting or already issuing central bank digital currencies, though most CBDCs are designed to complement or modernize existing payment systems rather than immediately replace cash.
While issuers of these instruments profit from seigniorage, the larger prize may be creating an alternative to existing payment systems: the data captured by a payment operator could easily be more valuable than all the transaction fees, particularly as competitive pressure drives those fees lower. Cloud computing infrastructure is an asset that could serve as the basis for a new form of money that has advantages over both fiat money and existing cryptos. Computing has become a fundamental resource to economic activity so it seems reasonable to think that it could provide backing to a new cloud currency that could be, but does not have to be, redeemed by the holder into a fixed amount of computing time on a particular cloud.
This design is available to any major cloud provider (CP): Amazon Web Services, Microsoft Azure, Google Cloud, Meta, Oracle Cloud Infrastructure, Alibaba Cloud, or any other provider operating at sufficient scale to credibly back a currency. The mechanics described below apply to any of them. Each has different competitive strengths and distribution advantages that would shape the specifics of implementation, but the core structure is provider-agnostic. While multiple companies could have successful currencies backed by their respective clouds, liquidity would likely concentrate in only a few with the best designs and strongest backing. First mover advantage could also matter, as early liquidity and acceptance may reinforce one another.
A cloud currency would augment existing gift certificates and reserved-capacity discount plans rather than replace them. Gift certificates and promotional credits are typically denominated in dollars or account credits, are not designed to circulate, and are subject to the issuer’s program terms. Even where consumer gift card laws limit expiration, the holder has a claim on a dollar amount, not on a fixed quantity of the underlying good or service. A cloud currency would be different because it would be denominated in compute and would have no expiration on redeemability.
Imagine getting a gift certificate from someone so you can buy some shirts. The gift certificate is redeemed for a fixed dollar amount, not a fixed number of shirts. If you hold onto the certificate and the price of shirts rises, you will get fewer shirts. Hold on to the certificate for too long or violate the program terms and you get nothing, a good deal for the producer of the certificate but not the holder. All things equal, the dollar value of a cloud currency will be higher with no expiration date than it would be with one.
While the cloud currency becomes a liability on the balance sheet when issued, unlike a typical liability, it requires no dollars to extinguish: redemption is settled in compute. Excess issuance of the currency would damage confidence in the redemption promise, likely imposing a cost by forcing the CP to add capacity, liquidity reserves or other backing.
The nature of this liability is also critically different from the dollar obligations being created to finance the explosion of capital expenditures by cloud providers and AI infrastructure companies. Estimates vary, but global data center capex is expected to exceed $1 trillion in 2026,4 with the largest US hyperscalers alone accounting for hundreds of billions of dollars. Some of this spending is funded by operating cash flow and equity. But much of the buildout is financed or contracted through debt, leases, vendor financing and long-term purchase commitments, obligations denominated in dollars, euros or some other fiat currency. These must be met regardless of whether the new infrastructure earns the revenue originally expected. Should this expansion lead to even a temporary glut of compute capacity, the potential for these obligations to become impaired, creating serious problems for both the CP and its creditors, is non-trivial. The railroads of the second half of the 19th century and the fiber-optic networks of the dot-com era left behind infrastructure of lasting value. Bondholders recovered only part of what they lent, and equity holders often nothing.
By contrast, a cloud currency liability is tied to the asset it helps finance. The holder has a claim on cloud capacity, not on the issuer’s future cash flows. This does not make the liability riskless, but it reduces the inherent mismatch between a financial obligation and the productive asset being built. If the price of benchmark compute goes up after the cloud currency is issued, the dollar value of the liability goes up too, but the value of the infrastructure asset backing it should also rise. If compute prices fall, the dollar value of the liability falls with them. The asset and liability are therefore far more closely matched than with ordinary debt, though not perfectly, because the price of the benchmark relative to other parts of the cloud will likely vary. Debt draws on those seeking yield, equity on those seeking capital appreciation, prepayments on compute users. A currency can draw on anyone seeking a store of value, expanding the universe of capital sources. It is an alternative to debt and equity financing, and a diversifier of both.
