We are involved in this machine; don’t get upset by this message. Look at what you can do to make things more “evenly distributed”.
The most important question about superintelligence is no longer simply whether the technology works.
The deeper question is who will own the infrastructure, who will control access to it, who will receive the wealth it creates, and what happens to the people whose work makes the system valuable.
SI is often described as an open revolution. Models are released publicly. Code is shared on GitHub. Datasets and experiments appear on Hugging Face. Developers around the world contribute improvements, documentation, and research.
But openness at the production layer does not necessarily mean openness at the ownership layer.
A project can be open source while the platform hosting it is privately owned. A model can be freely downloadable while the hardware and APIs needed to run it are controlled by a small group of companies. A developer can publish an idea openly while a much larger corporation turns that idea into a closed commercial product.
This creates the possibility of an enormous transfer of wealth from distributed creators to concentrated platform owners.
Sitting Ducks — Michael Bedard 1977
The emerging ownership structure
The SI economy is forming around several strategic layers.
There are the chip companies that manufacture the hardware. There are cloud companies that rent the computing capacity. There are model companies that train and operate the systems. There are developer platforms that host code, models and datasets. Finally, there are the applications that convert SI capabilities into products and services.
Many of these layers are controlled by a surprisingly small number of companies.
Nvidia’s agreement to acquire Hugging Face for approximately $12.9 billion is a major example. Hugging Face has become one of the central platforms for sharing open-source models, datasets and tools. Nvidia says the platform will remain open and that developers will not be required to use Nvidia hardware. However, ownership still gives Nvidia a major position inside the distribution and discovery layer of SI development.
Microsoft already owns GitHub, one of the most important repositories of source code and developer activity in the world. GitHub reportedly has more than 150 million users and hosts more than 420 million projects.
This does not mean Microsoft owns every project on GitHub or Nvidia owns every model on Hugging Face. It means that two enormous corporations have influence over where software and models are stored, discovered, ranked, tested, discussed and commercialised.
That distinction matters.
The future may not be controlled by the company that owns every piece of code. It may be controlled by the company that owns the roads through which code travels, the infrastructure.
The platform is the new landlord
A useful way to understand this is to think of digital platforms as landlords.
The developer creates the intellectual property. The platform provides the building:
hosting
search
identity
collaboration
rankings
analytics
deployment
payments
enterprise access
recommendations.
At first, the platform may be generous. Free accounts encourage participation. Open repositories create network effects. Developers invite other developers. Models become easier to find. The ecosystem grows.
Once the platform becomes essential, the balance of power changes.
The platform can introduce fees, change visibility, promote its own products, alter search rankings, place ads, limit competing services or change the terms of use. It may not need to ban a project. Making the project difficult to discover could be enough.
This is how practical control works in a digital economy. Formal ownership is only one kind of power. Visibility, distribution and default placement can be just as important.
A model may technically be available to everyone, but if it is ranked below a company-owned alternative, excluded from recommended lists or made difficult to deploy, its practical reach can be severely reduced.
The creator may legally give away the business
The open-source licensing model makes this particularly important.
If a developer releases a project under the MIT or Apache 2.0 licence, a company can generally use the code commercially, modify it, combine it with proprietary software and sell the result.
The company usually does not have to publish its modifications or contribute improvements back. It normally only has to preserve required notices and comply with the licence terms. Apache 2.0 adds clearer rules around attribution, notices and patents, but it still allows commercial use.
This is not necessarily theft under the law. It is often the intended purpose of a permissive licence.
The problem is that many creators do not fully appreciate what they are giving away. They think they are sharing a tool. In reality, they may be releasing:
the implementation
the technical architecture
the market concept
the product roadmap
the evidence of customer demand
the lessons from failed experiments
the knowledge needed to build a competing product.
A corporation may take the repository, preserve the licence notices and build a paid service around it.
The creator may retain copyright in the original code, but that does not mean the creator retains commercial exclusivity.
This is the difference between legal ownership and economic power.
Ideas are easier to borrow than code
Copyright generally protects the expression of an idea more clearly than the idea itself.
Suppose an independent developer publishes an open-source memory system for SI agents. The system includes a database schema, retrieval engine, API and documentation.
A large company might:
copy portions of the code under the licence
rewrite the system in another language
build a similar architecture independently
use the public documentation to understand the market
use its cloud, distribution and sales resources to reach customers first.
Even if the company does not copy the code directly, it may still benefit from the creator’s public research.
The original developer has taken the risk. They have paid for the experiments, written the documentation, attracted early users and demonstrated that the problem matters.
The larger company may then arrive after the uncertainty has been removed.
This is one of the quiet forms of intellectual-property transfer in the technology industry. The original creator may not lose ownership of a specific file. They lose the opportunity to be the only person able to commercialise the insight.
The new form of extraction
Traditional resource extraction involved taking oil, minerals or agricultural products from one place and concentrating the value elsewhere.
The digital equivalent can involve taking:
code
research
ideas
user behaviour
model evaluations
community knowledge
developer time
open datasets
The platform supplies infrastructure and monetises the resulting activity.
