AI and Monopoly
AI 与垄断
This paper reads AI, written as concentration in formation and its structural risks. The title uses "and" so as not to presuppose that this area is already monopolized; this paper judges neither that AI has already entered a stage of monopoly nor that it will not. AI divides into several layers: computing power and chips, cloud, models, data, distribution; each layer is read by the ways of examining in the earlier papers. Control in partnerships and investment is examined by what is actual: three kinds of material are examined, control, dependence and information; the presence of any one item does not by itself complete a conclusion of control (LD12). The United Kingdom Competition and Markets Authority's closure of its inquiries into four AI partnerships and investments, and the German Bundeskartellamt's conclusion on one of these partnerships, all stop at the jurisdictional threshold of merger review; each case is written by its own decision, and they are read neither as establishing that there is no control nor as finding that there is a problem. When the models of one firm or a few have become the entry point through which many people obtain information and make judgments, what they press on is the materials of judgment; checks increase with this depth; this proposition speaks of the entry point being held, not of machines replacing human judgment or creativity; when judgment is handed to machines, the bearing is not handed over with it (LD13). While concentration is still forming, checks must already be there: the normative reason is that checks are needed within the line, and that scale does not decide from what moment a construct must guard the line and be questioned; setting them while substitutes remain and dependence has not yet become entrenched keeps open the possibility of taking over and changing direction; none of this is a prediction of where an industry is going; research and public discussion are material and entry points for questioning, not final verdicts (LD14). Research notes several interrelated structural risks; there is dynamism now, and in segments where hardware, data and the like are concentrated, existing positions may become entrenched in the long run. The constructed example running through this paper is the four terms between a model company and its cloud partner; in the far future, after machines have taken over Base Layer work such as farming and drinking water, one household's day, with food, water and payments all passing through the same place. The far future is imagined, not predicted.
Writing Declaration: This paper was co-drafted with Claude (Anthropic). All intellectual decisions, framework design, and final editorial judgments were made by the author.
Introduction
In the spring of Year 27, four researchers who had come out of Oakgrove College founded Beacon in Apple City to make general-purpose models. In the autumn of the same year, Stone Granary, a computing and cloud company in the Highland State, invested in Beacon, taking about twenty percent. There were four terms:
Beacon's training and operation must use Stone Granary's cloud; Stone Granary supplies computing power at half the market price, and the discounted part counts as Stone Granary's investment.
Of the revenue from Beacon's models, Stone Granary takes fifteen percent, until it has recovered three times its investment.
Stone Granary holds one observer seat, without a vote, and may attend board meetings and obtain board materials.
Before Beacon gives another cloud company the right to use its models, it must first ask Stone Granary; within thirty days Stone Granary may take them first on the same terms.
In Year 28 Beacon Assistant went online; from Year 29 it charged companies and institutions, while individuals could use a basic tier free of charge. From Year 29 to Year 31, about seventy percent of Bluekiln's workshops used it to write product pages in foreign languages, keep accounts and answer letters from buyers elsewhere; four clinics in Apple City, and the township offices of Willow Ferry and Bluekiln, used it to issue queue numbers and answer questions.
In Year 31, about forty percent of Beacon's revenue came from sales through Stone Granary's cloud: Stone Granary had put Beacon's models into its cloud service catalogue.
Both faces are here. Stone Granary's money and half-price computing power let four researchers get started; Beacon Assistant opened a road to buyers elsewhere for the workshops. The four terms also tied Beacon to Stone Granary.
This paper asks: a minority stake, an observer without a vote, half-price computing power, the obligation to use Stone Granary's cloud, revenue sharing, a right of first refusal: taken together, where does control lie? Nominally Stone Granary has no vote; in fact, which of Beacon's decisions can it sway? Beacon opened roads for workshops and clinics: how is this face recorded, and can it offset the other?
There are several further questions: many people obtain information and make judgments through Beacon, and how is this depth examined; concentration is still forming, and when are checks set; and what will things look like after machines have taken over the work of the Base Layer?
§1 Layers
AI can be read in several layers.
Computing power and chips. The computing power needed to train and run models comes from chips and data centers. Paper 5 read Stone Granary's scheduling of computing power: seventy percent at one firm, "strategic partners first."
Cloud. Computing power is mostly rented through cloud. A position in the cloud layer stands at the entrance to the layer of models. Paper 5 established that control of one layer does not extend into the power to shut the door on the next (LD10).
