Authored by: Conflux
On May 31, 2026, the U.S. Department of Commerce issued new export control guidance: the channel for Chinese companies to purchase NVIDIA's advanced chips through overseas subsidiaries in places like Malaysia was officially closed.
In the same month, the President of Kenya halted a $1 billion geothermal data center project involving Microsoft—because once completed, it would consume one-third of the nation's electricity. President Ruto's exact words were: "It's like shutting down half the country."
Meanwhile, Huawei announced last week that its Ascend 950PR chip has entered mass production, with full-year AI chip revenue expected to reach $12 billion.
Three events, three continents, three completely different news stories. Yet they point to the same emerging reality: the competition for computing power is no longer just an internal matter for the tech industry.
A New Era of Oligopoly is Forming
Over the past two years, there has been a frequently overlooked reality in the AI industry: while it appears vibrant on the surface, the underlying resources are becoming increasingly concentrated.
The current AI industry chain can be roughly divided into four layers: GPU chips, cloud computing platforms, foundational models, and application ecosystems. In each layer, control is consolidating into the hands of a few players: in the GPU field, NVIDIA has become almost the sole choice; in cloud computing, AWS, Microsoft Azure, and Google Cloud dominate; at the model layer, OpenAI and Anthropic already hold the vast majority of the high-end model market.
In other words: the same set of companies is simultaneously controlling chips, cloud platforms, models, and distribution channels. University of Chicago law professor Eric Posner refers to this phenomenon as the "AI Octopus"—meaning these companies' tentacles cover the entire AI supply chain.
This differs from the platform monopolies of the internet era—internet platforms controlled traffic, while AI platforms control intelligence itself. This "oligopolistic monopoly" brings profound systemic risks:
- Concentrated Control and Pricing Power: A handful of companies control AI's pricing, API access permissions, and content moderation standards. Developers and businesses face severe "platform lock-in" risks, as giants can change rules or cut off access at any time.
- Infrastructure Fragility: Highly centralized computing power is prone to single points of failure that can affect the entire system (such as widespread cloud service outages) and places unsustainable pressure on regional power grids and energy supplies.
- Geopolitics and Compute Hegemony: Computing power is shifting from a neutral infrastructure to a strategic bargaining chip. Due to export control restrictions, countries without independent computing capacity (especially in the Global South) face the risk of being marginalized and experiencing a widening technological gap.
In the future, more and more enterprises will rely on AI for development, operations, customer service, marketing, and even decision-making. Once intelligence becomes a production tool, the importance of its control will far surpass that of search engines and social media.
The Deepening "AI Iron Curtain"
Over the past two years, U.S. actions on chip export controls have become increasingly fragmented. The Biden administration established an "AI diffusion rule," dividing global cooperation into three tiers; after taking office, Trump scrapped this rule, shifting to case-by-case approvals and temporary bans. Responses to this iron curtain vary significantly among nations.
Saudi Arabia directly declared 2026 its "Year of Artificial Intelligence": through the sovereign wealth fund's company HUMAIN, Saudi Arabia invested $3 billion in Musk's xAI, with one condition being the establishment of AI data centers with over 500 megawatts of capacity in Saudi Arabia. The UAE is building a 5-gigawatt AI campus in Abu Dhabi—reportedly the largest outside the U.S.—with the first phase launching this year. In May, the UAE received its first shipment of NVIDIA's latest chips exported from the U.S.
The logic of Gulf nations is straightforward enough: the last era relied on selling oil; this era relies on buying compute power.
EU anxiety stems from another direction: official data shows that over 80% of Europe's digital services run on non-EU infrastructure. The proposed "Cloud Computing and AI Development Act" (CADA) aims to triple Europe's computing capacity by 2030. France's Mistral published a strategic document in April titled "European AI: A Playbook to Own It."
The most difficult situation is for economies that barely have the qualifications to even participate in the competition: Kenya's $1 billion data center was halted; Malaysia allocated approximately $490 million to build a sovereign AI cloud. India is subsidizing researchers' GPU usage fees; Indonesia is preparing a domestic large language model—these investments are significant within their respective economic scales.
However, this year, the combined AI capital expenditures of just four companies—Microsoft, Google, Amazon, and Meta—total around $750 billion. This scale gap is itself part of the problem.
And the endgame of the compute power competition increasingly points to a more fundamental variable: electricity. The energy consumption of a single AI inference task can be up to 1000 times that of a traditional web search. To cope with the projected global data center energy consumption reaching 1050 terawatt-hours in 2026, tech companies have even started directly purchasing nuclear power plants.
Is There a Possibility to "Not Take Sides"?
It is precisely against this backdrop that Decentralized AI (DeAI) is beginning to gain attention. It attempts to answer a question: Besides entrusting the future to a handful of tech giants or a few nations, is there a third possibility?
If the internet can connect global networks through open protocols, can AI also connect global compute power through an open network? Can idle GPUs worldwide, independent developers, research institutions, and corporate data centers form an open AI infrastructure network?
The core idea of DeAI is not complicated: through open protocols, coordinate independent participants to create AI systems without a single controlling power center. By combining blockchain technology, cryptoeconomic incentives, and cryptographic verification mechanisms, it addresses the trust problem in anonymous networks, directly responding to the pain points of centralized AI:
- Breaking Market Concentration: Establish a network of distributed compute power, data, and model providers, forming a free-market pricing mechanism.
- Alleviating Physical Constraints: Distribute massive energy demands across power grids worldwide.
- Escaping Geopolitical Dependence: Build an infrastructure layer that transcends any single jurisdiction, offering possibilities for "sovereign AI."
- Improving Verification Transparency: Use provable technical means to replace blind trust in the corporate reputations of tech giants.
Proponents argue that this model can reduce dependence on single suppliers, enhance system resilience, and provide opportunities for participation by smaller nations and enterprises.
Meanwhile, institutional investor attitudes are shifting from curiosity to substantial investment. Venture capital firms (like DCG, a16z) are injecting hundreds of millions of dollars into DeAI protocols; traditional enterprises (like Deutsche Telekom) are beginning to participate in networks as validators; moreover, some national governments (like Kazakhstan) are exploring connecting their idle national supercomputing resources to decentralized compute markets.
Conclusion
As noted in the "State of DeAI 2026" report, the core value proposition of DeAI does not lie in immediately surpassing centralized systems in performance today, but in offering an underlying architecture that resists monopolies, rejects censorship, and disperses power.
With the declining cost of specialized AI hardware (ASIC) and the continued flourishing of open-source models, the time window for DeAI to solve operational challenges is opening. The work of building the DeAI foundation is just beginning.
Of course, DeAI still has a long way to go before becoming mainstream. Whether in terms of performance, stability, or business models, it remains in its early stages. But its significant meaning may not lie in immediately challenging OpenAI, but in providing an alternative.
Historical experience tells us: when an industry has only one choice, the question is often not *if* power will be abused, but *when*.
And the very existence of competition is itself a form of checks and balances.