Higher profitability is not guaranteed by removing an expiration date on the cloud currency vs. a gift certificate but to the extent holders are confident they can convert their currency into a valued non-depreciating asset in the future, they are less incented to do so today. That would allow the CP to issue more cloud currency for a fixed amount of computing assets than otherwise, which would presumably lead to greater profit for the amount of currency issued.
The amount of extra capacity required is proportional to the amount of currency issued, its average redemption rate, and the variability in redemption rate. The longer the holding period, the lower the average redemption rate, and the more profitable issuance becomes as the amount of float captured increases. However, even if the currency is redeemed almost upon issuance, it could still be profitable if it attracts new customers to the CP and/or increases the percentage of time that boxes/cores are used.
Existing reserved capacity discount plans such as AWS Savings Plans, Azure Reserved Instances and Google Cloud Committed Use Discounts offer discounts on usage for a commitment to using a certain amount of resources over a given time period and/or a specific spend over that time period. Unlike a gift certificate, the user does get a fixed amount of time but there is still an expiration date for using the time commitment made. The appeal of these plans is strictly for bulk cloud users with predictable usage. A cloud currency, by offering a stable store of value as well as all the other properties of money, could be appealing to everyone. As with currencies convertible to precious metals, holders would only have to be confident in the underlying asset for the currency to be widely adopted. Holders were not typically interested in using the underlying asset, just having something they could use in commercial transactions while protecting the value of their wealth.
This would allow a CP to broaden its reach far beyond potential users of cloud computing and expand its attractiveness to existing users. One possible step in that direction is offering a small amount of cloud currency with various consumer subscriptions such as Amazon Prime, Microsoft 365, YouTube Premium, Google One, etc. and/or as part of a loyalty program.
In terms of the mechanics of issuance, let us call a unit of the cloud currency a Landauer5 or dauer for short, and let it be infinitely divisible. Although the CP could require a minimum redemption amount, there would be no expiration on redeemability. Bulk discount plans could still exist to offer discounts to users who do not mind an expiration date. An additional option to seed the use of dauers would be to give them to such bulk buyers of cloud services, analogous to earning airline miles when booking a flight.
CPs have a growing number of offerings. From these a single benchmark would be selected, most likely the most popular offering at inception. The holder could redeem a dauer on demand for an hour of time on the benchmark computing platform. Other offerings would be priced in terms of dauers, and those prices would presumably vary over time.
A dauer created for an hour of time on a PC from the 1980s would be nearly worthless today. One way to address this is to pay additional dauers to compensate holders for this effect. These are bonus dauers (not dollars or some other currency), a percentage of the amount of dauers held. This bonus rate will be set by the CP but informed by market forces. If the rate is too low, nobody will want to hold dauers unless the price of compute is increasing. Too high a rate and the CP will have to increase its cloud capabilities more than it would otherwise want, so that the cloud assets it has match the increasing number of dauers outstanding. One way to discover the market rate would be to auction dauers for the following day. The spread between spot dauers and next-day dauers would reveal a market price for short-dated dauer exposure, incorporating yield, liquidity, credit and redemption risk.
If the CP needs to limit redemption due to extreme demand or an outage, it can offer additional dauers to holders who voluntarily defer redemption. The bonus rate could vary by time of day, day of week, etc. Only in truly extraordinary situations would the CP have the right to halt redemption, perhaps if there were a physical or cyber attack on a substantial amount of cloud resources.
Alternatively, if the CP is trying to attract usage when demand is low, it could lower the conversion price in dauers. For instance, it could set the conversion rate from 1-3am on a particular date to 0.8 dauers instead of 1. This creates a sale without diminishing the value of a dauer.
Theoretically, during weak cloud demand, the CP could instead charge a penalty rate (a negative bonus rate) to incent conversion but introducing the concept that the holder could lose dauers is unlikely to promote confidence among current and prospective dauer buyers.
Raising the rate of conversion above 1 could be done but it is, even temporarily, a devaluation that could signal to holders that more permanent devaluations could occur. Credibility on conversion might, like virginity, be impossible to reclaim once lost, damaging the perception of dauers as a store of value. Unlike a government that can compel citizens through legal tender laws to use a currency, private companies must rely on trust in the continued quality of their products.
