A Comprehensive Analysis of Investment, Financial Structure, and the Political Economy of Artificial Intelligence in the Age of Consciousness

1. The Overall Scale of Investment, Partnerships, and Mergers

Artificial intelligence has moved from the laboratory into industrialization: chips, data centers, networks, energy, foundation models, and enterprise software now form a single production chain. This transformation is not merely a new market technology; it is a productive force reshaping production, the distribution of knowledge, the organization of work, and therefore the possibility of a new social order.

Capital Expenditure (CapEx) by the Technology Hyperscalers

• Scale of investment: In 2026, the five major infrastructure players—Amazon, Alphabet, Microsoft, Meta, and Oracle—have plans and estimates for capital spending measured in the hundreds of billions of dollars. Exact figures change during the year, and not all of this spending is exclusively for AI; nevertheless, data centers, chips, networks, and electricity are the main drivers of its growth.
• Infrastructure concentration: Rather than repeat an unverified fixed claim of “75 percent,” it is more accurate to say that a large share of the increase concerns Nvidia and AMD accelerators, proprietary chips from Google, Amazon, and Microsoft, data centers, networks, and long-term energy contracts. This concentration compresses economic power into the bottlenecks of chips, cloud computing, and electricity.

Circular Investment and Strategic Partnerships

• Microsoft and OpenAI: Their relationship as investor, cloud supplier, and commercial partner places capital, computing capacity, and model distribution within an interconnected circuit. The financial details are contractual and variable, so the relationship should not be described simply as “receiving a share of profits.”
• Amazon, Google, and Anthropic: By the formally announced stage, Amazon had invested four billion dollars in Anthropic and made AWS its primary cloud provider; the partnership also involved Trainium and Inferentia. Google is likewise an investor and cloud partner. These transactions exemplify “circular investment”: capital flows to a model company, while part of that company’s demand returns to the investor’s cloud and chip services.

2. Revenue Models of Artificial Intelligence Companies

AI companies have organized their revenue streams around four principal channels:

• Consumer subscriptions (B2C): Tiered plans for access to models, coding, research, and content creation. Prices and usage caps change frequently; the more important analytical measure is the relationship between subscription price, the real cost of inference, and capacity constraints.
• API access (usage-based or token pricing): A principal source of revenue from developers and companies, based on the number of input and output tokens.
• Enterprise and on-premises contracts (B2E): Customized models, stronger data security, and commitments not to use company data for training.
• Licensing and infrastructure integration: The sale of tools integrated into enterprise software, such as Microsoft 365 Copilot.

3. Financial Analysis: Revenue and Profitability of Leading Companies

4. The Cost-Structure Challenge: Training Versus Inference

5. The Economics of Inference and the Effect of Prompt Caching

API cost is calculated according to the following formula:

Total Cost = (Tokens_Input × P_Input) + (Tokens_Cached × P_Cached) + (Tokens_Output × P_Output)

Comparative Example: Figures Depend on the Model, Date, and Usage Pattern

6. Capitalist Logic and a Political-Economy Critique of Artificial Intelligence

The Unprecedented Intensification of Private Ownership and Monopoly Concentration

In the Age of Consciousness, ownership no longer covers only the “physical means of production” such as factories and land. The “cognitive infrastructure and accumulated knowledge of humanity” have also become the private property of a small number of hyperscale corporations.

• Appropriation of a social asset: Large language models rest on the historical accumulation of knowledge, language, art, code, public data, and human labor. Not all data is necessarily public or free of rights, and its origins are not uniform. Yet the central divide remains: social and collective inputs can be transformed into private, closed, and monopolized outputs.
• Commodifying access to cognitive capacity: A token is a technical unit for processing text, not consciousness itself. Yet when access to models, data, and computation is available only through private payment and licensing, part of the capacity to produce knowledge, analysis, and code becomes a measurable and saleable commodity.
• The emergence of multilateral monopoly: Leading-model development depends on a restricted network including Nvidia, AMD, TSMC, Broadcom, ASML, and major cloud providers such as AWS, Azure, Google Cloud, and Oracle. “Digital feudalism” may serve as a metaphor for dependence, but the more precise concept is multilateral monopoly and private control over infrastructure bottlenecks.

