A covert war over the distribution of AI rights

Author: Zhixiong Pan

When we talk about AI, the public discourse can easily be diverted by topics like “parameter scale,” “ranking positions,” or “which new model has outperformed whom.” We cannot say that this noise is meaningless, but it often acts like a layer of froth, obscuring the more fundamental undercurrents beneath the surface: in today's technological landscape, a covert war over the distribution of AI rights is quietly unfolding.

If you raise your perspective to the scale of civilizational infrastructure, you will find that artificial intelligence is simultaneously presenting two completely different yet intertwined forms.

A “lighthouse” like a high-hanging coast, controlled by a few giants, pursuing the farthest illumination distance, representing the current cognitive limits that humans can reach.

Another kind of “torch” that can be held in hand, it pursues portability, ownership, and replicability, representing the intelligent baseline that the public can access.

By understanding these two types of light, we can break free from the confusion of marketing jargon and clearly determine where AI will lead us, who will be illuminated, and who will be left in the dark.

Lighthouse: The Cognitive Height Defined by SOTA

The so-called “lighthouse” refers to models at the Frontier / SOTA (State of the Art) level. In terms of complex reasoning, multimodal understanding, long-chain planning, and scientific exploration, they represent the type of systems that are the most capable, highest cost, and most centralized in organization.

Institutions like OpenAI, Google, Anthropic, and xAI are typical “tower builders”; what they construct is not just a series of model names, but a production method that “exchanges extreme scale for boundary breakthroughs.”

Why the lighthouse is destined to be a game for the few

The training and iteration of cutting-edge models essentially forcefully bundles together three extremely scarce resources.

First is computing power, which not only means expensive chips but also large-scale clusters, long training windows, and extremely high interconnection network costs; next is data and feedback, which requires massive corpus cleaning, continuously iterated preference data, complex evaluation systems, and high-intensity manual feedback; finally, there are engineering systems, covering distributed training, fault-tolerant scheduling, inference acceleration, and the entire pipeline for transforming research results into usable products.

These elements create a very high barrier to entry, which cannot be replaced simply by a few geniuses writing “smarter code”; it is more like a vast industrial system that is capital-intensive, has a complex chain, and where marginal improvements are becoming increasingly costly.

Therefore, lighthouses inherently have centralized characteristics: they are often controlled by a few institutions that possess training capabilities and data loops, ultimately being used by society in the form of APIs, subscriptions, or closed product formats.

The dual meaning of the lighthouse: breakthrough and traction

The existence of the lighthouse is not to “make everyone write copy faster”; its value lies in two more hardcore functions.

First is the exploration of cognitive limits. When tasks approach the edge of human capabilities, such as generating complex scientific hypotheses, conducting interdisciplinary reasoning, multimodal perception and control, or long-term planning, what you need is the strongest beam. It does not guarantee absolute correctness, but it can illuminate the “feasible next step” further.

Secondly, there is the traction of the technical route. Cutting-edge systems often pioneer new paradigms: whether it’s better alignment methods, more flexible tool calls, or more robust reasoning frameworks and security policies. Even if they are later simplified, distilled, or open-sourced, the initial path is often paved by lighthouses. In other words, a lighthouse is a laboratory at the social level that allows us to see “what intelligence can still achieve,” and forces the efficiency improvement of the entire industrial chain.

The Shadow of the Lighthouse: Dependency and Single Point Risk

But lighthouses also have obvious shadows, and these risks are often not disclosed in product launches.

The most direct aspect is controlled accessibility. The extent to which you can use it and whether you can afford it completely depends on the provider's strategy and pricing. This leads to a high dependence on the platform: when intelligence primarily exists as a cloud service, individuals and organizations effectively outsource their key capabilities to the platform.

Behind convenience lies fragility: network outages, service interruptions, policy changes, price increases, and interface modifications can instantly render your workflow ineffective.

The deeper hidden danger lies in privacy and data sovereignty. Even with compliance and commitments, the flow of data itself remains a structural risk. Especially in scenarios involving healthcare, finance, government affairs, and core business knowledge, “putting internal knowledge in the cloud” is often not just a technical issue, but a severe governance issue.