Models. General-purpose models are the center of this paper: one model can be connected to many places, to write, keep accounts, issue queue numbers, choose goods, read filings.
Data. Training needs data, users leave data, and data goes back into improving models. Paper 4 read that data about a person is not material for placing his leaving under another's control (LD8).
Distribution. How models reach people: an assistant, a default entry in a market, a government's system for public business. Paper 4 read Ask-the-Way: the default entry on the market's front page, with only one assistant allowed to place orders.
Each layer can be read by the ways of examining in the earlier papers: control recognized by what is actual, depth reaching the materials of judgment, examined as it is now (Paper 1); the road of questioning and the efficacy of questioning (Paper 2); names (Paper 3); a chokepoint making rules, data, several firms side by side (Paper 4); one layer extended into the next, integration and separability (Paper 5). This paper does not redo these, and reads only three questions AI adds: control in partnerships and investment; the materials of judgment; concentration in formation and early checks. Last it reads the far future.
Research has looked at what these layers look like together. The United Kingdom Competition and Markets Authority's update paper "AI Foundation Models," published on 11 April 2024, notes three interrelated risks: firms that control critical inputs restricting access; firms with existing positions in consumer- or business-facing markets using them to distort people's choice of foundation model services; and partnerships among key players strengthening market power along the value chain. The same paper also acknowledges that such arrangements can promote competition. What research notes are risks, not findings; this paper cites it as research and does not treat it as a final verdict.
§2 Control in Partnerships and Investment
Back to Beacon and Stone Granary.
Nominally, Stone Granary holds only about twenty percent, and its observer has no vote. About twenty percent of the shares and one observer seat without a vote cannot, on their own, complete a conclusion of control; what rights these shares carry, and how the rest of the equity is distributed, the example does not give.
Political Economy has set down a sentence: nominal holding and mobilization come apart, and how much is held is not how much can be mobilized; the position that can mobilize accumulation can be one person, a board, or a coordinated arrangement (Political Economy Paper 1, E2). Control is examined by what is actual, not by name.
How is it examined by what is actual? When Political Economy asks about control within the line, it has three questions: who can activate it; what actual efficacy it has on reasons, actions, rules or the holding of office; how the party questioned cannot cancel it on its own (Political Economy Paper 3, E16). The original asks whether those acted on by an economic position have actual control over its relevant decisions; here it is changed to the effect of a partner or investor on another construct (extension, own undertaking): over which of Beacon's decisions does Stone Granary have actual efficacy?
Take the four terms apart into three kinds of material.
Control. About twenty percent of the equity; one observer seat, attending the board, without a vote; a right of first refusal: before giving its models to another cloud, Beacon must first ask Stone Granary. The first two, on their own, cannot yet show what they can sway; the third falls directly on one of Beacon's decisions: on whose cloud its models go. If Beacon wants to sell its models on another firm's cloud, it must first get past Stone Granary.
Dependence. Training and operation must use Stone Granary's cloud; computing power at half price, with the discount counted as investment; revenue sharing of fifteen percent, until three times is recovered. Every training run Beacon does and every copy it sells pass through Stone Granary. By Year 31, forty percent of revenue came from sales through Stone Granary's cloud. Dependence needs no votes: to switch clouds, Beacon would first have to think whether its training system could be moved, whether half-price computing power exists elsewhere, and how the forty percent of revenue would be made up. Paper 5 read what came later: by Year 36, Beacon's training system could run only on Stone Granary's platform.
Information. The observer can obtain board materials. Stone Granary also runs a cloud business itself and has other customers; through this seat it obtains Beacon's board materials, and what is in them the example does not record. The right to obtain materials alone is not yet enough to judge that it sways Beacon's decisions; it may also fall elsewhere: if the materials contain Beacon's plans and prices, Stone Granary would thereby have information unavailable to others in its dealings with other model companies.
Read the three kinds of material together, and what is to be asked is: when the interests of Beacon and Stone Granary diverge, say Beacon wants to put its models on another firm's cloud, or switch to a cheaper supplier of computing power, can Beacon decide for itself? The example does not record that Beacon ever tried. This paper reads only this: the nominal "no vote" does not reveal the right of first refusal, the dependence and the information written into the terms; how much effect they actually have must be examined separately.
The presence of any one item does not by itself complete a conclusion of control. Being required to use a firm's cloud is a term many start-ups have signed; an observer seat is a common arrangement in investment; revenue sharing is a kind of return. This paper does not judge them by their names, and reads only what effect they have together in one place; nor does it say that Stone Granary already controls Beacon in fact, since the example records only the terms.
