Victor Paraschiv is an AI scientist, founder and CTO of Broadn, and was an early pioneer in the training of the first LLMs on legal datasets
Most observers misunderstand the AI race between China and the U.S. The dominant narrative suggests the conflict is about open-source versus closed-source models or whether Model A outperforms Model B on a given benchmark. That’s only the surface. The real competition is a structural clash between two fundamentally different economic models: scarcity versus abundance and expensive premium versus commodity, uncurated, open-source.
The practical reality and narrative line is that the West treats AI as a premium, high-margin luxury, whereas China is rapidly commoditizing it as public infrastructure, fully embracing the paradigm shift. The West knows how to monetize scarcity, but right now, it’s unprepared to compete with abundance and infinite availability.
The Western Model: AI As A Gated Luxury
In the Western model, AI is treated as a premium product characterized by expensive access and controlled subscriptions. This isn’t corporate greed—it’s a structural necessity. Western tech giants are engaged in historically unprecedented capital expenditure to build frontier intelligence. Initiatives like OpenAI, SoftBank and Oracle’s “Stargate” supercomputer, a $500 billion project, demand massive revenues to justify their existence, all while contending with extreme demand and proven shortage in AI chip availability.
Yet this infrastructure push is increasingly colliding with physical and political reality. Slow grid development and fierce social backlash over rising household electricity bills have triggered widespread resistance across Maine, Georgia, Maryland, Michigan, New York and South Carolina. In early 2026, Maine passed the nation’s first statewide moratorium on large data centers, while states like Texas and Oregon imposed strict regulations on “large load” customers to protect grid reliability. From local zoning freezes in cities like Tulsa and Bangor to proposed federal legislation like the AI Data Center Moratorium Act, communities are actively pushing back against the environmental and economic toll of AI expansion. When data center construction is blocked by constrained power grids and angry ratepayers, compute becomes structurally scarce and inherently expensive.
This closed-garden approach is reinforced by legitimate risk aversion. Stringent regulatory frameworks like the EU AI Act, alongside corporate fears of data exfiltration and prompt injection, mandate highly controlled deployments. As a result, an essential, transformative technology is turned into something scarce and exclusive. Enterprises pay significant premiums for “safe,” closed systems while erecting high barriers to entry that restrict the technology’s diffusion across smaller businesses and ordinary users who can’t afford to navigate the regulatory or financial hurdles.
The Chinese Model: AI As Public Infrastructure
China is moving in a radically different direction. AI and computing power are increasingly being treated like electricity, railways or the internet. AI is public infrastructure that every industry can and should access at a low cost. The internal competition in China is extreme, and the immense economic advantages offered by AI across diverse sectors are far too significant to be overlooked.
This shift is driven by aggressive cost reductions and architectural innovations. For example, DeepSeek V4, released in April 2026, utilizes a hybrid architecture and claims to cut inference FLOPs by 73% and KV-cache memory by 90% compared to DeepSeek V3.2. The pricing is disruptive: Its Flash variant “costs $0.14 per million input tokens,” often seven to 10 times cheaper than Western equivalents, while natively supporting a one-million-token context. A one-million-token context was first introduced by Anthropic’s flagship model, Claude Opus 4.6, as a premium feature in February 2026. Google Gemini released a model with a two-million-token context in June 2024. Now, a one-million-token context is being commoditized.
This low-cost access changes the decision matrix entirely. When AI becomes cheap enough, it’s no longer reserved for billion-dollar corporations. Small businesses, factories, schools and ordinary people can innovate. This is illustrated by the explosive adoption of OpenClaw, an open-source AI agent framework that reached 20 million monthly active users in just one month. The resulting “raising lobsters” craze (paywall) saw ordinary citizens queuing outside tech headquarters for installations, supported by local government subsidies like Shenzhen’s GPU credits. China is turning an open-source tool into national productivity infrastructure at a speed no other country is currently matching.
The Future Threat: Compute Dumping
If this trajectory continues, the marginal cost of “intelligence” will drop radically over the next three to five years. The likely stakeholder reading in the West will be to frame this as an unfair market practice. We’ll inevitably hear terms like “AI dumping” and “compute dumping,” as China floods the global market with capable, less-expensive AI models.
This “dumping” doesn’t necessarily mean selling below cost; rather, it means compressing reference prices through massive efficiency gains and state-backed infrastructure. By offering similar AI services globally at a fraction of the cost, this infrastructure model directly threatens the margins and pricing power of the current global tech hierarchy.
The secondary effect is economic. Armed with cheap AI infrastructure, small Chinese businesses can flood the market with products at price points that make Western competition unviable. Commoditizing AI is a structural bet on winning the long game. China’s state-backed ecosystem has the institutional patience to wait for that payoff. Western venture capital doesn’t.
The Western Dilemma
The vulnerability for Western tech giants is acute. They’ve mastered the playbook for charging monopoly prices in a scarce environment, but they have no playbook to compete with infinite abundance.
The defining question that will shape the digital age isn’t which model is slightly smarter today, but what kind of economic objectives is AI inducing? Should AI be a premium product or a public utility? Is AI adoption an inescapable gravity well?
We’re standing at the edge of a profound paradigm shift. Consider the possibility that the ultimate winner of the AI race won’t be the nation that builds the most brilliant, expensive supercomputer. It might be the nation that makes intelligence so affordable, invisible and ubiquitous that we stop noticing it entirely.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