Furthermore, as more industries entrust key decision-making processes to a few model providers, systemic biases, evaluation blind spots, adversarial attacks, and even supply chain disruptions will be amplified into significant social risks. A lighthouse can illuminate the sea, but it is part of the coastline: it provides direction while also implicitly defining the shipping lanes.

Torch: Open Source Defined Smart Bottom Line

Take your gaze back from the distance, and you will see another source of light: an open-source and locally deployable model ecosystem. DeepSeek, Qwen, Mistral, etc. are just some of the more prominent representatives; behind them represents a whole new paradigm that transforms considerable intelligent capabilities from “cloud-scarce services” into “downloadable, deployable, and modifiable tools.”

This is the “torch.” It does not correspond to the upper limit of ability, but rather to the baseline. This does not represent “low ability,” but rather signifies the intelligent benchmark that the public can access unconditionally.

The meaning of the torch: turning intelligence into assets

The core value of the torch lies in its transformation of intelligence from a rental service into an owned asset, which is reflected in three dimensions: privatizable, transferable, and composable.

The so-called private means that model weights and inference capabilities can run locally, on an internal network, or on a private cloud. “I own an intelligence that works,” which is fundamentally different from “I am renting the intelligence of a certain company.”

The so-called portability means that you can switch freely between different hardware, different environments, and different vendors, without having to bind critical capabilities to a specific API.

Composable systems allow you to integrate models with retrieval (RAG), fine-tuning, knowledge bases, rule engines, and permission systems, creating a system that meets your business constraints, rather than being confined by the boundaries of a generic product.

This falls into very specific scenarios in reality. Knowledge Q&A and process automation within enterprises often require strict permissions, auditing, and physical isolation; regulated industries such as healthcare, government, and finance have strict “data does not leave the domain” red lines; and in manufacturing, energy, and on-site operations in weak network or offline environments, edge reasoning is even more essential.

For individuals, the long-term accumulation of notes, emails, and private information also requires a local intelligent agent to manage it, rather than entrusting a lifetime of data to some “free service.”

The torch makes intelligence not just an access right, but more like a means of production: you can build tools, processes, and barriers around it.

Why does the torch get brighter and brighter?

The improvement of open-source model capabilities is not accidental, but rather the result of the confluence of two paths. One is the diffusion of research, where cutting-edge papers, training techniques, and inference paradigms are quickly absorbed and replicated by the community; the other is the extreme optimization of engineering efficiency, with technologies such as quantization (e.g., 8-bit/4-bit), distillation, inference acceleration, layered routing, and MoE (Mixture of Experts) continually bringing “usable intelligence” down to cheaper hardware and lower deployment thresholds.

Thus, a very realistic trend has emerged: the strongest models determine the ceiling, but “strong enough” models determine the speed of adoption. Most tasks in social life do not require “the strongest”; what is needed is “reliable, controllable, and cost-stable.” Fire Torch happens to correspond to this type of demand.

The cost of the torch: safety outsourced to the user.

Of course, the torch is not a natural justice; its cost is the transfer of responsibility. Many risks and engineering burdens that were originally borne by the platform are now transferred to the users.

The more open the model, the easier it is to be used for generating fraudulent scripts, malicious code, or deep fakes. Open source does not equal harmless; it simply decentralizes control while also decentralizing responsibility. Additionally, local deployment means you have to address a series of issues yourself, including evaluation, monitoring, prompt injection protection, permission isolation, data de-identification, model updates, and rollback strategies.

Even many so-called “open source” projects are more accurately described as “open weight,” which still have constraints in commercial use and redistribution. This is not only an ethical issue but also a compliance issue. The torch gives you freedom, but freedom has never been “cost-free.” It is more like a tool: it can build, but it can also harm; it can save oneself, but it requires training.

The Convergence of Light: The Co-evolution of Upper Limits and Baselines

If we only see the lighthouse and the torch as a dichotomy of “big players vs open source,” we will miss a more genuine structure: they are two segments of the same technological river.