Research has listed the features of such partnerships. On 17 January 2025, the United States Federal Trade Commission published a staff report on a 6(b) study of three partnerships between three cloud service providers and two developers of generative AI. The features the report notes include: significant equity stakes and some revenue sharing; varying degrees of consultation, control and exclusivity rights; a considerable part of the investment having to be spent on the partner's cloud services; discounted computing power; the possibility that such arrangements affect access to inputs such as computing power and engineering talent; contractual and technical switching costs; and the partner's access to sensitive technical and business information others cannot obtain. The materials the partners supplied run to September 2024, and public materials to January 2025 were used. What research notes are features and possible effects, not findings; this paper cites it as research. The four terms in the Kingdom State example are written as a combination of several features, and do not copy the terms or proportions of any real pair of companies.
In reality, one kind of regulatory decision already made shows how such partnerships are handled in one jurisdiction. The United Kingdom Competition and Markets Authority has made decisions closing its inquiries into four AI partnerships and investments in turn, all concluding that they did not constitute merger situations for it to review; the reasons are of three kinds, and each is written on its own.
Microsoft and OpenAI (5 March 2025). The Authority considered that Microsoft had had material influence over OpenAI's commercial policy since 2019; this time it found no evidence that this influence had risen to de facto control. Therefore, as to the change in the level of control under review, there was no relevant merger situation. The Authority stated expressly that this conclusion does not amount to finding that there are no competition concerns.
On the same partnership, the German Bundeskartellamt also reached a jurisdictional conclusion on 15 November 2023. It considered that Microsoft had acquired material competitive influence over OpenAI as early as 2019, or at the latest when the partnership was deepened in 2021; at that time OpenAI's activities in Germany were not yet substantial, becoming so only in 2023; the 2023 investment could not be considered to have deepened the existing link relevant to competition, so no notification was required. It also said that if Microsoft were to increase its influence on OpenAI in the future, it would have to be re-examined whether there is an obligation to notify.
Amazon and Anthropic (27 September 2024); Google and Anthropic (19 November 2024). In the former, neither of the two thresholds, turnover and share of supply, was met, and the Authority reached no conclusion on material influence; in the latter, the Authority did not consider that Google had acquired material influence over Anthropic through the partnership.
Microsoft and Mistral AI (17 May 2024). The Authority did not consider that Microsoft had acquired material influence over Mistral.
This series writes these as the two bodies have decided, and gives no verdict of its own. These decisions all stop at the jurisdictional threshold of merger review: on material influence and changes in the level of control, each case is written by its own decision; what they did not proceed to is a subsequent substantive assessment of competitive effects. They cannot be merged into a general reading that there is no control or no competition concern, nor read the other way as finding that there is a problem. The level of control the two bodies judged under their respective legal standards is not the kind of control this paper examines through three kinds of material either. Put together, they serve as an example for this proposition: several arrangements that look alike yield several different reasons at one body; arrangements that look alike must be read by their particular terms, their actual relations and the question under review.
LD12 Control in partnerships and investment is examined by what is actual. Three kinds of material are examined: control, dependence, information; they are examined by actual control, dependence and whether people are left unable to get around, not by name; the presence of any one item does not by itself complete a conclusion of control (per E2, "nominal holding and mobilization come apart", and E18; borrowing E16's way of examining actual efficacy: E16 originally asks whether those acted on by an economic position have actual control over its decisions; here it is changed to the effect of a partner or investor on another construct; extension, own undertaking).
§3 The Materials of Judgment
Paper 1 established that the depth of a chokepoint reaches not only prices and livelihoods but also information, attention and the materials of judgment (LD2). In AI, this depth is most evident.
Once a model is connected to many places, many people obtain information and form judgments through it. The examples already contain several such places.
A buyer on Crossways Market says what he wants, and Ask-the-Way chooses one or two items for him; he no longer looks at a list.
The Office of Model Affairs uses Beacon Assistant to read each firm's filings and write initial review summaries; Beacon's own filings are read by it too.
In the drought year, which district's water is cut first each time is decided by the dispatching model; the full calculation the City Council cannot read.
The daughter of one household works at a sorting station, and her shifts are set by Beacon's model.
In these places, what the model presses on is not a price, but where a judgment comes from: what to buy, what is wrong in a filing, whose water is cut first, who works on which day. When the models of one firm or a few become the entry point through which many people obtain information and make judgments, what they press on is the materials of judgment.
