In January 2025, a Chinese AI lab called DeepSeek released a large language model that matched the performance of OpenAI's best — at a fraction of the cost and trained on a fraction of the compute. The reaction in American technology circles ranged from alarm to denial. The stock market erased nearly $600 billion in value from AI infrastructure companies in a single day. The assumption that U.S. AI dominance was a fixed condition of the technological landscape turned out to be an assumption, not a fact.
The DeepSeek moment matters for equity not because of the geopolitics but because of what it reveals about the concentration of AI development. Until that week, the implicit model of AI progress was: a small number of American companies, backed by enormous capital and proprietary chip access, would build the systems that the rest of the world would use. Governance, access, and accountability would be structured around that concentration. DeepSeek didn't just challenge U.S. leadership — it challenged the premise that frontier AI must be expensive, closed, and controlled by capital-dense institutions.
The efficiency of DeepSeek's approach — using less compute to achieve comparable results — is significant beyond the competitive narrative. High compute requirements are one of the primary structural barriers to AI development outside of a handful of well-resourced institutions. If frontier capability can be achieved at lower cost, the geography of AI development changes. Universities, civil society organizations, public health agencies, and governments in lower-resource contexts have a different relationship to AI development when the compute barrier is lower. That is not a guarantee of better outcomes — cheaper AI can still encode harmful biases, still be deployed without accountability, still serve the interests of those who deploy it over those it affects. But the monopoly model was not producing equity either.
The deeper question the DeepSeek episode surfaces is: who controls the infrastructure that shapes public life? That question doesn't have a simple answer, and a Chinese lab is not an automatic improvement over an American one from an accountability or governance perspective. But the disruption of assumed concentration is worth examining seriously. The goal of Equity & AI has never been to argue for any particular company's dominance — it is to examine who AI serves, who it excludes, and what governance structures are adequate to the stakes. DeepSeek is a data point in that examination, not a resolution of it.
What comes next depends less on which country leads in AI benchmarks and more on whether the institutions deploying AI — wherever they are built — can be held accountable to the people they affect. That accountability infrastructure does not yet exist at adequate scale anywhere.