Once a sufficient number of dauers are created and accepted by the market, the dollar price per dauer should be set by the market. To increase the utility of dauers, holders should be able to buy/sell/transfer them with each other at will (unlike airline miles). While CPs should create a market to facilitate these transactions, and possibly even act as market makers in their own dauers, they should not stand in between all transactions on the market.
CPs could protect the value of dauers by increasing their bonus rate and/or buying them back, effectively decreasing their supply. If the dollar price of dauers got too high, the CP could do the reverse, sell more, and/or lower the bonus rate paid. Ultimately, the goal is to maximize the profitability of dauer production. Most likely, this is realized by having dauers be stable or perhaps slightly increasing in value with the bonus dauers received, relative to not just the dollar but to other goods and services in the economy.
Another tool for protecting dauer value is upgrading the benchmark every so often to a more valuable type of compute. Most likely this would be pre-announced by giving a certain future date for the transition or by the CP committing to upgrade if the price on the benchmark falls below some threshold for a certain number of days. Any forced benchmark change should occur only after long notice and under rules that guarantee or enhance the existing redemption promise. Protecting existing dauer holders should be paramount.
How much the CP leans on a bonus rate, supply control and benchmark upgrades will be a discovery process on how to maximize the value of its dauer production. Forward markets on the current and future benchmark could facilitate the discovery process. Competitive forces should mean that attempts to somehow devalue dauers will be punished by the market as happens with other products and services.
The market is already building supporting infrastructure. In the first half of 2026, CME Group,6 ICE,7 and the startup exchange Architect8 each announced compute futures, contracts tied to daily indices of GPU rental rates, with the US contracts pending regulatory review. Polymarket and Kalshi have hosted trading in GPU rental prices since early this year.9 These give compute standardized reference prices and, as of July 2026, its first forward curve, published by Kalshi. Such futures are different from dauers. They are strictly hedging instruments for compute users and speculators, cash-settled, never redeemed for actual compute, and store no value. But their arrival confirms the premise that compute has become standardized enough for the world’s largest exchanges to treat as a commodity, which makes it standardized enough to back a currency. What the exchanges have not built is a standardized spot instrument, a claim on the compute itself. That is the role dauers would fill.
Full disclosure of the total number of outstanding dauers will also be important. Showing the market that there’s enough compute capacity if every dauer were converted makes the promise of convertibility credible. The market could allow there to be more dauers outstanding than lifetime compute available if the CP has credible dedicated reserves to buy back these excess dauers if necessary.
A critical design element of the dauer system is ensuring it remains virtually arbitrage free. This means that the pricing of dauers, the bonus rate adjustments, and the conversion mechanism to compute time must be dynamically calibrated so that no market participant can exploit discrepancies for riskless profit.10 In practical terms, as technological improvements drive down the cost of compute time, the system’s automatic adjustments, particularly through dynamic bonus rate modifications, ensure that the value of a dauer consistently reflects its backing in real compute capacity. This arbitrage free condition not only mitigates short-term volatility but also reinforces market confidence, as it demonstrates the CP’s commitment to maintaining the integrity and stability of its dauer over time.
Business conditions and ultimately regulatory/legal considerations will rule, but dauers could be held anonymously on a blockchain or non-anonymously in an account with the CP. Over time, the CP might even license third-party “banks” to serve dauer holders.
Although the world is not heading this way, as a possibility, CPs could produce hard to counterfeit physical dauers with a unique serial number or watermark, analogous to cash. Increasing the versatility of dauers increases their utility.
Eventually the hope would be that dauers are widely accepted enough to be the basis for transactions away from the CP that created them and a substitute for dollars. For instance, the AWS dauer could be used as payment for goods and services on Amazon and elsewhere. Amazon could offer discounts for products paid for with its dauers to help seed this. Distribution through Prime and Amazon-wide acceptance would give non-compute users an immediate reason to acquire and retain dauers, while redemption into compute would provide the underlying value anchor. With continued success, it’s conceivable that dauers could be used in the full range of financial products such as loans and insurance.