Capitalism’s Structural Contradiction: Crushing Costs and Deepening Crises

• Pressure on returns to capital: Injecting hundreds of billions of dollars into chips, data centers, and energy raises fixed costs and heightens the risk of idle capacity or delayed returns. This fact alone does not prove a “falling rate of profit,” but it pushes companies to expand markets, lock in customers, and extract infrastructure rents.
• Overproduction and declining purchasing power: Capitalism uses AI to replace or weaken human labor in order to reduce production costs. But machines do not purchase AI goods or services; people and workers remain the final buyers. Job displacement and wage suppression reduce society’s aggregate purchasing power and expose the market to a crisis in realizing profit.
• Energy and environmental pressure: Data centers consumed about 415 terawatt-hours in 2024, close to 1.5% of global electricity use, and the International Energy Agency projects consumption near 945 terawatt-hours in 2030. The issue is not only the amount of electricity; location, energy source, grid pressure, water, and the distribution of costs also matter.

7. Operational Mechanisms for the Social Ownership of Artificial Intelligence

If artificial intelligence is removed from private monopoly, simply transferring it to the state is not enough. The goal must be “democratic social ownership”: society should participate in setting objectives and allocating resources through elected institutions, workplace and neighborhood councils, public oversight, freedom of criticism, and independent auditing. Success should be measured by well-being, consciousness, freedom, and human development—not merely profit or administrative power.

A Model for Social Allocation and Access

• Universal Basic Compute as a citizenship right: Every citizen receives a defined monthly amount of free computing capacity and tokens for educational, research, and creative needs.
• Allocation by social utility: Projects in medicine, public health, education, and the environment receive higher priority within approved capacity and transparent social criteria. “Unlimited priority” is neither technically feasible nor democratic.
• Transparent pricing based on social cost: Excess commercial consumption should be priced to reflect energy, depreciation, labor, security, development, compensation for environmental effects, and reserves for future investment. Promising a fixed reduction of “up to 90 percent” is not credible without valid operating data.

8. The Chain of Power: Which Companies Control What?

To understand the political economy of AI, we should not call every actor an “AI company.” The chain has several layers, and power accumulates at different bottlenecks.

Chip-manufacturing equipment: ASML builds advanced lithography machines; TSMC performs contract manufacturing for many leading chips; firms such as SK Hynix, Samsung, and Micron supply high-bandwidth memory. Disruption at this layer can constrain the entire chain.

Accelerator and network design: Nvidia has built an integrated ecosystem around GPUs, networking, and CUDA; AMD is an important accelerator competitor; and Broadcom plays a role in custom chips and networking. In fiscal year 2026, Nvidia reported about $215.9 billion in total revenue, roughly $194 billion in data-center revenue, and a gross margin of about 71.1%. These figures indicate that a large share of value is currently concentrated in the hardware and software-platform bottleneck. [Source: Nvidia financial report]

Cloud and data centers: AWS, Microsoft Azure, Google Cloud, and Oracle Cloud control computing capacity, storage, networking, and enterprise distribution. Companies such as CoreWeave have also grown as specialist GPU providers. This layer does more than host models: through contracts, cloud credits, and access to chips, it determines who can build models and at what cost.

Models and applications: OpenAI, Anthropic, Google DeepMind, Meta, xAI, Mistral, and Chinese model developers compete at the model layer, but many depend on the upstream layers for training and inference. OpenAI stated that its annualized revenue rose from $2 billion in 2023 to more than $20 billion in 2025. Reuters reported that Anthropic exceeded a $65 billion annualized run rate by the end of July 2026. These are annualized run rates, not necessarily audited revenue for a complete year. [Source: official OpenAI report] [Source: Reuters on Anthropic]

Distribution and customers: Microsoft 365, GitHub, Google Workspace, Salesforce, ServiceNow, Adobe, and other software platforms bring AI into everyday workflows. In its third-quarter fiscal 2026 report, Microsoft said its AI business had surpassed a $37 billion annualized revenue run rate, up 123% year over year. [Source: official Microsoft report]

The political conclusion is clear: competition among models does not necessarily create genuine competition in infrastructure. Dozens of apparently different products may ultimately depend on a few chipmakers, cloud providers, and distribution networks. Antitrust policy must therefore examine ownership, cloud contracts, chip access, data portability, and customer exit costs together.

9. Labor, Inequality, and the Problem of Purchasing Power

The International Labour Organization estimates that one in four workers worldwide is employed in an occupation exposed to generative AI to some degree, while 3.3% of global employment is in the highest exposure category. Exposure is higher in high-income countries, and in the highest category women face greater occupational change than men. The report’s main conclusion is not that all these jobs will disappear; transformation of tasks is more likely than complete job elimination. [Source: ILO report]

This distinction matters. The danger is not only direct unemployment. If the owner of intelligent tools can produce the same output with fewer workers, then without institutions that enable participation in power, productivity gains flow into profits and share values; worker surveillance intensifies; skills are separated from labor and embedded in corporate systems; and wage earners’ bargaining power declines. Even if total employment remains stable, job quality, professional autonomy, and labor’s share of income can fall.