Political Economy has established that checks increase with the power-face, and that the power-face is compared by its particular effects: whom it converges, down to which dimension, how many must pass through it, how great the cost of walking away (Political Economy Paper 3, E14). This paper follows it: an entry point that presses on the materials of judgment goes deeper; checks increase with this depth (per E14; extension of the depth of E7; own undertaking).
One misreading must be guarded against. This proposition speaks of the entry point for obtaining information being held by one firm or a few; it does not say that machines replace human judgment or creativity. The buyer can still go and look at the list; the Office can still read for itself; the City Council can still review the rules. Human judgment and creativity are not replaced by an entry point; but people's road to the materials of judgment can be narrowed by an entry point. What this paper examines is the latter: to what degree the road is held by one firm or a few, and how the holder is questioned.
When judgment is handed to machines, the bearing is not handed over with it. The Theory of Government says that a tool holds no bearing, and that the bearing falls on the positions that hand the judgment out, set how it is to be used, and act on its results (Theory of Government Paper 1, G8). The Theory of the Company specialist paper says that where a company deploys a system, "the address of attribution is the construct that deploys it; the company cannot present an algorithm or AI as another address" (The Company in the Age of AI, ZN2); and that a judgment entrusted to a position requires that the position be able to judge, that the information needed can be obtained and understood, and that it can change the outcome (ibid., ZN1). The originals speak of governments and companies; here they are extended to the various positions that make judgments through a model's entry point (extension). Where the Office uses Beacon Assistant to read filings, the bearing is with the Office; where the operator uses a model to allocate water, the bearing is with the City Council and the Water Department, which set the goals of allocation and handed over the dispatching, and the operator holds its own accountability toward the construct that engaged it; where Crossways uses Ask-the-Way to choose goods for people, the bearing is with Crossways. The model itself is not an address.
LD13 Where AI presses on the materials of judgment, checks increase with the depth. When the models of one firm or a few have become the entry point through which many people obtain information and make judgments, what they press on is the materials of judgment; checks increase with this depth (per E14; extension of the depth of E7; own undertaking). This proposition speaks of the entry point for obtaining information being held by one firm or a few; it does not say that machines replace human judgment or creativity. When judgment is handed to machines, the bearing is not handed over with it (by way of The Company in the Age of AI, ZN1, ZN2; Theory of Government G8, which originally speaks of government; extension).
§4 Concentration in Formation
What does AI look like now?
Research has given a "mixed picture." The OECD report "Artificial Intelligence markets," published on 10 July 2026, notes that the foundation-model layer has been dynamic over the past three years; that structural risks remain at each layer of the value chain; and that in segments structurally concentrated in hardware, data and the like, existing positions "may become entrenched in the long run." What research notes are risks and possibilities, not findings.
This paper writes accordingly: concentration in formation, and its structural risks. It judges neither that AI has already entered a stage of monopoly nor that it will not.
"Examined as it is now" must be used in both directions here. In one direction: dynamism now does not answer for later; early road-opening does not serve as a later name. In Year 29 Beacon opened a road for the workshops, and that is true; later it wrote an interface that allowed only itself priority, used a format for Notes that other firms cannot read, and the head of the review panel came from its safety department; these are examined by their own effects and are not offset by that face of Year 29. In the other direction: possible entrenchment later does not condemn the present. When research says "may," it means may; this paper does not fill it out into any firm's already being entrenched.
In formation means that substitutes remain. The Open Model Group was still releasing (before Year 34); the other two model companies are still there; a workshop owner who wants to switch has to rebuild from scratch. This stretch of time is when the readings are still changing.
§5 Early Checks
While concentration is still forming, when are checks set?
One view says: wait until it crosses the line. It has not crossed the line yet, there is no monopoly yet, and setting checks now is punishing in advance, and may also press down what has not yet grown.
This paper does not take that view. Early checks do not wait for the line to be crossed.
The normative reason has two places upstream. Political Economy says: "Growth must be subject to checks; this is needed within the line, not only once the line is crossed" (Political Economy Paper 5, E27). Checks are not a punishment for crossing the line, but an ordinary arrangement for a position with a power-face. The same series also says: "Scale and power-face affect the form and strength of checks; they do not decide from what moment a construct must guard the line and be questioned" (Political Economy Paper 3, E14). While a position is still small, checks can be light and few; but its being questioned does not begin only on the day it grows large.
