The Plummeting Cost of AI and the Geopolitics of DeepSeek
Destroying a market by making its core product a commodity?
In November of 2022, the world changed as, for the first time ever, OpenAI released a large language model that was designed for people, for it to be used by all users. While the technology was not new—after all, the model released was GPT-3, which means that we had a one, two, and a three—what changed is that, for the first time ever, a model was produced in a way that everybody could have access to machines that we can interact with verbally and naturally, the way we interact with other human beings, getting us closer to the dream of the talking C-3PO.
For a brief moment, the United States appeared dominant in artificial intelligence development. OpenAI, with the support of Microsoft, seemed like the unavoidable AI company, ready to dictate how the world would adopt and integrate American-centric AI technologies.
All of this changed this past weekend when a dingy startup from China called DeepSeek released something that caught the world off guard. While much of the tech industry remained focused on OpenAI’s GPT-4 and the eventual arrival of GPT-5, DeepSeek dropped a thinking model with open-source access.
Unlike proprietary systems like ChatGPT or GPT-4, DeepSeek is available for developers to train, customize, and deploy on their own terms. It isn’t locked behind paywalls or reliant on centralized infrastructure. And its API pricing is unprecedented—at $0.0011 per 1,000 tokens, DeepSeek costs 27 times less than GPT-4, which charges $0.03 per 1,000 tokens!!!
For developers processing 10 million tokens, GPT-4 would cost $300. DeepSeek, by comparison, costs just $11.
The Cost Revolution: DeepSeek and the Ubiquity of AI
DeepSeek’s API pricing of $0.0011 per 1,000 tokens changes the cost structure of AI systems by making AI tokens way cheaper (tokens are the units of measure used to price AI use). DeepSeek’s cost reduction is staggering. This kind of price difference lowers the barrier to entry for tasks previously considered cost-prohibitive. At this prices, using AI becomes much more attractive for applications that before would have been considered price prohibited.
How DeepSeek’s Efficiency Makes This Possible
DeepSeek achieves its cost reductions through technical innovations that optimize its computational demands. Unlike proprietary systems that depend on vast, energy-intensive data centers, DeepSeek reduces costs by rethinking how tasks are processed.
It processes phrases rather than individual words, cutting down on computational overhead.
It activates only the parts of the model necessary for each task, conserving energy and memory.
It uses simplified calculations that further reduce processing demands.
According to Morgan Brown’s analysis (see here), these innovations also dramatically lower training costs. GPT-4 is estimated to have cost $100 million to train, while DeepSeek required just $5 million. Additionally, DeepSeek is optimized to run on affordable gaming GPUs instead of costly, specialized hardware, enabling developers to use it without investing in expensive infrastructure.
DeepSeek’s aggressive pricing strategy exerts significant pressure on the economics of proprietary AI systems. With API prices this low, DeepSeek slashes the cost of AI access by a factor of 27 compared to GPT-4 restructuring of the AI landscape. For companies like OpenAI, DeepSeek’s cost model threatens their ability to sustain their operations.
As DeepSeek's adoption spreads, the viability of proprietary AI systems comes into question. Lowering prices to compete with DeepSeek would undermine the revenue needed to support large-scale proprietary models, while maintaining current pricing risks ceding significant market share. DeepSeek, therefore, catalyzes a shift from centralized, high-cost AI systems to more widely distributed and accessible AI capabilities.
In the fight between closed vs open LLMs, open LLMs took the lead two days ago.
Furthermore, the challenge is not over: DeepSeek’s open-source design empowers developers globally to adapt, refine, and expand the model . This open framework accelerates innovation, creating a dynamic ecosystem where improvements can be shared, tested, and deployed without centralized control.
The impact is akin to the transformation seen with Linux, where an open-source platform enabled a diverse array of innovations that proprietary systems couldn’t match in speed or breadth. DeepSeek invites the same kind of widespread collaboration, effectively decentralizing AI development and fostering an environment where the model evolves through global contributions.
A Thinking Model You Can Own
DeepSeek’ challenges the core of the OpenAI business model with something we thought would take years to have: a thinking model you can own. It is designed to engage in reasoning before generating responses, providing capabilities typically associated with top-tier proprietary systems. This allows for the integration of advanced AI into customized solutions without the need for dependency on a single provider.
Crucially, DeepSeek can be deployed on private infrastructure. In other words, you can download and run your own instance of DeepSeek. You can’t do that with OpenAI GPT. Developers and organizations gain full control over their AI applications, and no more centralized point of control will exist. This capability shifts AI from being a service that must be accessed through a gatekeeper to a tool that can be owned and modified directly.