Under the logic of a new social order, rising productivity should become shorter working time, higher social wages, lifelong education, and expanded public services. The central question should not be “How many jobs will disappear?” but rather: Who owns productivity, who decides how technology is used, and how will the fruits of liberated time be shared across society?

10. Energy, Water, and Public Infrastructure

The International Energy Agency estimates that data centers used about 415 terawatt-hours of electricity in 2024—nearly 1.5% of global consumption—and that the figure may reach about 945 terawatt-hours by 2030. An AI-focused data center can consume electricity comparable to approximately one hundred thousand households, while the largest projects under construction have several times that capacity. [Source: IEA report]

These figures should not lead to the simplistic conclusion that “AI is inherently anti-environmental.” AI can improve power grids, materials design, transportation, and energy efficiency. Governance is decisive: Which models are trained and for what social value? What is the source of their electricity? Who pays for grid reinforcement? How is water consumption reported? Does the local community have the right to reject or alter a project?

A clear example of the AI-energy connection is Google’s announced plan to invest about $15.1 billion in Finnish AI infrastructure over two years, together with a 22-year agreement to purchase part of the output of a nuclear power plant. Such contracts show that future cognitive infrastructure cannot be imagined apart from energy policy, territorial planning, and public oversight. [Source: Reuters on Google’s investment in Finland]

11. From Public Ownership to Social Ownership and Participation in Power

State ownership can limit private monopoly, but it is not by itself social ownership. If data, models, and data centers are controlled by an opaque state apparatus while citizens, workers, and local communities lack rights of decision and objection, only the form of concentration has changed. The more precise concept, extending the argument of A New Social Order in the Age of Consciousness, is the following:

Social ownership means linking rights of control and benefits among society’s institutions through participation in power: an elected national institution approves broad goals and budgets; workplace councils decide how AI is used in production; neighborhood and municipal councils oversee local effects on data centers, energy, and public services; universities and professional associations assess scientific and safety standards; and courts, media, and independent auditors protect the capacity to challenge decisions and expose wrongdoing.

This structure rests on basic rights: equal access to foundational computing capacity; privacy and protection from pervasive surveillance; explanation and appeal regarding automated decisions; data portability and provider choice; worker participation before automation; independent research into high-risk models; and society’s right to share in the economic return from common data and infrastructure.

The key term is therefore not a top-down “distribution of power,” but participation in power. Citizens are not passive consumers of a private or state system; they are partners in decisions about objectives, budgets, data, evaluation, and the sharing of benefits.

12. Data and Knowledge as a Social Commons

Foundation models cannot exist without language, knowledge, experience, and data produced throughout society. Yet the social origin of knowledge does not invalidate copyright, privacy, or consent. A new order must distinguish among four categories: personal and sensitive data, whose use without consent should be prohibited by default; copyrighted works, which require permission and compensation; publicly funded public data, which should serve the public interest as far as possible; and communities’ collective data, whose rules of use must be determined with those communities’ participation.

A practical proposal is to create “social data trusts.” Acting for their members, these institutions would negotiate access, anonymization, licensing, revenue shares, and the right to exit. Alongside them, a “public knowledge and compute fund” could be financed through taxes on infrastructure rents, licensing fees for public data, and returns on socially held equity, and could allocate resources to open models, low-resource languages, education, health, and basic research.

Openness is not binary. Open weights, documented data, evaluation code, energy reports, model cards, and auditability each represent a level of transparency. Extremely high-risk models may not require fully public weights, but they must be subject to multi-institutional oversight, independent testing, and public reporting. Transparency without real oversight capacity is insufficient.

13. Democratic Intelligent Planning

Artificial intelligence creates a new capacity for planning: the simultaneous analysis of inventories, demand, productive capacity, energy, transportation, health needs, and environmental effects. But algorithms must not replace politics and social choice. Intelligent planning should operate at three levels:

At the national level, measurable goals are set for housing, health, education, clean energy, food security, and reduced working time. Budgets and environmental constraints must be public and open to contestation.

At the sectoral and regional level, actual capacities and constraints are recorded. Trade unions, professional councils, cooperatives, universities, municipalities, and enterprises propose implementation options.

At the workplace and neighborhood level, information close to lived reality and feedback on consequences are provided. Final decisions must include channels for appeal, review, and correction.

In this system, AI is a tool for forecasting and comparing scenarios, not a ruler. A model should show the likely effects on waiting times, employment, energy, inequality, and quality when resources are allocated to option A or B. Values—such as prioritizing health over advertising or shorter working hours over private profit—must come from democratic processes, not from engineers’ hidden objective functions.