Being able to use dauers in a variety of ways increases their utility, which should further lower the redemption rate, increasing the CP’s margins. The ability to extract fees and valuable data from payments and financial products and services is gravy and could easily dominate the value of just having a low redemption rate. Consumers would benefit from a cheaper payment system.
Airline miles and credit card points generally receive favorable tax treatment because they are usually considered rebates or promotional benefits rather than separate investment property. That is one reason consumers can redeem miles without calculating a gain or loss on every transaction. Dauers are likely to receive this same treatment, particularly if at inception the CPs seed the adoption of dauers by creating them for a loyalty program.
If dauers are successful and grow so that they become transferable, market-priced, bonus earning and widely spendable, tax authorities may be more likely to treat them like digital assets than like airline miles. Similarly, at some point dauers could be deemed securities. While these issues might create a hurdle to widespread adoption, getting to this point would be a sign that dauers have become a mature and valuable product.
While it is early days for agentic commerce, expectations are for substantial growth in coming years. Bain estimates that US agentic commerce could reach $300 billion to $500 billion by 2030, while Morgan Stanley estimates $190 billion to $385 billion of US e-commerce spending could be driven by agentic shoppers. McKinsey’s estimate is broader still, projecting that up to $1 trillion of US B2C retail revenue and $3 trillion to $5 trillion globally could be orchestrated by agentic commerce by 2030.11 Specialized payment protocols for AI agents are being built by Coinbase (x402), Google (Agent Payments Protocol), and the major payment networks, suggesting industry consensus that the agent-payment segment will be material.12
Dauers should be competitive on payment transaction speed and perhaps offer the lowest fees, particularly as they become part of the CP’s flywheel, but they also have perhaps the ultimate advantage that, from the AI agent perspective, they are food. Agents using dauers would be holding a claim on the compute needed to complete their tasks. By holding dauers, the agent reduces variability in the cost of its food rather than merely holding dollars and buying compute later at whatever price prevails. An agent with lower cost variability may thus have an economic advantage over agents not using dauers. As agents compete against each other, it is likely that only those that are profit maximizing will survive, just as with any other economic actor.
Few financial instruments have so directly unified a payment medium with the primary operating input of the user. An AI agent holding dauers holds a store of value and hedges its primary operating cost. This is categorically different from any dollar-pegged payment system, which by construction cannot protect the holder against changes in the price of what is being bought. Even if a central bank had a credible promise of keeping a basket of goods at a constant level, the fact is that the basket would have to be generic. Any individual’s basket is likely to differ.
The market seems ripe for this type of product. The explosion of cryptos coincided with the aftermath of the 2008 financial crisis. Since then, through mid-2023, interest rates around the world dropped so dramatically that adjusted for inflation, they were negative. In those conditions, putting money into a savings account guarantees that your wealth will lose purchasing power, i.e., diminish.
Although real interest rates (nominal rates minus inflation) are now positive, there is still a fear that central banks will allow for more inflation as a way for governments to deal with the explosion of debt since Covid. Many consumers and investors are desperately looking for alternatives to protect and grow their wealth. Cryptos, stocks, and real estate have generally prospered, but none of these qualifies as money due to their volatility and/or illiquidity. Precious metals have increased in value but are now strictly investments rather than circulating money. Federal law prohibits private gold or silver coins intended for use as money, and the government no longer issues currency redeemable in gold.
As of early June 2026, the estimated market value of above-ground gold was roughly $31 trillion,13 while Bitcoin’s market capitalization was roughly $1.4-1.5 trillion and the total crypto market was roughly $2.4 trillion.14 These figures move quickly, but they show the size of the markets investors already use as alternatives to fiat currencies. Airline mileage programs show how valuable a private currency-like system can become even when the currency is limited and imperfect. In 2025, American reported $6.2 billion in cash payments from co-branded credit-card and other partners, roughly four times its adjusted operating income, while Delta received $8.2 billion from American Express through its SkyMiles relationship.15 The March 2026 rankings from On Point Loyalty, a global loyalty consultant, valued Delta’s SkyMiles at $31.8 billion, American’s AAdvantage at $26.7 billion and United’s MileagePlus at $25.3 billion or about $84 billion in total.16 The combined market capitalization of these three airlines is around $100 billion and their enterprise value about $160 billion (as of 7/23/26), so the market values the programs at roughly as much as the flying business itself. The approximate total for all mileage programs globally is about $240 billion. These programs are valuable despite the fact that miles are often illiquid, hard to redeem at attractive rates and subject to devaluation.