By being open-source, DeepSeek ensures that its innovations aren’t confined to a single company; in fact, DeepSeek could disappear tomorrow, and its model would outlive the company in hundreds of permutations. Developers can continue to build upon, adapt, and optimize the model, regardless of what happens to DeepSeek as a company.
AI technologies that were once restricted by proprietary controls are now available for widespread adaptation and use (how long before we see the proliferation of NSFW thinking models? two more weeks?) DeepSeek accelerates the democratization of AI, setting in motion a process where advanced AI capabilities become a standard part of the technological landscape, rather than a premium product controlled by a few tech leviathans.
How good are Deepseek thinking capabilities?
AI Technopolitics: A Bipolar AI ecosystem has emerged.
DeepSeek creates a clear opposing pole to the dominant role of American OpenAI up to this point. In fact, without a clear pricing response by OpenAI, American dominance is over.
This moment signals the emergence of a bipolar AI ecosystem, where Chinese-origin systems directly challenge U.S.-led proprietary models. For the last two years, OpenAI systems like GPT-4 dominated the global AI landscape, giving the United States a unique advantage in shaping technological adoption worldwide. DeepSeek, however, offers a radically different model—one that isn’t tied to U.S. infrastructure or pricing, breaking the centralized control that proprietary systems once commanded.
The release of DeepSeek is not just about technology; it reflects broader frustrations with U.S. leadership. Over the last two years, the transactional nature of Donald Trump’s policies has alienated key allies and eroded trust in American commitments. Countries eager to assert their independence from the United States now see DeepSeek as an opportunity to diversify their technological dependencies and reduce reliance on U.S.-based platforms.
For countries in the Global South, DeepSeek represents a rare chance to leap into the AI revolution. The barriers posed by high pricing and centralized infrastructure, which previously kept AI adoption out of reach, are dismantled by DeepSeek’s affordability and open-source design. Nations that were locked out of AI development due to the high costs of systems like GPT-4 can now integrate advanced technologies without relying on American or Chinese cloud providers.
The open-source nature of DeepSeek its its killer feature. By allowing developers and nations to deploy AI on local infrastructure, DeepSeek shifts control away from specific companies.. DeepSeek high high-performing fully downloadable open source architecture enables developers to bypass these constraints entirely. Removing the single point of regulation and control.
Given the staggering decrease in cost and DeepSeek’s unprecedented pricing, it is hard not to imagine that this is part of a state-sponsored capitalism strategy to undermine U.S. hegemony at a critical juncture.
By pushing prices this low, DeepSeek destabilizes the profitability of proprietary AI systems, leaving companies like OpenAI vulnerable. These firms depend on high-margin revenues to sustain their research, development, and infrastructure, all of which are now under immense pressure. Without the ability to compete on price, their dominance in the global AI landscape faces erosion.
This strategy serves a larger geopolitical purpose. By demonetizing AI, DeepSeek changes the power dynamics of the technology sector. Lower costs make AI more accessible, enabling nations, businesses, and developers outside the usual centers of power to integrate advanced systems into their infrastructure. This especially impacts countries in the Global South, where affordability has historically limited the adoption of cutting-edge technologies.
At the same time, DeepSeek’s open-source design amplifies its disruptive potential. By allowing developers to deploy AI on their own infrastructure, it removes the need to rely on centralized systems controlled by U.S.- or Chinese-owned cloud providers. This decentralization shifts technological control into the hands of those who deploy and customize the systems, further undermining the gatekeeping role of established AI powers.
If the Chinese government is subsidizing DeepSeek, the implications are profound. Such subsidies would represent a calculated effort to weaken the foundations of American AI dominance while embedding Chinese-origin technology as the new global standard. The demonetization of AI, driven by state-backed strategies, challenges the balance of power in AI development and adoption, reshaping the competitive and geopolitical landscape.
But subsidies or no subsidies, the financial foundation of proprietary AI systems is cracking. DeepSeek’s aggressive open-sourcing policy has made Antrophic’s and OpenAI business models obsolete in one weekend.
This shift signals the commoditization of Large Language Models and the end of US's uncontested role in the AI market. DeepSeek disrupts this equation by providing advanced models at a time there is little international support to US-led initiatives, by providing alternatives that are not under the control of any single company. The ball is now in OpenAI court, to respond in a way that demonstrates they can compete with open-source alternatives that, unlike previous open-source models, sacrifice only very little in performance while offering massive cost savings.