14. Sharing the Gains of Productivity

To prevent AI from producing a demand crisis and further concentration of wealth, four mechanisms must operate together:

First, the workweek should be reduced gradually, without a proportional reduction in pay, in line with sustained productivity growth in each sector. Second, a job-transition fund should turn training into a right and paid time rather than a personal loan. Third, workers and society should receive a share of automation-derived profits through social ownership, cooperatives, or taxes on rents. Fourth, basic services—health, education, housing, internet access, and computing capacity—should be guaranteed so that citizens’ ability to live does not depend solely on selling their labor power.

Within this framework, a basic income can play a complementary role, but it cannot replace the right to good-quality work, public services, and participation in power. Cash payments without changes in ownership may simply return part of public income to rent, subscriptions, and monopolistic platforms.

15. A Five-Year Transition Roadmap

Year One—Transparency and rights: Create a public registry of widely used government models; disclose cloud contracts and energy and water consumption; require algorithmic impact assessments; establish a right of appeal; prohibit dismissal based solely on automated decisions; and form worker councils for AI decisions.

Year Two—Shared infrastructure: Create public computing capacity for universities, health, education, and cooperatives; establish data-portability standards; adopt multi-vendor public procurement; and prevent lock-in to a single cloud provider or model.

Year Three—Social ownership: Launch a knowledge and compute fund, data trusts, socially held equity in publicly subsidized projects, and mechanisms that return part of the benefits to society.

Year Four—Participatory planning: Connect anonymized and audited data from energy, health, transportation, and production; publish scenarios; and conduct multilevel voting and review of priorities.

Year Five—Sharing time and wealth: Reduce working hours in sectors that have achieved sustained productivity growth; expand basic services; measure inequality, freedom, job quality, and sustainability; and reform institutions in light of real results.

This transition is neither a command imposed suddenly from above nor a passive wait for markets to collapse. It is an experimental, lawful, reversible, and auditable process that builds social capacity step by step.

16. Responses to Three Objections

First objection: “Planning always fails.” Large corporations already use data and algorithms to plan supply chains and investment. The issue is not whether planning exists, but its scale, information, accountability, and objective. Democratic planning must avoid administrative centralization, a single hidden goal, and secrecy, while preserving local experimentation and revision.

Second objection: “Social ownership destroys innovation.” Much of AI’s foundation was built through universities, public research, the internet, semiconductors, and a workforce educated with social resources. Innovation requires motivation, but monopoly and personal wealth are not the only motivations. Scientific recognition, professional autonomy, public purpose, fair pay, and the ability to build are motivations as well. The proposed system can retain private and cooperative enterprises while placing foundational bottlenecks under public rules and fair competition.

Third objection: “AI is too complex for democracy.” Complexity is not a reason to exclude people; it is a reason to build accountable expert institutions. Just as budgets, medicine, and electricity grids are governed through combinations of expertise, law, and public oversight, AI requires independent expertise, stakeholder representation, and rights of appeal.

17. Conclusion: A New Productive Force and the Possibility of a New Order

Artificial intelligence reveals the longstanding contradiction between social production and private ownership at a new level. Knowledge, data, language, labor, universities, electricity grids, and the global chip supply chain all participate in its production, while decision rights and economic returns are concentrated in the hands of a small number of infrastructure owners. This concentration is not an inevitable result of technology; it results from rules of ownership and power.

A new social order in the Age of Consciousness cannot be achieved through simple nationalization or algorithmic government. Its foundations are democratic social ownership, participation in power, freedom of criticism, pluralism, data rights, independent auditing, and the sharing of productivity gains. AI should increase society’s ability to see alternatives and coordinate resources; decisions about objectives must remain human, political, and public.

The final question is not whether machines become more intelligent than human beings. It is whether society becomes conscious and organized enough to turn the machine’s new power into freedom, leisure, equality, and shared flourishing. This is the point at which the “Age of Consciousness” can move from the name of a period to the foundation of a new social order.

Selected Sources

• OpenAI’s official report on revenue growth
• Microsoft’s third-quarter fiscal 2026 financial report
• Nvidia’s fiscal 2026 financial report
• Reuters reporting on Anthropic’s annualized revenue run rate
• Amazon’s official announcement of its investment in Anthropic
• International Energy Agency report on energy and AI
• International Labour Organization report on generative AI and jobs
• Reuters reporting on Google’s investment in AI infrastructure in Finland

Methodological note: Revenue figures for private companies—especially OpenAI and Anthropic—are often annualized run rates or media reports and are not equivalent to realized, audited revenue for a complete year. Figures and prices in this industry change rapidly; the data in this text was reviewed through September 2026.

Return to Home