On the side of fact there is another kind of reason, which must be stated separately. Setting checks while substitutes remain and dependence has not yet become entrenched can keep open the possibility of taking over and changing direction; what is kept, how much is paid, and how much is saved later must be examined separately for the same arrangement. When the Office of Model Affairs was set up in Year 33 it had no testing equipment of its own; the laboratory was built only in Year 36. Had the laboratory been built three years earlier, the Office would have had an extra pair of eyes of its own in those three years of reading filings, looking at self-tests and granting approvals; the one result it found not to match might have come out sooner.
Neither of these reasons is a prediction. The normative reason says that checks ought to be there; the factual reason says that setting them early can keep open the possibility of taking over and changing direction, at a cost examined under E17. Neither says that early checks necessarily avoid monopoly, nor that without them things will necessarily end in monopoly. What an industry will become this paper does not predict.
Setting checks early also has costs, stated as they stand. Political Economy says that if checks affect the capacity to produce, to serve and to give conditions, the cost, the alternatives available and the reasons for the choice are to be stated as they stand (Political Economy Paper 3, E17); it also says that too many veto points can also leave a position unable to do anything (ibid.). Early checks must state their effects on innovation and on the capacity to produce and to supply; a set of rules that comes too early and weighs too heavy can press down what has not yet grown: the safety rules read in Paper 3 have this face to be examined. Early checks likewise must state their costs, and leave room to change them as things are now (under E17; under LD3).
A few arrangements that can be made at the rule layer are given as examples: keeping access, interoperability and preparation for separability, so that other firms' models can connect to the entry points of distribution and the interfaces between cloud and models are public; building in advance the reviewer's capacity to obtain information and test independently, like that laboratory; setting up in advance channels of supervision and questioning, so that those stopped by a gate can ask. These are all examples, and their form is examined by effect, cost and alternatives.
One more thing is already being done: research and public discussion. The several studies cited above, the Competition and Markets Authority's list of risks, the features of partnerships listed by the Federal Trade Commission, the OECD's mixed picture, are all questions raised while concentration is still forming. They are material and entry points for questioning; they are not final verdicts. That a study says "there are risks" is not a finding that anyone has crossed the line; but it points out early the places to be asked about.
LD14 Early checks do not wait for the line to be crossed. While concentration is still forming, checks must already be there. The normative reason is that checks are needed within the line (per E27, "Growth must be subject to checks; this is needed within the line, not only once the line is crossed"; E14: scale does not decide from what moment a construct must guard the line and be questioned). Setting them while substitutes remain and dependence has not yet become entrenched keeps open the possibility of taking over and changing direction; this is the side of fact that can be examined; the particular costs, and the effects on innovation and on the capacity to produce and to supply, are examined under E17; neither is a prediction of where an industry is going. Arrangements that can be made at the rule layer include, for example, keeping access, interoperability and preparation for separability, building in advance the reviewer's capacity to obtain information and test independently, and setting up in advance channels of supervision and questioning. Research and public discussion are material and entry points for questioning; they are not final verdicts.
§6 The Far Future: After Machines Have Taken Over the Work of the Base Layer
This section is an imagined scenario. What it writes is: supposing machines can do Base Layer work such as farming and drinking water, what form monopoly could take. "Machines can do all the work of the Base Layer" is the premise of the scenario, not the conclusion of an argument; this section is written as facts in a fictional world, not as a prediction of the real future. The far future is written only for the Base Layer: food, water, payments.
The earlier papers have already read several of these matters separately.
Food. From Year 40, Hundredcraft's farm machines are driven by Beacon's model, and Fieldworks, which the two set up together, operates about seventy percent of the Kingdom State's farmland by Year 50; the machines accept only the seed and fertilizer specifications Fieldworks designates; from Year 45, the state's farm machinery subsidies, grain purchasing and storage, and subsidies to secure the harvest in disaster years are mostly received and carried out through Fieldworks; at the autumn harvest of Year 47 an update went wrong, the combines in two counties stopped for nine days, and no one in the state could be found with the people and machines to take over; after the outage, the state gave Fieldworks a special "food security" fund, to be used for backup systems. Paper 5 read one layer extended into the next, and Paper 2 read being unable to fail.
Water. From Year 42, Apple City's Water Department handed its operation and maintenance to the operator set up jointly by Hundredcraft and Beacon; the department's engineers fell from forty to four; at the renewal in Year 50 the operating fee rose by about forty percent. In the drought of Year 53, whose water is cut each time is decided by the dispatching model. Paper 2 read rules written into programs, and Paper 7 reads taking over.