There should be little doubt that Amazon, Microsoft, Google, Alibaba, Meta and Oracle could create a more generally useful product. They are far bigger, more agile and more successful, with better reputations and stronger credit ratings. If even one of these CPs were to launch its version of a dauer, it does not take much imagination to picture it creating much more economic value than airline miles.
The case for a cloud currency rests on reinforcing foundations. For consumers, it is money backed by a productive asset, protection against inflation and a cheaper way to make payments. For AI agents, it uniquely unifies the payment medium with the operational fuel, hedging the agent’s largest input cost. For the cloud providers, it is a way to extract more value from their infrastructure and to help finance what is already the largest private infrastructure cycle since the railroad boom, and may yet surpass it relative to the size of the economy. Any one of these alone would justify the design. Together they point to a category of money that does not yet exist, waiting on the handful of companies able to create it.
Full Disclosure: I first filed for patents on this design in September 2012, nine months before the original op-ed, and hold several granted patents on its mechanics. They exist to prove provenance, not to obstruct. Evaluation and pilots are free, terms for anyone building at scale will be fair and reasonable, and the IP will never be the obstacle. I have no token and no fund. I think someone should build this, and I am happy to help whoever does.
Notes
Inferred from the reserve holdings listed and prevailing 2025 short-term rates; Tether does not disclose the split between reserve interest and asset appreciation: https://assets.ctfassets.net/vyse88cgwfbl/20d2BoOAd28ZfkiQPYPjGN/4ed12f5939e1e06ee5aceccad4effbe4/ISAE_3000R_-_Opinion_Tether_International_Financial_Figure_31-12-2025.pdf
Dell’Oro Group, June 2026 forecast raised to more than $1 trillion: https://www.delloro.com/news/ai-infrastructure-buildouts-and-memory-cost-inflation-drove-data-center-capex-higher-in-1q-2026
Named for Rolf Landauer (1927-1999), the IBM physicist whose principle that “information is physical” holds that information always exists in a physical embodiment and is therefore governed by the laws of physics. From this he derived that erasing a single bit carries an irreducible minimum energy cost (Landauer’s principle, 1961), experimentally confirmed in 2012.
In practice the CP need not perform this calibration alone. When competing demands on its capital make buybacks unattractive, most likely in stressed markets, when dauers trade cheapest, private arbitrageurs can buy discounted dauers and redeem them for compute, enforcing the same floor at little cost to the CP, since redemption draws on capacity that would otherwise sit idle. Either way pricing discipline is enforced.
American Airlines Form 10-K for 2025 (cash payments from co-branded credit card and other partners of $6.2 billion): https://www.sec.gov/Archives/edgar/data/6201/000000620126000014/aal-20251231.htm
Delta Air Lines FY2025 results ($8.2 billion American Express remuneration): https://ir.delta.com/news/news-details/2026/Delta-Air-Lines-Announces-December-Quarter-and-Full-Year-2025-Financial-Results/default.aspx
The proposal has one central weakness from which nearly every other problem follows. Every mechanism meant to stabilize the currency ultimately depends on the discretion of the issuer. Hayek's proposal was competition among currencies whose issuers earned trust by maintaining stable rules. This proposal repeatedly asks holders to trust the issuer's judgment instead.
There is an older problem underneath the discretion problem. Smith observed that the qualities making money a good store of value are not the qualities making it a good medium of payment. A store should be hard to produce and worth withholding. A payment medium should be cheap to move and never worth holding back. The better an asset stores, the less it moves. One instrument cannot do both jobs, and every monetary system that worked separated them, gold in the vault and notes in circulation. The dauer is asked to be the store, the payment rail, the yield instrument, and the financing vehicle at once. The market will assign it one of those functions regardless of the design, and the float economics tell you which one the issuer is betting on. The payment story is the marketing.