The lighthouse is responsible for pushing the boundaries, providing new methodologies and paradigms; the torch is responsible for compressing, engineering, and disseminating these results, turning them into widely applicable productivity. This diffusion chain is already very clear today: from papers to reproduction, from distillation to quantification, then to local deployment and industry customization, ultimately achieving an overall elevation of the baseline.

The elevation of the baseline will in turn affect the lighthouse. When a “sufficiently strong baseline” is accessible to everyone, it becomes difficult for giants to maintain a monopoly solely based on “fundamental capabilities” for a long time, and they must continue to invest resources in seeking breakthroughs. At the same time, the open-source ecosystem will generate richer evaluations, counteractions, and user feedback, which in turn will promote frontier systems to be more stable and controllable. A large number of application innovations occur within the torch ecosystem, where the lighthouse provides capabilities and the torch provides the soil.

Therefore, rather than saying these are two factions, it is more accurate to say these are two institutional arrangements: one system concentrates extreme costs in exchange for breaking limits; the other system disperses capabilities in exchange for inclusiveness, resilience, and sovereignty. Both are indispensable.

Without a lighthouse, technology can easily fall into a stagnation of “only optimizing cost performance”; without a torch, society can easily fall into a dependence on “capabilities being monopolized by a few platforms.”

The harder but more critical part: What exactly are we fighting for?

The contest between lighthouses and torches, on the surface, is about the similarities and differences in modeling capabilities and open-source strategies, but in reality, it is a covert war about the distribution of AI power. This war does not take place on a smoke-filled battlefield, but unfolds in three seemingly calm yet future-determining dimensions:

First, competing for the definition of “default intelligence”. When intelligence becomes infrastructure, the “default option” means power. Who provides the default? Whose values and boundaries does the default follow? What are the default's censorship, preferences, and commercial incentives? These questions will not automatically disappear just because the technology is stronger.

Second, the way to bear externalities in competition. Training and inference consume energy and computing power, data collection involves copyright, privacy, and labor, and model outputs affect public opinion, education, and employment. Both lighthouses and torches create externalities, just with different distribution methods: lighthouses are more centralized, more regulatory but also more like a single point; torches are more decentralized, more resilient but harder to govern.

Third, the competition for individuals' positions within the system. If all important tools must “connect, log in, pay, and comply with platform rules,” an individual's digital life will become like renting: convenient, but never truly their own. Torch provides another possibility: allowing individuals to possess a portion of “offline capabilities,” keeping control over privacy, knowledge, and workflow in their own hands.

The dual-track strategy will become the norm.

In the foreseeable future, the most reasonable state is not “fully closed source” or “fully open source”, but rather a combination similar to that of an electrical system.

We need lighthouses for extreme tasks, to handle scenarios that require the strongest reasoning, cutting-edge multimodal capabilities, cross-domain exploration, and complex research assistance; we also need torches for critical assets, to build defenses in scenarios involving privacy, compliance, core knowledge, long-term stable costs, and offline availability. And between the two, there will be a large number of “intermediate layers”: proprietary models built by enterprises, industry models, distilled versions, and hybrid routing strategies (simple tasks handled locally, complex tasks handled in the cloud).

This is not a compromise, but an engineering reality: the upper limit seeks breakthroughs, while the baseline seeks popularity; one pursues excellence, while the other pursues reliability.

Conclusion: The lighthouse guides the way ahead, the torch guards the steps below.

The lighthouse determines how high we can push intelligence, it is the offensive of civilization in the face of the unknown.

The torch determines how broadly we can allocate intelligence, and that is society's self-restraint in the face of power.

It is reasonable to applaud the breakthroughs of SOTA, as it expands the boundaries of what humanity can think about; it is equally reasonable to applaud the iterations of open source and privatization, as it allows intelligence to not only belong to a few platforms but to become a tool and asset for more people.

The true watershed of the AI era may not be “whose model is stronger,” but whether you have a light in your hand that doesn't need to be borrowed from anyone when night falls.