Payments. In Year 48 the Kingdom State set up a Basic Payment, issued through Beacon Assistant; the service windows were cut from a hundred and twenty to eighteen; in Year 52 the payment account was tied to the assistant account. Paper 4 read the manner of payment.
Each matter, taken on its own, has its effects and costs, and each has two faces: grain prices fell by thirty percent, deaths and injuries in the fields almost disappeared, leakage in the water mains fell from twenty percent to five, three villages in the north of the city were connected to water, and the payment was never late. This section brings them together in one place, set in Year 55.
One household's day. A household in Apple City's East Market: an elderly couple, their daughter, and a child at school.
The rice and vegetables in the morning come from fields Fieldworks operates, ordered through Ask-the-Way on Crossways Market and delivered; the only assistant that can place orders for her on Crossways Market is Ask-the-Way, and Ask-the-Way is driven by Beacon Assistant.
The water is supplied by the Apple City Water Department, whose operator is the one Hundredcraft and Beacon set up together. The water bill is paid in Beacon Assistant; the payment system uses Beacon's interface, and other firms' assistants face call limits and priority ordering when using it.
The two elderly parents' Basic Payment arrives through Beacon Assistant. The daughter works at one of Fieldworks' sorting stations, and her shifts are set by Beacon's model.
In the evening, the daughter wants to switch the family's assistant to another firm's. She works it out: the two parents' payment would have to be verified again at a service window, which takes about six weeks, and the nearest service window is two hours away by road; the family's three years of Notes cannot be taken along; with another firm's assistant, paying the water bill would be put further back in the queue; and with another firm's assistant, the only assistant that can place orders for her on Crossways would still be Ask-the-Way. She does not switch.
She writes a complaint in the assistant: "I want to switch, and I can't." The assistant replies: "Your request has been recorded; the relevant department will handle it." By the end of Year 55 there is no response.
What this household adds is only her day, her reckoning and her complaint; the other places are all matters the earlier papers have already read. Taken together, they show something that cannot be read from any one place on its own.
Whether there is a way out. The several burdens she reckons with add up: the payment must be verified again, Notes cannot be taken along, paying the water bill would be put further back. There is also one place that is not newly added by this switch: with a different assistant, the only one that can place orders for her on Crossways is still Ask-the-Way; this dependence does not disappear with this switch. Under the limit of Gate One, that there are terms, conditions, procedures or exit costs does not mean that exit has been sealed off (Political Economy Paper 4, E21). These items are each set by different positions: the payment and the service windows by the state, the water supply by the city government and the Water Department, the admission of assistants in the market by Crossways, while the operator, Fieldworks and Beacon each have effects of their own. That they all use one firm's model does not of itself prove that these controls have been acquired together by the model's provider. What is to be examined is the difficulty of leaving that these arrangements form together, and also, separately, the segment each position can change. She did not switch; this is not filled out into every structural channel having been sealed off.
Whom securing in the Base Layer is for. Food, water and payments are all in the Base Layer. Political Economy says that for a construct doing the work of the Base Layer, the institution must not only check but also secure this supply; "The object of securing is the supply, not that construct" (Political Economy Paper 3, E14a). The state's subsidies, purchasing and storage, and special fund flow out through Fieldworks; the service fee for the payment is paid to Beacon; the operating fee for the water is paid to the operator: is the money securing the supply of food, water and payments, or that firm? Are backup systems built inside Fieldworks still preparation for taking over? Paper 7 takes up this question.
Gate One. Political Economy says: accumulation is not converted into domination over people's Base Layer; what people need to sustain their lives is not converted into control that compresses their life-substrate, nor may money, housing or the manner of paying wages be used to place people's ability to leave, refuse or revoke a standing bond under another party's control (Political Economy Paper 4, E21). This household's payment arrives through one firm's assistant, its water bill is paid in that assistant, and its groceries are bought through Ask-the-Way, driven by the same firm's model; if she wants to leave that assistant, the payment must first be verified again. This reading falls close beside that gate. This paper does not say that the gate has already been crossed: the service windows are there, the road is there. What this paper says is that if several arrangements in the Base Layer come together in one place like this, that gate is the place to be examined item by item.