The issuer controls the unit. Gold required no committee to decide what an ounce became next year. Here the CP chooses the benchmark, sets the bonus rate, determines when the benchmark is upgraded, controls issuance and buybacks, and prices every other service in dauers. Those are not implementation details. They are monetary policy.
The airline example demonstrates what issuers actually do with that discretion. Miles were continually devalued, redemptions restricted, and the programs became enormously profitable anyway. Competition did not discipline issuers into protecting holders. It disciplined them into maximizing the value of the float. The $240 billion valuation of airline loyalty programs is evidence of how much value issuers extracted from holders, not of how well the currencies served them.
The unit is administered rather than fixed. A dauer represents an hour on a benchmark whose economic value continually declines as technology improves. The proposal compensates for that decline through bonus payments determined by the issuer and benchmark upgrades chosen by the issuer. The purchasing power of the unit therefore depends on continuing administrative decisions rather than an objective standard. That is not commodity money. It is a managed currency.
The redemption mechanism exposes the same problem. When compute becomes scarce, precisely when holders most value redemption, the issuer proposes paying holders additional dauers to defer conversion. A bank paying depositors more not to withdraw is a bank in a run. The proposal does not eliminate liquidity risk. It manages it.
The liability matching improves the issuer's balance sheet, not the holder's. It is true that a compute-denominated liability matches compute assets better than dollar debt. But that transfers the infrastructure risk to currency holders. If compute prices collapse because the industry overbuilds, the dollar value of dauers collapses as well. Holders who believed they owned a stable store of value instead discover they financed part of the cloud provider's capital program without receiving the upside of equity ownership.
The trust argument collides with the business model. The paper offers protection from inflation and surveillance while proposing that the currency be issued by companies whose competitive advantage is surveillance. The essay's most revealing sentence is that "the data captured by a payment operator could easily be more valuable than all the transaction fees." That is not a side observation. It identifies the issuer's dominant economic incentive. The people most interested in an alternative monetary system are precisely those least likely to trust Amazon or Google to become their payment intermediary.
The redemption promise remains discretionary. The proposal permits temporary redemption halts, issuer-controlled benchmark changes, KYC accounts, and licensed third-party banks. All of these may be commercially sensible. None reduces reliance on a trusted intermediary. Satoshi's solution to the trusted third party was to remove it. This proposal replaces the state with a hyperscaler.
There is one omission that deserves attention. The proposal assumes compute remains the scarce economic resource. That is true today, but money lasts for decades. If energy, proprietary models, bandwidth, robotics, or some future bottleneck becomes more important than raw compute, the monetary anchor weakens with it. Gold did not have to remain the economy's most valuable industrial input to remain money. Compute might.
Gresham settles how this ends. Give people two monies and they spend the worse and hold the better, every time, without instruction. If dauers hold value, they will be hoarded and dollars will do the spending, and the circulating currency the paper envisions never circulates. If dauers are spent freely, it is because holders expect the bonus rate to lag the depreciation, which is the airline outcome. Either branch defeats the design. The only version that works is the layered one, a store at the base and payment instruments built on top, and that architecture already exists.
What is genuinely novel deserves recognition. The insight that an AI agent should hold claims on its own primary operating input is original and important. Likewise, the emergence of compute futures at CME, ICE, and Kalshi strongly supports the argument that compute has become a standardized commodity. Those observations point toward a future in which compute is widely hedged and financed through commodity markets. They do not point toward the largest seller of compute becoming the issuer of money.
The Hayek question the paper invokes is therefore the right one. Hayek imagined currencies competing for holders who were free to leave because issuers could not casually rewrite the rules. A currency whose unit, supply, yield, and redemption terms are all ultimately determined by one issuer, redeemable only into that issuer's product, is not denationalized money. It is monetary sovereignty relocated from the nation-state to the hyperscaler.
Could you begin by stating the objective? THEN describe the proposal and only THEN the advantages and disadvantages. This seems to start at the end.