This does not mean platforms contribute nothing. Hosting, security, compute, storage, search and collaboration are expensive. The issue is the imbalance between the value contributed by millions of participants and the bargaining power of the company operating the platform.
A developer might contribute years of work to an open project. The platform can earn revenue from enterprise subscriptions, advertising, compute usage, developer tools and strategic intelligence.
The creator receives visibility and perhaps reputation. The platform receives recurring revenue and a strategic view of the entire ecosystem.
That is an uneven exchange when the platform becomes indispensable.
The state is entering the ownership structure
The government is now becoming part of this story.
US officials have discussed whether the government should receive equity stakes in major AI companies. Reports have described proposals involving OpenAI and a public investment vehicle that could distribute future returns to American households. Reuters also reported that Anthropic was not involved in discussions about providing the government with equity at that stage.
This is important, but it should not be described too simply as nationalisation or as a public utility.
A government holding shares in a private company is not the same as the company becoming a regulated public service. The company may still set prices, control access, exclude users and pursue private commercial objectives.
The key question is what the public receives in exchange for public involvement.
There are several possible arrangements:
the government buys shares with taxpayer funds
the company donates shares
the government exchanges subsidies or contracts for equity
the state receives board representation
the state receives only passive ownership
future returns are distributed through a public wealth fund
taxpayers absorb losses while executives retain operational control
These arrangements are economically very different.
If the government takes a stake using public money, taxpayers carry investment risk. If the company donates shares, the public may receive future upside without paying directly, but the donation could still be exchanged for political influence, regulatory protection or public contracts.
Either way, ownership alone does not guarantee public control.
Public risk and private control
The most dangerous arrangement would be one where public money supports private companies without creating meaningful public rights.
Consider the possible sequence:
Venture capital funds the early company
Government subsidies support chips, energy, data centres or research
Public agencies become major customers
The company becomes too strategically important to fail
The government accepts equity or provides emergency support
Taxpayers absorb part of the downside
Executives and private investors retain control over operations and pricing
This would amount to socialising risk while preserving private control.
The government may describe the arrangement as an investment. But taxpayers may not receive the same protections as private investors. They may have no board seat, no meaningful ability to influence strategy and no guarantee that returns will be distributed to the public.
A small passive stake can create the appearance of public participation without giving the public real power.
The Intel arrangement is a useful warning. The US government agreed to purchase an approximately 9.9% stake in Intel, with the investment associated with federal semiconductor funding. The announced structure was described as passive and did not provide ordinary governance control.
That model may be appropriate in some circumstances, but it demonstrates the difference between owning part of a company and controlling its decisions.
From technology oligopoly to political oligarchy
The concern is not only economic concentration. It is the interaction between economic concentration and political power.
A Silicon Valley oligarchy would not necessarily look like a government dictatorship. It would operate through ownership, influence and dependency.
A small number of companies could control:
the chips
the cloud
the models
the developer platforms
the software distribution
the enterprise contracts
the government relationships
the funding channels
They would not need to own every startup. They could own enough infrastructure to shape the choices available to startups.
**Independent companies might have to buy compute from the same **corporations that compete with them. They might have to distribute models through platforms owned by major infrastructure providers. They might rely on cloud credits, marketplace rankings and enterprise partnerships controlled by potential competitors.
This produces a system in which competition is allowed, but the terms of competition are privately administered.
The smaller company can exist, but it may never become independent enough to challenge the system.
The potential near-future outcomes
Several outcomes are plausible over the next few years.
The controlled public wealth model
Governments receive minority stakes in major AI companies. Those stakes are placed into public funds and future returns are distributed to citizens.
This could give the public a genuine claim on AI-generated wealth. But it may also legitimise a small group of private companies as permanent national champions.
The result could be public dividends combined with private oligopoly.
The protected national champion model
Governments decide that frontier AI is too important to leave to ordinary competition. They provide subsidies, contracts, energy access and regulatory protection to a handful of companies.
Smaller competitors remain technically free to operate but cannot match the capital and infrastructure advantages of the protected firms.
This would resemble an industrial policy system centred on AI.
The platform enclosure model
Open models and code remain publicly accessible, but discovery, compute and deployment become concentrated.
The open ecosystem continues to exist, but the economically valuable part is captured by the platform owners. Developers can contribute to the system, but cannot easily monetise their work independently.
The AI bubble and rescue model
Large investments in data centres and AI companies fail to generate the expected returns. Some firms collapse. Governments intervene because the infrastructure is considered strategically important.
Taxpayers then fund rescues, while the surviving companies acquire failed competitors and become even more concentrated.
This would be the clearest version of the feared rug pull.
The decentralisation response
Developers begin moving toward self-hosted Git platforms, independent model registries, federated identity, local inference and cooperative infrastructure.
Open-source projects distribute their repositories and models across multiple locations. Creators adopt more restrictive or commercially structured licences.
This outcome would reduce platform power, but it would require developers to sacrifice some convenience.
The problem with calling everything theft
There is a danger in using the word “theft” too broadly.
If a creator releases code under MIT or Apache 2.0, commercial reuse is generally authorised. The company may be behaving aggressively, opportunistically or unfairly, but not necessarily unlawfully.