Whether questioning has effect. The complaint has a road: the assistant received it, and "the relevant department will handle it." By the end of the year there is no response. The Prequel said that being open to questioning does not cancel monopoly; Paper 2 said that the presence of the channel of questioning does not prove that questioning has effect. To which position able to change the outcome was this complaint delivered? The example does not record.
The same form as despotism. The Prequel read the same form through this household's day. Here Beacon is not identified as a despot; only the structures that limit exit and questioning are compared; that the payment arrives on time and the water supply has improved is another face to be recorded as it stands, and does not take the place of examining this structure. The same form lies in the structure, not in names, nor in whether the service is good or bad. Whether there is a way out and whether questioning has effect have not been fully examined for this household: the places to be asked about have been put side by side, and the judgment must be made place by place.
This household is what AI's monopoly over the Base Layer could look like. It brings the far future back to this paper's subject: once the models of one firm or a few drive the supply and issuing of food, water and payments, several places of control may come together in one place, reaching down to the life-substrate; whether they have come together, and how far, must be examined position by position.
Two suppositions are left for Paper 7: suppose the state took all of these into state ownership; suppose that around Year 40 preparations had been made to let it be able to fail. The first looks at what changes once the name has changed; the second at whether what is reduced is dependence or capacity.
This section does not prove that machines will do all the work of the Base Layer, nor that monopoly will necessarily follow then. It can prove nothing about the real future; it only writes one possible form clearly, so that the places to be asked about show themselves early.
Where This Paper Stops
This paper has established three propositions: control in partnerships and investment is examined by what is actual, through the three kinds of material, control, dependence and information, and no one item by itself completes a conclusion of control (LD12); where AI presses on the materials of judgment, checks increase with the depth, without saying that machines replace human judgment or creativity, and the bearing is not handed over (LD13); early checks do not wait for the line to be crossed, the normative reason and the factual reason are kept apart, and neither is a prediction (LD14).
What it has not shown: that the AI industry has already entered a stage of monopoly; that any AI company has crossed the line; that early checks necessarily avoid monopoly; that controlling an entry point for information proves that creativity has been replaced; the far future is imagined: it does not prove that machines will do all the work of the Base Layer, nor that monopoly will necessarily follow then. Stone Granary has not been judged to control Beacon in fact; the Competition and Markets Authority's four decisions closing its inquiries and the Bundeskartellamt's conclusion have been read neither as establishing that there is no control nor as finding that there is a problem.
Handed to Paper 7: all three are needed, and letting it be able to fail; the household's two suppositions; public support in the Base Layer and its grounds.
The remainders it leaves: how the list of three kinds of material is carried into the rules of merger review and partnership review lies at the rule layer; how deep counts as deep in pressing on the materials of judgment is examined place by place at the judgment layer; how concentration in formation will go later this paper does not predict, and examines as it is now.
The Three Propositions of This Paper
LD12 Control in partnerships and investment is examined by what is actual. Three kinds of material are examined: control, dependence, information; they are examined by actual control, dependence and whether people are left unable to get around, not by name; the presence of any one item does not by itself complete a conclusion of control (per E2, "nominal holding and mobilization come apart", and E18; borrowing E16's way of examining actual efficacy: E16 originally asks whether those acted on by an economic position have actual control over its decisions; here it is changed to the effect of a partner or investor on another construct; extension, own undertaking).
LD13 Where AI presses on the materials of judgment, checks increase with the depth. When the models of one firm or a few have become the entry point through which many people obtain information and make judgments, what they press on is the materials of judgment; checks increase with this depth (per E14; extension of the depth of E7; own undertaking). This proposition speaks of the entry point for obtaining information being held by one firm or a few; it does not say that machines replace human judgment or creativity. When judgment is handed to machines, the bearing is not handed over with it (by way of The Company in the Age of AI, ZN1, ZN2; Theory of Government G8, which originally speaks of government; extension).
LD14 Early checks do not wait for the line to be crossed. While concentration is still forming, checks must already be there. The normative reason is that checks are needed within the line (per E27, "Growth must be subject to checks; this is needed within the line, not only once the line is crossed"; E14: scale does not decide from what moment a construct must guard the line and be questioned). Setting them while substitutes remain and dependence has not yet become entrenched keeps open the possibility of taking over and changing direction; this is the side of fact that can be examined; the particular costs, and the effects on innovation and on the capacity to produce and to supply, are examined under E17; neither is a prediction of where an industry is going. Arrangements that can be made at the rule layer include, for example, keeping access, interoperability and preparation for separability, building in advance the reviewer's capacity to obtain information and test independently, and setting up in advance channels of supervision and questioning. Research and public discussion are material and entry points for questioning; they are not final verdicts.