The stronger argument is about bargaining power.
A creator may voluntarily publish the code, yet still be operating inside a system where:
the licence is poorly understood
the platform owns the audience
the company has superior legal resources
the company can see early signals of commercial demand
the creator cannot afford enforcement
the code is legally reusable but the creator’s business model is not protected.
This is better described as asymmetric extraction than simple theft.
The transaction may be legal while the economic outcome is deeply unequal.
How creators can protect themselves
Open-source creators do not need to abandon openness, but they should separate what they want to share from what they need to own.
Possible strategies include:
releasing a useful core while retaining a proprietary hosted service
using dual licensing for commercial customers
considering AGPL where network-service sharing is important
keeping proprietary data and evaluation sets private
protecting the product name with a trademark
maintaining multiple repository mirrors
avoiding dependence on a single model registry
keeping customer relationships outside platform marketplaces
publishing only the components that benefit from adoption
using contributor agreements that clarify ownership
separating public demonstrations from commercially valuable infrastructure.
A repository should be treated as a strategic asset, not merely as a code dump.
The question should be:
What must be open for adoption, and what must remain controlled for the business to survive?
The deeper question
The future of SI will not be decided only by model intelligence.
It will be decided by ownership.
Who owns the compute? Who owns the data? Who owns the distribution channel? Who controls the ranking system? Who receives public subsidies? Who determines which projects receive visibility? Who captures the productivity gains when SI replaces or augments human work?
If the answers remain concentrated among a few technology companies and their financial partners, SI may produce tremendous technical progress while also creating one of the largest wealth transfers in modern history.
The public may receive better software and cheaper services, but still lose economic power. Independent creators may produce the innovations, while platforms capture the recurring revenue. Governments may support the infrastructure, while private companies retain the strategic control.
That would not be a conventional revolution in which power is distributed.
It would be a transfer from many creators to a small number of owners.
Conclusion
The central danger is not that SI companies will suddenly steal every piece of code or that open source will instantly disappear.
The danger is more gradual.
Millions of developers may create the intellectual foundation of the SI economy. A small number of companies may then control the hardware, cloud infrastructure, model platforms and distribution channels through which that work becomes valuable.
Government equity stakes could either give citizens a share of the upside or help legitimise and protect a private oligopoly. The difference depends on governance, transparency, competition and whether public risk produces public control.
Open source can remain legally open while becoming economically enclosed.
That is the issue to watch.
The critical test for the next decade will be whether SI becomes a widely owned productive infrastructure, or whether it becomes a privately controlled utility-like system in which society supplies the data, labour, capital and public support while a small Silicon Valley oligarchy captures the wealth.
Sources and factual context
Reuters reported that US officials had held preliminary discussions about the possibility of the federal government acquiring stakes in major artificial-intelligence companies. The report described the discussions as preliminary, rather than confirming that the government had completed an investment in OpenAI or Anthropic.
Reuters reported that OpenAI had discussed a possible 5% stake for the US government, according to reporting by the Financial Times. The proposal was described as part of broader discussions about national-interest involvement in advanced SI.
Reuters separately reported that the Trump administration and Anthropic had not discussed the government taking a stake in Anthropic at that time. This distinction is important because public statements and media reports have sometimes grouped OpenAI and Anthropic together despite the different status of the discussions.
Nvidia agreed to acquire Hugging Face for approximately $12.9 billion. Hugging Face is a major platform for hosting and distributing open-source SI models, datasets and development tools. The acquisition gives Nvidia a significant position in the SI software and developer ecosystem.
GitHub states that its platform is used by more than 150 million developers and hosts more than 420 million projects. This helps explain why control of GitHub has strategic importance beyond ordinary code hosting.
The MIT licence permits commercial use, copying, modification and redistribution, provided the required copyright and licence notices are preserved. The Apache 2.0 licence also permits commercial use and modification, while adding more detailed requirements concerning notices, modifications and patent rights.
GitHub’s documentation explains that software without an open-source licence remains protected by ordinary copyright law. A public repository is not automatically free for unrestricted commercial reuse merely because it can be viewed online.
GitHub’s generative SI terms state that GitHub will not use customer inputs or outputs to train generative SI models under the relevant service terms. However, developers should distinguish between different GitHub products, account types, settings and contractual terms rather than assuming one universal data policy.
The US government announced an agreement to purchase approximately 9.9% of Intel for about $8.9 billion. The arrangement was described as a passive investment without ordinary board representation or governance control, illustrating that public ownership does not automatically equal operational control.
The facts are developing, but the direction is already visible: SI is becoming not only a technology race, but a contest over ownership, distribution and political power.
Founder’s note
The article’s claims about wealth transfer, platform enclosure, intellectual-property extraction and Silicon Valley oligarchy are analytical arguments rather than settled facts. The cited material supports the underlying events and legal or commercial structures, while the conclusions about future control and wealth concentration remain potential forecasts and not financial, legal, political or personal advice.
VEKTOR Memory builds local-first persistent memory infrastructure for AI agents. Documentation and downloads are at vektormemory.com.
Si
Open Source
LLM
Hugging Face
Github
Top comments (0)