References
All SAE works below are cited by concept DOI, which always resolves to the latest version.
I · This system
Qin, H. (2026). Judgment (SAE Theory of Government, Paper 1) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23412130
Qin, H. (2026). Accumulation (SAE Political Economy, Paper 1) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23874138
Qin, H. (2026). Chokepoints (SAE Political Economy, Paper 2) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23874140
Qin, H. (2026). Checks (SAE Political Economy, Paper 3) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23874142
Qin, H. (2026). Conversion (SAE Political Economy, Paper 4) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23874146
Qin, H. (2026). Growth (SAE Political Economy, Paper 5) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23874150
Qin, H. (2026). The Company in the Age of AI (SAE Theory of the Company, Specialist Paper) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24117816
II · This series
Qin, H. (2026). What a Monopoly Is Not (SAE Theory of Monopoly, Prequel) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24377536
Qin, H. (2026). Monopoly: Formation (SAE Theory of Monopoly, Paper 1) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24377542
Qin, H. (2026). Monopoly: Exemption from Questioning (SAE Theory of Monopoly, Paper 2) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24377546
Qin, H. (2026). Monopoly: Names (SAE Theory of Monopoly, Paper 3) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24377553
Qin, H. (2026). The Digital and Monopoly (SAE Theory of Monopoly, Paper 4) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24377556
Qin, H. (2026). Technology and Monopoly (SAE Theory of Monopoly, Paper 5) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24377558
Qin, H. (2026). Monopoly: Remedies (SAE Theory of Monopoly, Paper 7) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.24377564
III · Regulatory decisions
Competition and Markets Authority. Microsoft / OpenAI partnership merger inquiry, decision of 5 March 2025, paras 8, 14–15, 67–68. (Full text of the decision; tier A.)
Bundeskartellamt. Microsoft / OpenAI: no merger control notification required, press release, 15 November 2023. (Tier A.)
Competition and Markets Authority. Amazon / Anthropic partnership merger inquiry, decision of 27 September 2024; Alphabet / Anthropic partnership merger inquiry, decision of 19 November 2024. (Decision summaries and the Authority's case pages, tier A; the full texts of the decisions were published on 17 October and 24 December 2024 respectively.)
Competition and Markets Authority. Microsoft / Mistral AI partnership merger inquiry, decision of 17 May 2024. (Full text of the decision; tier A.)
IV · Research
Federal Trade Commission. (2025, January 17). AI Partnerships & Investments 6(b) Study, staff report. (Press release tier A; the full report yet to be read through.)
Competition and Markets Authority. (2024, April 11). AI Foundation Models: Update paper. (Tier A, government press release page.)
OECD. (2026, July 10). Artificial Intelligence markets: Recent developments and competition issues. (Tier B, summary page; the full text yet to be read.)
Interfaces
Main upstream: Political Economy Paper 1, E2; Paper 3, E14, E16 (extension, own undertaking), E17, E18, E14a; Paper 5, E27; Paper 4, E21; The Company in the Age of AI, ZN1, ZN2; Theory of Government Paper 1, G8 (extension).
Also used: this series' Paper 1, LD2, LD3; Paper 2 (capability dependence; rules written into programs; being unable to fail); Paper 3 (in the name of safety); Paper 4 (data; the manner of payment); Paper 5 (scheduling of computing power; one layer extended into the next; machines in the fields).
Examples: the four terms between Beacon and Stone Granary; Beacon Assistant in workshops, clinics and township offices; Crossways Ask-the-Way, the Office of Model Affairs and the allocation of water in a drought year (pointed to); the Oakgrove Testing Laboratory (pointed to); machines in the fields, water and the Basic Payment (far future, imagined, pointed to); one household's day (far future, imagined). All are constructed examples. Real-world: the United Kingdom Competition and Markets Authority's closure of its inquiries into four AI partnerships and investments; the German Bundeskartellamt's conclusion on one of these partnerships. Research: the United States Federal Trade Commission's 6(b) study, the Competition and Markets Authority's 2024 update paper, the OECD's 2026 report.
This series: the Prequel (the same form as despotism; one household's day first used); Paper 7 (all three are needed; letting it be able to fail; the two suppositions).
Handed to Paper 7: diagnosis and the three areas have been read; the remedies.